AMD's Dual-core Athlon 64 X2 reviewed

ChocolateJesus writes "Weeks after formally announcing its dual-core Athlon X2 desktop processor, reviews are finally trickling out. The Tech Report's coverage tests two flavors of the Athlon 64 X2 against a whopping 17 competitors, including AMD and Intel's fastest single- and dual-core offerings. They've even thrown in a handful of dual-processor systems (and dual-core, dual-processor systems) for good measure. Testing focuses on multi-threaded applications, and the X2s deliver remarkable performance. Perhaps even more impressive is the fact that unlike Intel's dual-core Pentiums, AMD's X2s consume no more power than single-core chips." Looks like this story has come out of embargo - if you've find more reviews, post them in comments.

Comments

Cooling

[Anonymous Coward]

I don't get how this can run on the same power level as the single core chips. Can someone explain on how this is possible?

Re:Cooling

[masklinn]

Specific design and use of a modified version of the most recent AMD core (Venice). Venice's consumption is much lower than it's parent (Winchester core), check the graphs, Dual Cores' power consumption is a bit higher than the 3800+ Venice processor.

On top of that, A64 platforms are known for their low power consumption compared to Netburst based processors.

Re:Cooling

[0x461FAB0BD7D2]

Easy. Multi-threaded electricity.

Re:Cooling

[Kobun]

Over at Anandtech, they have a similar article up.

http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2410&p=2

On that page they compare a 130nm single core Athlon to a 90nm dual core. Even under a full load, the 90nm dual core uses less power than the single core 130nm chip.

Re:Cooling

[moz25]

Well, if you design them to use less power overall and make use of the fact that they're on the same chip then it'll roughly be about the same power level. Also keep in mind that a slower-running (or idle) CPU will consume less too. In actual use, both processors probably won't be running at their full speed anyway. Such techniques are already common for mobile chips so they can improve battery life.

Re:Cooling

[Xoro]

I don't get how this can run on the same power level as the single core chips. Can someone explain on how this is possible?

It isn't.

Under load, the dual core system consumes about 25 watts more power than the single (178 watts vs. 154) -- and 25 watts is just less than what a single-core A64 consumes under load.

I think the poster was looking at the numbers for idling.

Re:Cooling

[Zemplar]

Nanotechnology and further advances in chip design resulting in more efficient transistors and better interconnects.

Re:Cooling

[drakyri]

For a smaller manufacturing process (90nm), the transistors are smaller. On the single-transistor level, at least, they require less power to operate than 180 or 130nm transistors.

Other considerations factor in to determine the power consumption (total number of transistors, other elements, arrangement, etc.), but the smaller size drops the power level quite a bit beforehand.

Re:Cooling

[Wesley Felter]

Most of the other responses are wrong; the manufacturing process and core revision have nothing to do with it. The single-core chips are faster (2.8GHz) than the dual-core chips (2.4GHz); since higher frequencies use much higher power, the power ends up being the same between the two chips.

Re:Cooling

[Anonymous Coward]

Close, it's the quantum nature of the electrons; each core shares all it's electons. In core 0 a quantum filter is applied so computations are done on spin up electrons; core 1 uses spin down. In order to move larger numbers of cores you have to use something with more than two states -- QCD confinment makes it tricky to get the necessary free quarks. That, however, is just an engineering issue and we'll soon be able get cpu's in any color you want.

Re:Cooling

[ThosLives]

The really interesting thing is they measured system power consumption, not chip consumption. They specified that the power supplies were the same, but the systems have different specs.

It's hardly accurate to judge a CPU's performance based on a "power drawn at the wall" measurement.

Re:Cooling

[smittyoneeach]

I knew there was something to that whole super-string theory business...

Re:Cooling

[ceswiedler]

I know very little about chip design, but I've heard that power and heat are becoming the big limiting factors in the scalability of processors.

Does that mean that AMD's processors have much more room to scale, since they draw less power and generate less heat today?

Are there any estimates for how high, and when, AMD will be able to scale up to signifigantly higher frequencies?

Why doesn't AMD ramp up the frequencies now, and put out processors now with approximately the same power draw as Intel's chips, but with signifigantly higher performance? Except in laptops (which Intel owns) and blade servers (which maybe AMD is going after), low power isn't a big deal. Performance is a big deal everywhere.

Re:Cooling

[zardo]

When they make the chips, the ones that perform poorly and produce more heat are just downgraded to lower clock speeds and sold for less. As manufacturing is perfected, all the chips are performing better, so each year there are advancements to be made, whether AMD utilizes the advancements to produce faster clock speeds or lower power consumption is up to them. These days I think power consumption is becoming more an issue than clock speed, and it also lends itself to clock speed eventually.

Re:Cooling

[orderb13]

Look at the numbers for the Athlon's. They used the same had the same specs, just changed out the chips. The Dual Cores ran under less load than the FX did.

Re:Cooling

[Short Circuit]

Nah...that would be quantum mechanics.

Re:Cooling

[nokiator]

Comparing the X2s against 130nm Athlon 64 is not fair. As expected, the new X2 Athlons do burn more power than a single core Athlon 64 built on the same core/process at the same frequency. The amazing thing is that the difference appears to be only around 20%, which is almost unbelievable.

You would expect to see less than 100% increase in the case of a dual core CPU due like the shared components which are not replicated:

- The X2 chips still have a single, 128-bit wide memory controller. Since the memory controller charges/discharges external bit lines going to DIMMs, they do burn quite a bit of power. This power consumption is not duplicated in the case of a dual core CPU.

- The X2 chips still have a single HyperTransport bus. The power consumption of this bus is the approximately the same between a dual core and single core CPU.

However, power scaling due to these shared components would probably not explain how a dual core chip can burn only 20% more power. For both of the above cases, you could argue that one should expect to see higher utilization of the memory bus and the HyperTransport bus, so the exact power consumption contribution is not entirely clear.

One thing to note that, AMD Athlon 64 cores tend to burn much less power in idle state compared to Intel chips. This is probably due to choices AMD made both in architecture and process. So the fundemantal reason why AMD X2 chips have such minimal incremental power consumption over single core chips is that one of the cores is typically underutilized most of the time and therefore burns much less power.

Re:Cooling

[warrior]

Smaller transistors, smaller capacitive loads to drive, lower supply voltages, and shorter wires. A good baseline for power in a synchronous digital system is P=C*V^2*f. From physics we know that q=C*V, therefore current (I) dq/dt = (C*V) * f, where f is the frequency the digital circuit is switching at. Power = current * voltage = C*V*f*V. Power goes down linearly with smaller devices and by the square of the supply voltage.

There are other factors in making smaller devices that keep you from getting a linear decrease in power, all in the form of various leakage currents. Some solutions to these are SOI (semiconductor on insulator) to reducue substrate leakage, high-k gate dielectric to reduce gate leakage.

There are also some circuit techniques to decrease power consumption. One example is clock gating (keep ckts that aren't in use from switching). Another is shying away from dynamic ckt design, a technique that is high-speed but switches twice per clock cycle, even when logic values aren't changing.

Re:Cooling

[lakeland]

Are you sure? Because I've read one (very poorly done) benchmark coming to this conclusion, and one (very well done) benchmark coming to the opposite conclusion.

Certainly Venice is cooler than (what was it before winchester? clawhammer??) but I don't yet believe venice runs cooler than winchester.

Re:Cooling

[Jherek Carnelian]

For what it is worth, those charts are moderately deceptive. Like most amatuer sites, Anandtech doesn't have the equipment to measure actual cpu power consumption. So they measured the consumption of the entire system.

So, assuming they used the same system for all measurements and just swapped out the cpus, the relative differences are accurate. But you can not draw any conclusion about the absolute power requirements of the cpus based solely on Anandtech's review.

Maybe no one cares, but it would be easy to read that article and come away with the idea that the dual core cpu consumes (and thus must dissipate) 150 watts under load. While that might be in the realm of possibility for Intel's cpus which are little micro-furnaces, the AMD chips are significantly less hot than that.

Re:Cooling

[aceh0]

the relationship between frequency and power is not linear. scaling the frequency back a couple steps results in a 40% reduction in voltage. check out this page for a graph of this from WinHEC. most of the articles show that this plays out in practice but few have tackled why.

Re:Cooling

[afidel]

That's not always true. Your leakage current loss can go so high that a smaller process processor with the same transistor count can use MORE current then the larger process chip. Intel mostly fought this with the use of strained silicon and by using nickel silicide instead of cobalt silicide when introducing Prescott at 90nm.

Re:Cooling

[dam]

Apart from the smaller core size; remember that the cache on the CPU designs takes up a large part of the chip space.

Looking at the pictures I assume that the second core adds only about 20% extra real-estate.

I did not find a thermal picture of a current AMD chip but assuming that the switching speed is the most significant source of power consumption, then lowering the speed (200 MHz) probably cancels out the extra power consupumtion to a large extent.

The new Venice architecture with its power consuption and speed optimizations will do the rest (as mentioned in the other posts).

Re:Cooling

[bbbl67]

Simple, they run the dual-cores at a couple of Mhz increments lower than the fastest current single-cores. That way the two slower cores together don't use up much more Watts than the fastest single core out there.

Re:Cooling

[be-fan]

Wow. Not a single correct answer to this question :)

The answer is that the single core Athlon in the article runs at 1.5 volts, while the dual cores run at 1.26 volts. Since power increases by the square of the voltage, we can see that the 1.26 volt chip uses only about 70% the power of the 1.5 volt chip. The remaining 20% comes from optimizations to AMD's 90nm SOI process.

Re:Cooling

[be-fan]

Oh, one more thing. It also runs at 2.4GHz vs 2.6 for the single core chip. Power is linear to clockspeed, so the dual core processor uses 92% of the electricity of the single core processor based on that. 92% of 70% is about 64%. The remaining 14% is from the process optimizations.

Anand's Take

[Anonymous Coward]

Here's Anandtech's review of the X2.

Re:Anand's Take

[Zocalo]

Trusted Reviews has its review up to, also being mirrored by The Register.

Re:Anand's Take

[Zocalo]

Oops. Pasted the wrong link for Trusted Reviews; the correct direct link is here, and here's a link to Bit tech's review with a photo of the chip in question with its cover off, for those who get excited by such things... :)

What's that burnt intel smell?

[MaceyHW]

will they be able to outmarket AMD again?

Re:What's that burnt intel smell?

[eyegone]

Yes.

Re:What's that burnt intel smell?

[ackthpt]

will they be able to outmarket AMD again?

Intel is obviously relying on fat vendors like Dell, but with performance like this and power consumption like that, buyers will be asking Dell what their problem is. When Dell finally cracks, you'll know Intel have spent too long fixating on their stock price rather than their products. It's a tough thing to recover from, too, and will call for a major shake-up.

Pity is, companies which go though this usually are considerably weaker. AMD looks good, but you have no idea what may be coming out of Japan/Taiwan/China in 10 years.

I'll wait for the next version...

[Eagle5596]

While the AMD 64 X2 Dualcore is impressive, I am still waiting for the AMD 69 XXX Hardcore myself.

Sorry, it just had to be said.

Re:I'll wait for the next version...

[RealProgrammer]

>AMD 69 XXX Hardcore

They only come in pairs.

Also, after you've had it a while one of the pair refuses to go into "69" mode, even though it's still fine in standard mode. Something about being incompatible with a daughtercard.

Re:I'll wait for the next version...

[l0b0]

I'd like to have a 30 GHz myself, just to pronounce it "triple-x jeeze".

Re:I'll wait for the next version...

[Prof.Phreak]

Beh, I'm waiting for AMD X3 86 myself.

Re:Mmm.... dual core.

[maharg]

forget longer pipelines, it's girth that really counts

Row, row row your boat, gently down...

[Anonymous Coward]

No actually, they're going to be launched in June. The fact that this would be lost on the submitter was so obvious, I was able to prepare this message in advance and just paste it in.

These look to be amazing CPUs. After the initial linpack-with-large-matrices benchmark, you have to go thirteen pages into the benchmarks at TechReport to find some of note where the Intel solutions are able to score off a win!

Don't Forget the [H]

[Unholy_Kingfish]

The cold dark [H]OCP also has their preview up.

Or you can jump right to their conclusions.

Re:Don't Forget the [H]

[Ziviyr]

It seems to me that several of their written conclusions are based on the poor multitasking scheduling of Windows, or his inability to use task priority settings.

I like powerful CPUs, I don't like validating their purchase with "I'm too lazy to push this slider to low/idle priority".

Rollout process

[fbody98]

I'm relieve to see at least one thing out of this launch, and I would hope that other companies would do as much. AMD has clearly defined their rollout process so there will be no confusions and hopefully no false expectations.

1. Announcement
2. Technical Preview (benchmarks Appear)
3. Launch (OEM Availability)
4. Ramp-up and Reseller Availability

They even give dates, if they can keep to those dates then we might actually have a product launch that doesn't antagonize the community with accusations of a 'paper launch'.

I'd like to see more companies be more upfront about this.

Re:market for this?

[Anonymous Coward]

More power means less needed optimisations! Programmer's dream :-)

Re:Redsigning your applications.

[mattmentecky]

That a lot of applications are not multithreaded. Thus wont get the speed advantage of the Duel-Core processor.

Thats because the two cores are too busy fighting.

Re:Redsigning your applications.

[Enigma_Man]

My spelling and grammer combined with the fact that I have college degree, proves a problem with the educaion system.

duel-core

Wow, couldn't have reinforced that point any better. These cores don't fight eachother, there are just two of them (duel/dual).

-Jesse

Re:Redsigning your applications.

[BigGerman]

it does not have to be the same application. I have seen the environments where the business users are crippled because they have over-agressive anti-virus running in the background. Their apps would fly with a dual-core (or even hyperthreading).

vs

[Doc Ruby]

This is all getting very complex in the "Pentium compatible" world. Where's a chart of direct CPU performance comparisons across manufacturers (Intel, AMD, etc), so I can look up a potential purchase? Eg, I see that PriceWatch has an "Athlon XP 3000" at $102, and a P4/2.26GHz at $111. How much faster/slower will my LAME encoder server run for the $9 difference? At the very least, where's a chart showing which makes/models are direct competitors?

Re:vs

[fbody98]

The best example of what you're looking at that i've found is at http://www23.tomshardware.com/index.html

It's an interactive chart of all major processors available now and plenty that aren't available, it's a good idea to compare what you might have not and what an upgrade could do for you.

Re:vs

[Anonymous Coward]

Try here

http://www23.tomshardware.com/index.html

Not everything you want, but a good start.

Re:vs

[Doc Ruby]

That is interesting, though my listeners have only MP3 players. Is there a comparison chart of encoder/format performance? I need to know about 128Kbps MP3, but other formats could be interesting for context.

Re:vs

[fnj]

That chart doesn't have any Pentium M's in it. What I'm looking forward to is the Yonah.

Re:vs

[loose_cannon_gamer]

Don't forget, there's more than the price of the processor to consider. I just upgraded to the top of the line Asus socket A motherboard for an Athlon XP 3200+, and their top of the line socket 478 motherboard (for a P4 3.2 northwood, for example) is twice as much, similarly featured (roughly $90 vs. $180, for an A7N8X-E Deluxe vs. a P4C800-E Deluxe).

In fact, it was this fact alone that pushed me to AMD over Intel.

Re:Redsigning your applications.

[natrius]

What a lot of people dont realize (Including a lot of programmers). That a lot of applications are not multithreaded. Thus wont get the speed advantage of the Duel-Core processor.

Correct. Instead of executing the code in parallel, both cores will fight to the death for the privilege. Since only one core survives, you don't really get much benefit from duel-core processors.

Wow... those are fast

[tinrobot]

I do a lot of reviews of dual processor machines for publications that cover 3D animation and graphics. Usually the dual processor machines kick the single proc machines to the curb in every test. Dual CPU machines also give better interactivity, and the machines we use at the studio always are dual cpu for that reason.

This really is going to make me think twice about the need for separate CPUs. I really want to get my hands on one of these to test.

Re:market for this?

[59Bassman]

Who will EVER use more than 640K?

The fact of the matter is if you build it, they will come. I'd bet that it won't be more than a couple of years before you see a recommendation for 2 processors on games.

Re:whitebox linux on dual core Opterons

[Anonymous Coward]

This is the guy who's selling the domain trying to get more money for it. He's even put an anti-Intel/MS comment in there in a vain attempt to get his post modded up.

Keep this spamming fuck at -1, and certainly don't bid for his domain.

Cooler than the old AMD 130nm designs and Intel

[Brian Stretch]

Sayeth Anandtech: ...the Athlon 64 X2 will consume less power than a 130nm Athlon 64, and less than 20% more power than a 90nm Athlon 64. Note that the Athlon 64 X2 4200+ compared here also consumes less power than all single core 90nm Intel Pentium 4 CPUs, even the Athlon 64 X2 4800+ consumes less power than all single core 90nm Pentium 4 CPUs.

Re:whitebox linux on dual core Opterons

[masklinn]

Crappy ad for a crappy distro by a crappy AC in a thread about processors sounds as off topic as one can get, even for /. standards

Re:market for this?

[Stibidor]

If the general public can be convinced that these new dual-core chips are "better" than the old single-core stuff (not hard to do in a culture always seeking the so-called "latest and greatest"), then there is certainly a market for this. People don't always buy things for practical purposes.

Besides, in an industry where if you don't come out with something new frequently you die, it seems likely that it won't be too many years down the road before dual-core may be the only option for consumers in the market for a brand new machine.

Just my two cents...

It's about the interactivity

[tinrobot]

A dual CPU machine provides such a smooth operating environemnt. Never hiccups or pauses. I'm hooked on them. I hope dual core provides the same interactivity.

Re:It's about the interactivity

[snookerdoodle]

Hit the nail on the head.

Certain image process apps (e.g.: enblend, a panorama enhancing app) drive my single athlon into a state where it is sluggish and hard to use (1.8ghz athlon, 768 mb ram, fedora core 3).

So, while the Gimp can be compiled to do some multiply-threaded stuff, the real boost is that my computer should still be useable for other things while it's off fixing my panoramas.

Since some of my panoramas take over 2 hours to fix, I'm looking into a faster system and will definitely be trying to make sure I can get something that lets me plug one of these guys in when they become available.

Mark

Re:It's about the interactivity

[D. Book]

What about the operating system's role in multitasking efficiency? Just a few days before joining the latest dual-core drumbeat, Scott Wasson of The Tech Report posted the following item:

We've already asked you for some input on our possible multitasking tests, but let's talk for a sec about that creamy smoothness that comes from having multiple processors in a well-tuned system. I've said many times that it smoothes over potholes and allows the user experience to feel friction-free. In fact, if you pick up the latest copy of PC Enthusiast magazine, my column this month extols the virtues of dual-core CPUs for multitasking. I use an example of a problem with my own PC slowing down to a halt while checking mail, caused by the convergence of too much client-side spam, virus, and mail filtering. Dual-core processors should make problems like this almost a thing of the past.

However...

After writing that article, I decided to troubleshoot the mail-checking slowdown problem one more time, and I realized that I hadn't applied some basic tweaks to this installation of Windows XP Pro. Once I set the OS scheduler preferences to optimize for "background tasks" instead of "applications," my mail problem was largely resolved. I also used registry tweaks to increase the size of the system disk cache and to disable paging of the Windows executive, and all told, my system is much more responsive now.

Now, I still think dual-core CPUs will be a great thing for multitasking, but this raises the question: How much creamy smoothness can you squeeze out of a box with only one CPU, with or without Hyper-Threading? And what proportion of PC slowdowns and performance "hiccups" are really caused by inadequate CPU power as opposed to lousy OS scheduling, hard drive bottlenecks, running out of RAM, lousy drivers, or the like? Is multitasking nirvana really just a second CPU core away? What, in your experience, has the most impact on your PC's responsiveness, and what upgrades have helped the most?

Unfortunately, it seemed the question was mostly rhetorical, as The Tech Report prompted their users to "discuss" the issue subjectively rather than getting some multitasking benchmarks going to back up the anecdote.

Re:It's about the interactivity

[ArbitraryConstant]

"I hope dual core provides the same interactivity."

I imagine it will, but I don't have any interactivity complaints on Linux 2.6.

I had a Pentium 4 with HT, I noticed a big interactivity boost on Windows but on Linux I couldn't really tell the difference between HT enabled and disabled. I often ran with HT disabled because it was faster for the stuff I was doing. I can't stand it when stuff happens like music skipping (the reason I never used Linux on the desktop pre-2.6), so I'd notice if it happened and it's been fine.

My P4 motherboard kicked it and I'm now on an Athlon64 3000+, and it's also fine for interactivity purposes.

Everything pre-2.5 was pretty bad for interactivity, but 2.6 is excellent. That's why I put up with bugginess. I'm no Linux zealot, I bitch about the buginess of 2.6 a lot more than I praise its interactivity, but it just happens to do that one thing really well.

Obviously not everyone can use Linux and this is a big win for them, and it's a big win for Linux users in plenty of other ways. But it's really amazing how terrible Windows is at providing a responsive system.

Re:It's about the interactivity

[maraist]

Everything pre-2.5 was pretty bad for interactivity, but 2.6 is excellent.

Don't really share your user experience.. Being a college student at the time, I salivated over a dual system for years, and finally found the opportunity with the dual Pentium II-class celeron motherboards by Abit. That brief window in history when you could have a full dual processor system for under $250. It was dual 433's overclocked to 466. At the exact same time, I had an AMD K5-400 as my main machine. The dual ran on Red Hat 8 I believe originally, and is still running to this day as my main home-server, and I've only upgraded it to Red Hat 9.0 (main because if it ain't broke, I ain't fixen it). So as I look at uname, it's still only running kernel 2.4.20-31.9smp. I remember running this puppy side-by-side to the K5-400, and later my K6 Thunderbird 800MHZ. The Thunderbird should have blown away with dual 466, but it didn't. I had better throughput of mp3 encoding on the duelies (which I was doing a lot of at the time, farming out all my machines at home and at work 24/hours a day some weeks). One of the features I specificly played with was encoding single-threadedly with grip+lame v.s. dual threadedly. When dual threaded. Obviously single-threadedly the system was almost perfectly responsive (since lame isn't HD or even memory bound), but even when dual-threaded, the system was more responsive than my faster single-CPU K6. I quickly fell in love with the dual processor concept, and used it as my main home-station for just about anything that wasn't video games.. When my K5 literally exploded one day due to moisture damage, I was rather forced to migrate over to the new machine; but it was a welcome change from a mostly windows unresponsive environment.

I am convinced that even Linux 2.4 was more smooth operating with multiple CPUs than windows. Perhaps it is because X is single threaded, but graphical thinking occurs in the application-space and is thus inherently multi-processed. Thus you get the best mixture of non-race-conditions streamlined code with concurrent processing capability. This is purely speculative. Whatever the deal, it was great.

Unfortuntaely, I don't remember if the standard Linux benchmark of doing a parallel make of the kernel was faster on the dual 466 v.s. the single 800. I guess one of these day's I'll have to fire that 800 back up again to check; the dual's still chugging along fine as my server.

Unfortunately I haven't had the luxury of having ANY affordable dualies in the past 5 years, so I've just gone for greater single-threaded horse-power for work-stations.

As for the point of this thread. I seem to recall that the 2.6 kernel had more overhead than the 2.4 kernel. This along with my anxiety for changing a back-end special-purpose servers' OS kept me from wanting to up the now ancient machine. Most likely this overhead is compensated for by the better MT-support, and is especially unnoticeable at the 2GHZ range. But I find it hard to believe a perceptible difference in UI responsiveness could be found between the 2.4 and 2.6 kernels. Perhaps measureably in application benchmarks, but surely not on the GUI.

Sadly, as I've said, I can not provide empirical data as I don't have $1,500 to spend on a simple file-server.

Re:It's about the interactivity

[ArbitraryConstant]

"Don't really share your user experience."

It should be noted that Red Hat has now backported several interactivity improvements from 2.5/2.6

My first attempt at having a Linux desktop machine was on a PII-400, with a version of Slackware that I don't recall (I do recall that it had a 2.2 kernel with the option of a 2.4 kernel circa 2.4.10, I used 2.4). I didn't care about throughput, but I wanted a system that was responsive. I define responsiveness as the ability to respond to me and programs I care about (like XMMS) with a small delay. I'm not sure we're talking about the same thing, because you seem to be talking about thoughput in CPU-bound tasks.

For example, if I was browsing slashdot while compiling the kernel, the mouse might stop updating for a few seconds, and XMMS would start skipping. Because of that, I went back to Windows as my desktop machine and telnetted to the Linux machine. The cause of this was the scheduler. It didn't give enough priority to processes and threads that the user would notice if they were delayed even a few milliseconds. That's why dual-CPU systems were more responsive, they had a greater probability of being able to run a thread when it needed to be run.

2.6 is much better about this. Threads and processes that consistently sleep before their CPU time is used up (either through sleep calls or system calls that block) are run as soon as they become runnable.

2.6 does have more overhead, but it's faster in the one way that matters to me: it remains capable of responding to me with very little delay no matter what is going on in the background. It might have slightly lower throughput, but if I can use the computer while it's busy doing other things, it doesn't matter to me.

I've used it on Pentium 4 systems with hyperthreading, which has most of the same responsiveness advantages. I can tell the difference on Windows, but not on Linux.

Re:It's about the interactivity

[ArbitraryConstant]

I can't go through the system and nice anything that might potentially consume lots of CPU. That's just not practical.

Re:Redsigning your applications.

[mellon]

A lot of the CPU burn with AV software involves doing I/O or scanning memory, neither of which are speeded by a dual-core processor. So you might get some speedup from this, but it won't be the difference between sluggish and speedy - it'll be sluggish versus less sluggish.

Re:market for this?

[Blue-Footed Boobie]

SERVERS!

Nothing sexier than a 16slot Blade server running Dual-Core Opterons. Equivelent of 32 cpus in 5u of space.

MMMmmm...mmm...mmm...mmm...SEXY!

Re:Redsigning your applications.

[masklinn]

What a lot of people dont realize (Including a lot of programmers). That a lot of applications are not multithreaded. Thus wont get the speed advantage of the Duel-Core processor.

And what YOU don't realize (including... duh... yourself?) is that running two or more applications at the same time will make use of dual core system, even if the apps themselves are single threaded (which is mostly true for games, quite a lot of desktop apps are at least a bit multithreaded).
And a singlethreaded badly written application will be less prone to lock your computer, too, since the other apps will still be able to run from the second core.

The main issue is not the multithreading abilities of the applications, but the multithreading abilities of the OS itself. If the OS handles multithreading well, multicore (physical or virtual) will always give a slight to impressive improvement over single core.

Re:Redsigning your applications.

[EpsCylonB]

Any programmer that doesn't understand what multithreading is isn't a very good programmer.

Adware&Virus: hardware makers win!!

[amcdiarmid]

The market for this is everyone who uses an agressive anti-virus program. The AV will run on one prcessor, what you are doing on the other.

It's a sad case that as malware becomes more previlent, hardware vendors win. Really, you can be productive with (for example) Win2K on a 1GHz machine and 256MB, in an office. Now add the wait as every file is scanned on access for viruses (per corporate policy), and the machine somehow becomes "too slow."

OH well. I guess it's time to put all productivity applications on a Server & run them remotely. Again;-(

Re:Adware&Virus: hardware makers win!!

[EXrider]

And this is teh #1 reason why I don't use Windows.

Re:Adware&Virus: hardware makers win!!

[swillden]

The AV will run on one prcessor, what you are doing on the other.

The only problem is that most people only have one disk, so your AV will still bog down any disk accesses you need to make, and that is bigger than the CPU hit. Using a pair of mirrored drives would mostly solve that problem, I suppose. The AV would slow down writes, but reads should still be fast (assuming you only have one other program doing reads).

Oh, and the AV program would probably still end up filling your disk cache (whatever spare RAM you have) with the data it reads for scanning. On most OSes, once the disk cache has consumed all free RAM, the OS will start looking at what other stuff in RAM hasn't been used lately and swap that out to disk, so as soon as you click back on an app that you haven't used in a bit, it'll have to be swapped back in.

Hmmm... what's really interesting about that is that data is cached in RAM on the theory that it is likely to be needed again soon, but the AV program only scans once, and won't look at any of that data again. If the AV program could tell the OS "Don't cache the files I'm opening", then the OS could avoid wasting RAM on files that won't be touched again. Do any operating systems provide an API for that?

With an OS that supports non-cached reads, an AV program that requests non-cached reads, mirrored drives and two CPU cores, you could *almost* work uninterrupted right through an AV scan!

Re:Adware&Virus: hardware makers win!!

[amcdiarmid]

AV will bog down disk access. However, *if* it reads the file to memory, then this will not be as big a hit. A RAID array would help, but is prob. out of the league of most users.

Otherwise good comment. (I assume you meant raid array, and not a mirror anyway.)

Re:Adware&Virus: hardware makers win!!

[swillden]

AV will bog down disk access. However, *if* it reads the file to memory, then this will not be as big a hit.

I disagree. Reading the data into memory is a bad idea. It's much more efficient to read the data from the disk a small chunk at a time (just big enough to scan) and to convince the operating system not to bother caching such reads, because we know that the data will not be needed again.

A RAID array would help, but is prob. out of the league of most users. Otherwise good comment. (I assume you meant raid array, and not a mirror anyway.)

Mirrored == RAID-1.

Usually, when people say RAID, they're referring to RAID-5, which helps read performance some but in this case, I don't think. In general, striping (RAID-0) provides the best read (and write) performance, but I think it would also not work as well as mirroring in this case. Why? Well, the reason striping (and, to a lesser extent, RAID-5) helps read performance in general is because each read pulls part of the data from each drive. That is, if you're reading 256KB and you have four drives striped together, each drive will only have to find and send 64KB. If they can all do this in parallel, you can get your 256KB approximately as fast as one drive could give you 64KB.

But that assumes a single process is doing the reading. In the AV case, you have two processes, doing different reads. The AV is gradually working its way through the entire disk, while the user process is doing something else. With mirroring, for each read the OS (or RAID controller) will decide which drive to actually pull the data from; typically it will pick the drive whose head is currently nearest the data. That way, the user process can be reading from one portion of the disk while the AV process is reading from another portion at the same time. With only a single disk the head has to skip around to support both. With striping or RAID-5, since every read has to come from all of the disks (well, all but one for RAID-5), an AV process plus another process will require the heads to skip around constantly, though the data will actually come faster once the heads are in position.

I haven't tested it, but I suspect that because seek times are so large, net performance with an AV+other process workload would be better with mirrored disks.

I have a machine at home with RAID-0, RAID-1 and RAID-5 on it. It doesn't run any AV, but I could probably do something like starting a process to cat all files to /dev/null that would effectively do the same thing. Maybe I'll try it sometime. Unfortunately, the process catting all of the files would blow the disk cache and slow down other processes that way. Without that, I think mirrored disks (and dual procs) would make the performance hit almost unnoticeable.

Does dual core == 2xProcessor or hybrid?

[amichalo]

Does dual core have to mean 2 of the SAME processor?

I recall reading a /. comment on a previous news day that suggested using dual core to allow the OS and anti-virus software run on one proc, while applications share another, thus improving stability/security/performance.

But does a vendor HAVE to make a dual core chip with two of the same processor? Perhaps gains could be made using a less powerful, commodity chip core and pairing it to a top of the line core.

Costs would be lower and they could sell more of this hybrid dual core because they would only need 1 top of the line cores.

Oh, you get what I am saying.

Re:Does dual core == 2xProcessor or hybrid?

[DarkBlackFox]

That makes sense from a logical standpoint. The problem is, virtually all cores in any given line^1 have the same production cost. The core of an Athlon 64 2800+ cost the same to produce as an Athlon 64 4000+. Essentially, all cores are the same. After fabrication, they are tested at each power level (4000, 3800, 3500, 3200, etc) and typically marked as the highest stable rating. Sometimes if the market demands more mid-low end chips, some of the higher rated cores will be re-marked to lower rated cores, and sold as low end CPUs, which is why some people have great luck overclocking certain CPUs.

The point is, it wouldn't matter if you plant a 2800+ core with a 4000+ core, because as far as AMD is concerned, they cost the same to produce. The bigger question for them is what the market value is for a CPU of any given speed, e.g. how much would a dual core CPU be worth.

^1 For a dual core setup, I'd assume it can really only be done with cores of the same processor line, e.g. Athlon 64 vs. Athlon XP. The 64's have an onboard memory controller and Hypertransport, Athlon XPs require a memory controller in the chipset and don't support Hypertransport. It'd be like trying to mate an orange tree to a potato plant.

Re:Does dual core == 2xProcessor or hybrid?

[Fittysix]

While it probably doesn't no, i would think it's better off this way, i belive the clock across the entire processor is synchronus. I'm no engineer but i would think communication between the 2 processors would become a difficult task if the 2 weren't the same, also, the a64 core was designed from the start to be used in a dual core application, so a different core would somewhat negate the advantages.

Re:Does dual core == 2xProcessor or hybrid?

[realmolo]

I'm sure that would be possible, but you'd need all kinds of other support chips to make it work. When both "cores" are the same, it's easy to schedule/distribute things between them. When they're not, now you have to figure out which processes are high-priority, and should run on the faster core, and which processes are low-priority, and should run on the slower core.

Which, needless to say, is probably pretty damn hard to do. So hard, that it'll never happen.

Re:Does dual core == 2xProcessor or hybrid?

[jlebrech]

I thought that both cores where on the same die!!

Re:Does dual core == 2xProcessor or hybrid?

[Wesley Felter]

But does a vendor HAVE to make a dual core chip with two of the same processor? Perhaps gains could be made using a less powerful, commodity chip core and pairing it to a top of the line core.

But the less powerful core does not exist, so they'd have to design it. And the design cost is killer.

However, assuming unlimited design budget and schedule, there are some academic papers showing that heterogeneous cores are a good idea.

Re:Does dual core == 2xProcessor or hybrid?

[bbrack]

The cores on a processor do not have to be identical, but having different cores would GREATLY increase the difficulty of the design/debug process.

Having to do (in effect) 2 separate processor designs, then ensuring that the arbitration logic works correctly, the list just goes on and on. Having both cores identical greatly reduces design costs.

In reality, manufacturing costs are not that different from a top of the line core and a commodity core, one just bins out at a higher speed. What might be doable is running one core on one clock multiplier, and runnning the other core on a different (much lower) clock multiplier, and disabling the faster core when the proc is under a low load...

Re:Does dual core == 2xProcessor or hybrid?

[kesuki]

dual cores but only one bus.. So you can only have one front side bus multiplier, which means both cores run at the same frequency, with the same multiplier. It's true, in a dual CPU board you can have one high end CPU, and pair it with a 'cheaper' cpu, (if the bios allows it) however, the dual core cpus work in the same motherboards as their single core counterparts, and they share a single socket, and that socket doesn't have pinouts or a configuration allowing seperate halves of the dual core to run at different frequencies. They also share a memory controller, making it even less possible to have the cores on different frequencies.
The bonus is that dual core cpus will work in standard, commodity single cpu boards, there is no need for an extra expensive dual cpu board to run them. and dual cpou boards are less complex than quad cpu boards, and so on and so forth.. so dual core is a good idea, and they've already done as much as they could think to do to make it work as good as possible.

Re:Does dual core == 2xProcessor or hybrid?

[kyojin the clown]

mmm orangey potatoes...

Re:Does dual core == 2xProcessor or hybrid?

[Wesley Felter]

What might be doable is running one core on one clock multiplier, and runnning the other core on a different (much lower) clock multiplier, and disabling the faster core when the proc is under a low load...

Some of my coworkers did some research showing that the best thing to do for server apps is to run both cores at the same frequency, and adjust the frequency to match the workload. For desktop machines, I suspect you should either run both cores at full frequency or disable the second core (as in Yonah). I think there are few situations where you'd want different frequencies.

Re:Does dual core == 2xProcessor or hybrid?

[jericho4.0]

Yes, they have to be the same processor. The chipset design assumes that both processes are equally capable. Last time I built a dual processor machine, it was recommended that the CPUs have the same stepping number (same batch off the line).

Re:Does dual core == 2xProcessor or hybrid?

[Anonymous Coward]

It'd be like trying to mate an orange tree to a potato plant.

or maybe a tomato plant with a tobacco plant?

Re:Does dual core == 2xProcessor or hybrid?

[fitten]

Define "lose overall". With asymmetric processors, you have to worry about compiling the code multiple times (one for each architecture), loading the right executables onto the right processors, and having all the code in-place to deal with communications between different architectures. That's a lot of work.

Re:Does dual core == 2xProcessor or hybrid?

[TheRaven64]

Not only that, but scheduling algorithms for heterogeneous processors are a whole lot more complicated than those for homogenous sets (see the problems with getting good performance out of HyperThreading for an example). It might be possible to do something fairly simple, like run all processes on the slow processor and migrate them to the fast one when they use more than a certain percentage of the CPU speed, but in this case why not just down-clock the faster one when it is not in use?

The only time when heterogeneous processors are really useful is when each is better than the others at a sub-set of tasks. Current PCs are usually a set of 3 different processors in a single box[1]. They have a reasonably fast general purpose CPU, and on the same die a simple vector processor (e.g. MMX, SSE, AltiVec), which has a different instruction set to the main processor and must be invoked explicitly. They also have a highly parallel large vector processor on a separate chip, which is usually used for graphics. No automatic scheduling is performed between these - it is up to the programmer to explicitly code for each one. Ideally, a heterogeneous processing environment would require code to be JIT compiled for each processor, and then moved between them depending on run-time profiling information.

[1] Yes, this is an oversimplification.

Re:Does dual core == 2xProcessor or hybrid?

[ExoticMandibles]

[...] there are some academic papers showing that heterogeneous cores are a good idea.

That's what the IBM/Sony/Toshiba "Cell" architecture is all about. It has one "real" PPC core and eight mini-PPC (SPEs) cores all in one processor. You can read more about it here.

larry

Re:Does dual core == 2xProcessor or hybrid?

[wisdom_brewing]

isnt that sort of like what the cell is? one main processor and 8 'assistants'?

Re:Does dual core == 2xProcessor or hybrid?

[Wesley Felter]

I was talking about heterogeneous microarchitecture but homogenous ISA chips (e.g. one big x86 and one small x86). Heterogeneous ISAs look good for embedded systems but I don't think they'd work in the desktop/server market.

Re:Does dual core == 2xProcessor or hybrid?

[kalgen]

But does a vendor HAVE to make a dual core chip with two of the same processor? Perhaps gains could be made using a less powerful, commodity chip core and pairing it to a top of the line core.

But the less powerful core does not exist, so they'd have to design it. And the design cost is killer.

What about dual-socket setups? Can you pair an Opteron 152 with an Opteron 275 and get a similar effect?

Re:whitebox linux on dual core Opterons

[masklinn]

Yes, a point that's been valid ever since last year.
Doesn't make his post less offtopic

Re:Mmm.... dual core.

[fbody98]

Is that why I keep getting SPAM trying to sell me and Ultra-Fast Wide pipeline?

Re:market for this?

[jtpalinmajere]

For Intel, your argument definitely holds water. Their whole business plan has been based off of their vast number of production plants and relatively cheap process of putting hordes of chips on the market... hopefully making the chips pervasive enough to strike a profit level in the end (more like early mid-life with their price schemes though). It is only after a processor has been found tried and true that Intel migrates it to server land.

AMD on the other hand has always started out chips on the enthusiast / enterprise market because they simply don't have the fabrication capacity that Intel does. Thus they market first for the high end users and over time the processors find their way into the desktop market when they've been dated by yet another new, improved processor being marketed at the first group. Their whole revenue plan is based off of the 'rich' people niche (which includes many medium to large businesses). Based on their success, I'd say that they've done really well with this business model and continuing to do so would likely continue to work for them.

The common misnomer that is latched onto with many processor reviews nowadays is that both AMD and Intel are prodcing processors for the desktop platform, when in reality their business goals for their processors are on opposite spectrums. Intel starts desktop side, AMD starts server side. It is only after both have matured to some degree (and software caught up to both of them) that the processors can be meaningfully compared for the average joe user that just bought a new computer (or had one built for him).

Most people who go crazy over these new technologies are either wanting it for pure bragging rights, or simply aren't aware of how little it will actually do for them... or both in all liklihood.

Re:market for this?

[masklinn]

Even 16 slot blade servers running octo dualcore Opterons systems?

YeS!

[fredan]

An 32-bit comparison of an 64-bit processor. This is exactly what I look for when I need to know which cpu to buy...

Highlander!

[Anonymous Coward]

There can only be one core!

Re:market for this?

[Malc]

If they ever make it to a significant size of the market, you will see more of the CPU intensive tasks that people do today becoming multi-threaded. Some of the of long running processes that are common on home computers lend themselves nicely to divide and conquer, such as ripping music or video. By going dual core or SMP, one can halve the processing time without having to wait a few years for the processing power of CPUs to double.

The new Anandtech article

[Kobun]

Myself, I enjoyed reading the new Anandtech article that went up today. The new AMD CPU's are put through their paces, and are compared against the best Intel has to offer. For some good top end (or dual core, as it has become the same thing) comparisons, this is a good place to start.

http://www.anandtech.com/cpuchipsets/showdoc.aspx? i=2410

Fast and INEXPENSIVE to run!

[Senor_Programmer]

"... AMD's X2s consume no more power than single-core chips."

This is significant if you live in say Honolulu where electricity is 14cents/KWh or on Kauai where it's close to 22cents/KWh.

Re:Fast and INEXPENSIVE to run!

[myukew]

a year has 8760 hours. if you leave your pc on 24/7 and it's always under maximum load it consumes, say 400W, 400J/s. this is 0.4kWh/h, that'll be 3504kWh/year or 770.88$ (an kauai) 490.56$ (on honolulu). however you'll probably get away with 100$/year...

Re:Fast and INEXPENSIVE to run!

[Jedi Alec]

Holland: close to 20 eurocents/kWh.

Re:Fast and INEXPENSIVE to run!

[otis wildflower]

This is significant if you live in say Honolulu where electricity is 14cents/KWh or on Kauai where it's close to 22cents/KWh.

Last time I looked, that's about on-par with NYC (given fuel surcharges).. Too bad we don't get enough sun for solar :/

Re:Fast and INEXPENSIVE to run!

[MikeBabcock]

Its even more significant if you're buying a few million of them.

Think Google.

Re:Fast and INEXPENSIVE to run!

[hackstraw]

This is significant if you live in say Honolulu where electricity is 14cents/KWh or on Kauai where it's close to 22cents/KWh.

But everything else is dirt cheap in Hawaii because a majority of the US population works and lives there.

Re:Fast and INEXPENSIVE to run!

[wisdom_brewing]

In the UK electricity costs a bit more in honolulu. At least here you have the advantage of using the computers as multifunction tools, pc and heater in one... Not very efficient but at least its something.

Re:Redsigning your applications.

[robnauta]

If you use Windows, open the task manager and go to the 'processes' tab. Choose View - select columns and tick the 'threads' column.
You'll see that almost all running processes use threads. Only the tiniest systray apps may be singlethreaded. Apps like Internet Explorer use dozens of threads, and will render a page with many jpg's or flash applets faster than a comparable single CPU/single core system.

The 'not many apps use threads' myth keeps on being spread, but anyone can see for himself just how many apps use threads.

Re:market for this?

[segmond]

You are right that the market for dual core processors for home users is really not there, but it could be. Think of the BeBox/BeOS, It was a system that was designed for dual core processors and all applications compiled for BeOS automatically benefited from it. What is missing in the x86 world for home users is such an OS. I believe that Intel/AMD are well aware of this, and this belief leads me to the conclusion that they are not really pumping out such systems for the average home users, it is more for businesses. Internet/Enterprise servers will definitely benefit more from it.

Longhorn

[yabos]

Have you seen the specs required for Longhorn!

Re:Mmm.... dual core.

[millennial]

Your spam is selling you ultra-fast pipelines? Wow. Mine tries to sell me "pipelines that are slower so they last longer."

Re:Redsigning your applications.

[Jeff DeMaagd]

What you don't understand is that hyperthreading, dual core & dual processing is good for multi tasking. Some people actually run more than one app or service at a time, and the computer will be more responsive. It looks like AMD has no wattage penalty for this, which is good.

If readers didn't get the above replies, duel and dual are two very different words. Dual is a word used for "two", duel is a word for a fight.

I think it is a chicken and egg thing, developers of performance intensive software will try to optimize for multicore if there is demand for it.

Re:market for this?

[be-fan]

Programmers? Multimedia people? Scientific computing folks? There are quite a few markets that can use dual-core right now. Basically, anybody who buys a PowerMac :) Moreover, in the future, everyone will have to move to dual core (including gamers), because AMD and Intel cannot ramp up the clockspeed of single core chips. So AMD's strategy makes quite a bit of sense. Sell dual-core chips to the high-end now (notice how all of these CPUs are high-end chips that carry quite a price premium), and start getting the ball rolling on multithreaded software.

Re:Redsigning your applications.

[smackjer]

And any programmer that doesn't understand what multithreading is, but tries to do it anyway, is a very very BAD programmer -- and generally much worse than a programmer who avoid MT altogether.

Windows Licenses

[Timberwolf0122]

As I understand it M$ only allow upto 2 cpus on a standard licence. I hope they will release an update to allow for 2 dual core chips.

Re:Windows Licenses

[QuiK_ChaoS]

Acording to this month's Computer magazine Microsoft won't be charging extra for dual cores.

"Microsoft and Novell agree and don't charge extra for using their software on multicore processors."

"BEA Systems and Oracle, on the other hand, charge more to use their software on multicore chips for per-processor licensing."

Re:market for this?

[Austerity Empowers]

Gamers need dual core processors. Many games are still single threaded but that's probably the past. There is plenty to be gained from a single game running multiple threads. The only concern is that RIGHT NOW I don't think Windows will do The Right Things. I could be wrong.

The question is, other than gamers and graphic artists, who needs them? You have a point in that almost every other application that the average guy uses has been saturated in terms of (quite prolific) features for years. I really don't think that will change much, hence MS is having a hard time maintaining the "Buy a New Office Suite every 3 Years" business model.

More and better multimedia applications MAY be the next killer app that requires this power, but a significant amount of work needs to be done to make tools for this accessible to the masses.

Check out Linux

[Nailer]

ps -eLFwww

A lot more common apps are multithreaded than people think. Nautilus, Firefox, OpenOffice, Gnome Terminal, and, um Gnome Weather Applet are all mutithreaded.

Even if no apps on your system are multithreaded, if you're like the 99% of users who run multiple processes simultaneously, you'll still get an advantage. Your updating app runs on one core while your desktop runs on another, for example.

Re:Redsigning your applications.

[UrgleHoth]

Let me guess, if one gets hurt, is it only a flash wound?

Re:Redsigning your applications.

[Xoro]

Dual CPU systems tho are useless to the home users, it's for businesses and scientists with more computing need. Real enterprise applications are multithreated.

Not so!

I was one of the lucky people buy a cheap dual Celeron setup right after that hack was first discovered and I can tell you that multiprocessors on the desktop rock. My old system was a dual Celeron 400, and while it couldn't compete with a modern system in terms of benchmark speed, it had my current 1400 MHz Celeron system beat bloody when it comes to interactivity and responsiveness -- that elusive "feel".

The price is steep now, but don't let arguments about application benchmarks dissuade you from trying out multicore when prices go down. The Anandtech review cited about has some really telling benchmarks about how well a dual system performs when loaded down with multiple tasks.

Unlike the unnoticeable 200 or 400 MHz incremental bumps you usually see with processors, dual core really brings something of value to the desktop user. Try it and you'll see.

Re:Funny from TFA:

[Enigma_Man]

Hahaha, maybe it's just the dayquil talking, but damn that's funny.

-Jesse

Re:Redsigning your applications.

[EpsCylonB]

Oh I agree, I avoid multithreading because its a bit beyond me, but the concept of multithreading isn't difficult to grasp. Any programmer that doesn't understand it and its relevance to dual core/CPU computers shouldn't really call himself a programmer.

RISC

[delirium of disorder]

Packing all this circuitry will cost more in heat and fabrication costs then conventional cpus. SPARC and MIPS CPUs get more flops, mips, and overall thoroughput per watt and per millions of transistors on a die. Maybe we will see a resurgence of eligent RISC designs as dual/quad/oct core chips become more previlent.

Re:RISC

[tomstdenis]

There is hardware for decoding but a lot of it goes into scheduling windows, register renaming, etc... the OOE support.

It's not quite fair to compare a cpu that doesn't have any of that to a recent x86 that does.

Cuz you know what, an AMD64 can hold it's own against alpha just fine. And with CISC instructions it does so with less code space pollution.

lw $0,blah
work
sw $0,blah

boring...

ADD [eax],ebx

much more efficient ;-)

I mean something like an ARM processor.. you won't see that at 2Ghz anytime soon [even though the ARM has a good ISA].

Tom

Re:RISC

[Wesley Felter]

Of course, you could also put a bunch of simple, power-efficient x86 cores (VIA/Centaur C3) on one chip. It has nothing to do with RISC vs. CISC.

Re:RISC

[Fittysix]

A modern x86 processor is basically a RISC processor internally, the core design probably has more in common with chips mentioned in the parent than they do with the 486.

http://arstechnica.com/cpu/4q99/risc-cisc/rvc-1.ht ml

Re:RISC

[be-fan]

That has nothing to do with RISC. SPARC and MIPS CPUs have few transistors because they're exceedingly simple chips. The SPARC is an in-order chip, and the MIPS are rather limited superscaler execution. RISC processors that are actually highly superscaler and out-of-order (PowerPC, Alpha), also require a lot of transistors like x86 chips do. The PPC970 is a good example. Clock-for-clock, its comparable to an Athlon 64 in performance. At the same time, its 58M transistor is comparable to an Athlon64's 68.5M transistor count. The comparison is even more even if you consider that the Athlon64 has 32KB more of L1 cache plus an on-die memory controller.

MythTV?

[rkrabath]

Would this chip have a usefull application in realtime video encoding/decoding? Would I be better off with a high clocked single core AMD proc? Anyone with experiance?

Re:MythTV?

[Kinthelt]

It depends on how your video encoding/decoding software has been programmed. A very well studied problem in parallel computation is parallelization of the fast fourier transform.

So any encoding/decoding software based on the FFT (and DCT) is easily parallelizable. Whether or not it has been coded to take advantage of any hardware parallelization is another question. And the answer to that question is a "probably not".

That said, the OS and other user programs could be running on one processor, while the encoder/decoder could be running on another processor.

Re:MythTV?

[falser]

You don't need anywhere near the performance of an AMD64 (let alone dualcore) for recording standard NTSC video. With a hardware encoder the cpu usage is negligable, and even with a software encoder my old Thunderbird 1.4GHz did just fine. And decoding is less taxing than encoding. Point being, get a hardware tuner and don't bother with an expensive CPU.

Re:MythTV?

[Mad_Rain]

Not if you're doing HDTV. Check Jarod Wilson's howto for mythtv hardware recommendations.It even says in the MythTV Hardware docs: You'll need a faster processor for HDTV.

The hardware reports that I've read say that the AMD64 has some advantages over Intel's offerings, although they don't explicitly say that it's good for a mythtv box. The fact that AMD64's often run faster and cooler than Intel's chips seems to me to be useful, especially given if you're building a single box to do capture and playback, because it means a quieter machine.

The caveat to all this is the mythtv code hasn't been optimized for 64-bit processors yet, so it's required a lot of extra effort to get running. Check the mailing lists, there are several threads about that issue.

Re:market for this?

[Vellmont]

who is going to buy computers with these new ultra powerful dual core processors?

I will. I'm often running applications that take 100% CPU. Having another core around to make the system nice and responsive would be wonderful.

gamers don't need dual core

Right, and when video cards that supported an accelerated transform and lighting (i.e. the GeForce) came out, they didn't need that either since current games didn't support it. You can bet the next core of games will be multi-threaded.

everyday users have had plenty of power for the last 5 years

That's true, but thats been the case for several years now (as you yourself said), and Intel/AMD are still selling processors. Dual cores will make an OS seem a lot more responsive though. That alone isn't enough to convince the average user to upgrade, but multiple cores is the only place to go now that the heat dissipation problem has reached a crisis point for CPU makers.

Re:market for this?

[Fittysix]

Originally there weren't going to be dual core A64s for now, though they would appear when the market demanded, only the Opteron was going to be dual core. I'm sure you can understand where dual core could be useful in a server environment. Intel decided to target the destop environment with the thier own dual core processors, mostly for multimedia applications, AMD not to be outdone decided to take the A64 to dualcore as well.

At least, that's as well as my memory tells me about all the dual core press.

Re:Redsigning your applications.

[LurkerXXX]

One of the things they should have tried in the review was to set the known single-threaded apps (games, etc) to run on a set CPU. Bouncing the thread back and forth between CPUs cost them some performance.

I usually do that on my dual-cpu systems when I want max performance out of a known single-threaded app.

No 'update' necessary

[Anonymous Coward]

Because of all the recent news regarding dual-core CPUs and the licensing implications, this question has come up a lot. Microsoft's official licensing stance remains that one die = one CPU. The company adopted this view about the same time the Pentium HTs hit the market, bringing emulated dual processing (multi-threading) to the mainstream world.

Re:No 'update' necessary

[prisoner-of-enigma]

Microsoft's official licensing stance remains that one die = one CPU.

No, Microsoft's official licensing policy is one socket = one CPU. Therefore, a dual-socket Opteron motherboard with two dual-core chips would be licensed as a dual-CPU system, even though it has four separate cores.

I think your post was trying to get that idea across, but your statement of "one die = one CPU" is misleading to that effect.

What's odd about this is if you bought a dual-core, dual-CPU Xeon system supporting HyperThreading. If you opened up Task Manager you'd find eight CPU graphs. Not that you'd get anything near the performance of a eight-way system, though...

Microsoft's licensing is a bright spot when it comes to commercial software and multi-core CPU's. There are several firms still clinging to the "one core = one CPU" model, and dual core chips are going to immediately make such software very expensive.

I contacted Oracle a couple of weeks ago to clarify their position, and I was told then that dual-core chips would be considered a single CPU for the purposes of licensing. It seems that Microsoft's adherence to the "one socket = one core" idea is forcing its competitors into the same pricing model. Who woulda thunk Microsoft would actually be helpful in this situation?

Re:No 'update' necessary

[fitten]

Yup. If you look down the Intel product timelines, they show an MCM (multiple chips in a single package) processor in there. Since Microsoft (and others) are licensing per-socket, MCMs and multi-cores in one socket are all treated as one processor.

Re:Redsigning your applications.

[TheRealMindChild]

Your application doth NOT need be multithreaded. As a matter of fact, processes are light in most Unixes and clones, so often you will find that applications take advantage of SMP via multiple processes and some sort of IPC mechanism, such as shared memory.

This is one of the sadder things about projects such as OpenSSI and Mosix. Only whole processes can be migrated to another node... not threads.

something wrong with benchmarks

[paronomasia5]

They claim a range of 42 to 75 fps for UT2004? I have a Pentium 2.4C and 5900 Ultra, and I get way over 100 frames per second at that resolution. Heck, I can probably get 100 frames a second at 1920x1200 if I tweaked everything.

I'm not sure that UT2004 is a CPU / GPU intensive product, so I'm a little surprised at their benchmarks.. link

Re:something wrong with benchmarks

[Luthair]

I would imagine they're running specific demos that stress the system more than normal.

Re:something wrong with benchmarks

[mako1138]

They make their demo publicly available on the test notes page, but the link seems to be down right now. I'd recommend you try it out yourself when it works again.

Re:market for this?

[quarkscat]

Scientific workstations?

Anyone involved in matrix math (circuit design, mechanical engineering, fluid dynamics, etcetera) would love to be able to do this on their desktop instead of shared time on an HPC. Or combine the computational power of an office full of these machines at night or over weekends for the really big jobs. What's not to like?

Any scientific organization that has been holding off on capital expeditures while waiting for a clear winner to emerge ((AMD vs. Intel) vs. (PPC vs. SPARC)) will have come that much closer to making a decision.

Intel's IA64 gambit has not panned out -- their marketing hype has brought down some of their competition (PA-RISC and MIPS), but it has not proven to be the market leader Intel would have hoped. But like a wildfire in the woods, Intel's IA64 has opened up competition for diversity and some new leadership.

more reviews up front.

[hruske]

http://anandtech.com/cpuchipsets/showdoc.aspx?i=24 10 http://www.bit-tech.net/hardware/2005/05/09/amd_a6 4x2_4800/1.html

two more reviews.

simple. It can't.

[Anonymous Coward]

The only way to accomplish this is to compare apples and oranges. Comparing this chip to their previous 130nm chip? When they've had 90nm versions of their previous chips for some time now?

Clearly, as little power as this uses, if you turned off one of the cores (permanently) it'd use less. So it uses more power (probably almost twice the power) of a single core processor.

That's not to say that it doesn't run relatively cool.

For the gentoo users out there...

[Phil John]

...make -j2 springs to mind, using two threads to compile, you'd see ~70-80% speedup in compile time.

Re:For the gentoo users out there...

[coolGuyZak]

-j3 actually. make.conf (5) reccomends setting the number of concurrent compiles between [Number of CPUs]+1 and 2*[Number of CPUs]+1

Server's slow, but we have a mirror

[Dr. Damage]

Sorry about our server's inability to keep up right now. We have a mirror here: http://www2.techreport.com/

Re:Server's slow, but we have a mirror

[arose]

Does it run on the other core? ;)

Re:Server's slow, but we have a mirror

[bach37]

Apparently it does, as it too is slashdotted. :)

Re:market for this?

[Chatterton]

> gamers don't need dual core

Right, and when video cards that supported an accelerated transform and lighting (i.e. the GeForce) came out, they didn't need that either since current games didn't support it. You can bet the next core of games will be multi-threaded.

Right and false, in the test DOOM III doesn't seems to benefit from the dual core, but Far Cary and UT seems to benefit from it.

See: Gaming performance

this is a software problem...

[Anonymous Coward]

Not a number of cores problem.

Your machine always has 1 CPU free because someone didn't write their code well enough to use both. If they did, your panorama might take only an hour to fix instead of two.

Keeping your machine useable is a separate issue from number of cores. A single core machine could be useable if the system scheduled it such that there was still some CPU available when other tasks needed it. For example, if it only dedicated 50% of the CPU to the GIMP there would always be CPU available. Of course, it'd be wasting CPU, but then again, it is in your dual core example too.

I have to say, I've used plenty of dual-core machines (Macs) and I don't see the effect you speak of anyway. You can still chew up your CPU. And besides, non-responsiveness usually comes when there is a fight over the drive head (swap fight, etc.) and add all the cores you wish, you won't fix that situation.

Re:this is a software problem...

[Big_Breaker]

Apps that are written with multiple threads are typically also written with care to avoid locking up the computer.

Single threaded apps are typically written with far less care and don't leave cycles free for the GUI and OS functions.

That is why having the second processor is nice. It has free cycles when an app is hogging the other one. A multi-threaded app will use both but will probably not hog both, leaving the GUI still "snapppy".

Re:this is a software problem...

[platypus]

Bull. The vast majority of really bad locking issues with multithreading never surface with singe CPU systems. Only when run on multiple cores these problems really come up. This is because only with several cores working on the same data your system has to rely on the correct thread programming to assure cache coherency.

Re:this is a software problem...

[ultranova]

A single core machine could be useable if the system scheduled it such that there was still some CPU available when other tasks needed it. For example, if it only dedicated 50% of the CPU to the GIMP there would always be CPU available. Of course, it'd be wasting CPU, but then again, it is in your dual core example too.

Or the system could do round-robin scheduling, where each program gets to run for a while, and then the system switches to the next program in the circle; this way GIMP gets 100% of CPU as long as it's the only program running. If another program needs the CPU, it's inserted to the circle, and they share the CPU 50/50. Insert a third program, and each gets 33% of CPU time, and so on.

A more advanced system could include the possibility of altering the length of time each program gets to run before the CPU is given to the next one. This would allow one to give programs different priorities; for example, make GIMP a low-priority program, and it will still use 100% CPU as long as it's the only program running, but when the user does something that the user interface must react to, the CPU is divided, say, 80/20 between user interface and the GIMP.

An even more advanced scheduler could divide programs to CPU-bound programs (which consume all the CPU they can get, like the GIMP) and I/O bound programs (which spend most of their time waiting for input/output request to complete), and give a priority bonus to the latter, since they are much more likely to be waiting for user input, so giving them higher priority will make the system seem faster. Naturally the program's behaviour needs to be monitored, so that the GIMP gets moved from CPU bound to I/O bound as soon as it finishes its current work and resumes waiting for user input, and back to CPU bound again when the user gives it another task and it resumes burning CPU time. This way the machine stays responsive even under load. Linux 2.6, for example, has this kind of scheduler.

Sorry if I'm stating the obvious here, but your example with "wasted CPU time" simply didn't make sense.

Anyway, the real problem with single-core machines is the "uneveness" and latency of scheduling. If you have a lot of low-priority programs/threads running in the background, latency will grow a lot, and what's worst, it will vary wildly. For example, running Freenet (with maximum niceness) in the background will cause the sound in Neverwinter Nights to play brokenly, at least on my Duron 1GHz/512MB machine. If I had a dual-core machine, and could bind all of Freenets threads to a single core, and NWN to another, they wouldn't interfere with each other anymore, I think.

Of course, a cheaper solution would be a better scheduler with a real "idle only and if that means that the damn thing will starve then so be it !" priority.

Re:this is a software problem...

[maraist]

Bull. The vast majority of really bad locking issues with multithreading never surface with singe CPU systems.

I disagree.. I've often managed to discover race conditions on a single-CPU machine.. Thanks to pre-emptive multi-tasking. The fact that there's a lower probability of context switching out doesn't preclude situations when un-safe race-condition code-blocks are called extremely often (such as memory management, etc). Push the machine to memory-swap such that every hundreth instruction will incur a context-switch and your problem exacerbates.

win 3.0 and MacOS 8 (and maybe even 9; don't remember) used cooperative multi-tasking so they didn't have this problem as badly.

Re:this is a software problem...

[maraist]

Of course, a cheaper solution would be a better scheduler with a real "idle only and if that means that the damn thing will starve then so be it !" priority.

Not to be sacraligeous, but didn't windows (and probably the MAC) have event-based threads which were only called when there was NOTHING else the OS had to do. I don't believe they were threads, so much as registered GUI-event callbacks. Such handlers were used for background disk-defragmentation, screen savers, and a few other things I've forgotten. I don't remember ever seeing anything like this in a UNIX/Linux API, which seemed like a waste.

Re:this is a software problem...

[platypus]

Yes, you are right. It's not correct to say that the bugs "never surface", they are just a lot harder to trigger. And even if you create a system state like the one you described, it's still a lot easier to expose these kinds of bugs when testing on MP.

The point I was trying to make is that if one really wants to generalize like the OP did, then assume that multithreaded programs are more fragile, not more robust, especially on SMP machines.

Re:this is a software problem...

[snookerdoodle]

Well, actually, I've tried "nice"ing the thing, and it helps a tad. But the fact that I'm not really swapping madly is somewhat of a clue that I'm not in contention for the hard drive. cpu was at 100%.

You're correct that, if enblend were designed so that it could, say, work on two borders simultaneously using two cpus, it would run faster on two cpus and still hurt the responsiveness of the system. AFAIK, it's not (and, alsa AFAIK, the Gimp can be compiled to where only parts of it take advantage of multiple cpus).

I think I'm ready for a little exra bandwidth and a simple upgrade path to a dual core is very welcome lagniappe...

Mark

Re:Redsigning your applications.

[NanoGator]

"What a lot of people dont realize (Including a lot of programmers). That a lot of applications are not multithreaded."

Well we realize it here, because it's BROUGHT UP every single time there's a mention of more than one processor running!! Yeesh. Heh.

On a lighter note: When these processors become more popular, multi-threaded apps will come. Besides, its not like our machines aren't keeping up with apps today. Except for my 3D rendering, I don't have anything that would benefit from a faster processor, and I doubt many other people do either.

Re:Redsigning your applications.

[jejones]

Well... yes, but:

1. If you do more than one thing at a time, you'll benefit--two single-threaded programs can run without getting in one another's way.

2. This might be a good time to brush up on your pure applicative language skills...don't they lend themselves to easier parallelism than imperative languages?

Re:Redsigning your applications.

[Phu5ion]

True dat! For anyone that has developed concurrent thread applications know it's a whole other animal. But concurrent programming is a worth skill to know and will only become more important over the next few years.

We will probably see some new languages being developed to make concurrence easier for the masses in the not-to-distant future.

Re:Redsigning your applications.

[Courageous]

Mod +1, Factual.

Ah well, I guess they don't have that mod point.

*wink*

C//

Re:Redsigning your applications.

[LWATCDR]

But the only time you are likely to be running only a single thread or task on a system is if you boot in DOS.
Yes you are not likely to see a HUGE increase in the performance of a single threaded app you will see some performance increase.
Are you running a Virus checkers?
What about a firewall?
If so then you will get at least some advantage from a multi-core system.

I have to wonder if the next step will be asymmetrical multi processors. I could see adding a few low speed, low heat cpus to handle things like virus scanning, firewalls, TCP/IP, and other small light weight house keeping tasks to free up the main CPUs to run the big honking nasty tasks like Doom4.

Re:Off topic, but...

[Anonymous Coward]

It's not raw energy cost, for the most part.

A large part of the greater expense results because Hawaii can not trade electric power between the islands. On the mainland, Georgia can sell power to Florida and buy it from Tennessee.

Not being able to trade makes for additional and more expensive infrastructure if reliability levels are to approach those of the mainland. The reliability if important for the tourist trade and for developing industry other than agriculture.

There are windfarms. The usual arts and letters environmentalists are all for them as long as they don't have to look at them. 'We can afford the electricity, let the poor people have the eyesore', is an attitude shared with the mainland N.E.

There is also heat pump from deep ocean, other geothermal, state tax breaks for solar (you know federal breaks are gone), and on Oahu, a pretty darn good bus system. The climates a big plus. You don't need HVAC in private homes. Sure, it gets so ht in some places that all you want to do is sleep in the afternoon, but overall it's comfortable.

Re:market for this?

[dgallina]

I dunno, the market has been here for a few years for me. I've been using dual-cores on my home machines since I switched to Mac in 2002. It definately *does* provide a performance boost, as even single-threaded apps leave a processor relatively free for the OS and/or other things. Much less bogging-down even on slower processors like my original 866mhz G4s.

Re:Redsigning your applications.

[drew]

a spell checker with all of the spare cpu cycles in the world won't help someone who doesn't care.

in this case, duel is spelled coorectly, it's just the wrong word....

I have a spell in checker,
It came with my PC.
It plainly marks four my revue
Mist takes I can not sea.
I ran this poem threw it,
I'm shore your please to no
Its letter perfect inn it's weigh,
My checker tolled me sew.

Re:market for this?

[diegocgteleline.es]

In fact, Intel's dual core offerings are just for desktop, which is why it doesn't surprises me that AMD is faster here. I mean, CPUs for servers are supposed to be "faster", so I'm not suprised that Intel cpus are being slower here, there must be a reason why Intel dual core is much cheaper than AMD's. It's being sold for desktops, it's much cheaper, why everyone is "surprised" that it's slower?

The problem is that right now they don't have a decent server product, but they have 90%+ of market share in that area so they can't lose all that suddenly. Specially when the one decent chipset for AMDs is their AMD chipset (rare to found in motherboards, at least here in spain) while Intel has much better offerings, their "platform" is stronger, even if their CPUs are slower. In the server world many people wants reliability, quality etc, not a Nvidia chipset. AMD has a great CPU but it is going to take some time to everybody to build a "reputation" and more products (altough sun and HP seem to have done that already) - I guess Intel is trying to use that time to build a decent server product

Anyway, AFAIK Intel is preparing a "second generation" dual core CPU based in the Pentium centrino platform for 2006 or so. My opinion is that what Intel is selling is just something to stand up in front of amd for now, they're somewhat late in the dual core game (opteron was designed to support dual core from scratch they say, P4 not) so they're bringing a hack to have something to show, anything else.

(most of this is just speculation of course, except the centrino-based future dual core)

Re:Redsigning your applications.

[fitten]

I have to wonder if the next step will be asymmetrical multi processors. I could see adding a few low speed, low heat cpus to handle things like virus scanning, firewalls, TCP/IP, and other small light weight house keeping tasks to free up the main CPUs to run the big honking nasty tasks like Doom4.

Like a step back to 1970 or so? Back then, it was *very* common for a machine to have a CentralPU and (many) I/O Processors (which weren't the same hardware as the CPU). For example, I could run a program on an old UniSys machine and use, like, 0.00001 second of CPU time but like 1.0 seconds of I/O time if I were doing a lot of disk/tape access. The nifty thing was that the system handled it all. I just wrote in Fortran (or C) just like any other system. The IO calls 'did the magic' to use the IO Processors without my having to deal with it.

Multiprocessor machines (along with AsymmetricMultiProcessing) have been around a *long* time. They're just now making it into the commodity desktop world.

Does anyone buy performance anymore?

[dtjohnson]

The performance of the AMD X2's is absolutely amazing but...will anyone really buy them? The big computer companies seem to be offering mostly P4's at about 3 Ghz using some elderly Intel core. The newspaper this morning carries an ad from Fry's Electronics offering a wimpy '2800+ Sempron with motherboard' for $69 and that's the only AMD thing listed in their ad. Can't be much money for AMD at that price. It just doesn't look like the desktop computer market cares much about performance anymore.

AMD might be turning out some pretty good products but they are not making any money selling them and it is only a matter of time before they have to fold their tent and leave the field to Intel.

Re:Does anyone buy performance anymore?

[ebrandsberg]

Read the article, it was the flash memory business that caused the loss, not the cpu business.

Re:Does anyone buy performance anymore?

[Anonymous Coward]

From your link:
AMD's processor business had another excellent quarter, posting record revenue and profit figures. Overall company revenue was $1.23 billion, slightly higher than the estimates of analysts polled by Thomson First Call. The processor business accounted for $750 million of that revenue in the company's first quarter, which ended on March 27.

Now does that sound like "not making any money" to you?

Re:Does anyone buy performance anymore?

[SirTalon42]

Do you expect an ad in the newspaper for an AMD 64? They aren't going to try and get you in with the most expensive one, but the one that looks good but is dirt cheap (may or may not be the best price/performance ratio though)

Re:Does anyone buy performance anymore?

[BCW2]

buy any 3.2GHZ P4 from Delland any AMD 2000+ will smoke it. I don't know what Dell does to them but they are painfully slow, and just generally run like crap.

AMD has beaten Intel in desktop sales for the last 3 quarters in a row. More and more buyers are realizing there is no reason to pay 20% - 30% more for the same or worse performance. Someday the so called majors (Dell, HP, Gateway)will catch on. Most white box sores are selling AMD hand over fist and ordering Intel when specifically asked. The one I work in never keeps Intel processors in stock. Less and less call for them.

Re:Does anyone buy performance anymore?

[BCW2]

They just came out in the last 6 months with a viable chip for laptops. The problem is that most laptop makers are also the "major" desktop makers. They are as locked into the Intel "discount" as they are to M$'s. This would be a good place for a startup with ambitions to become the replacement for IBM in laptops. Don't forget that the Centrino is a modded P III, not a P4.

Server sales are climbing in Opteron and Athlon 64 CPU's. The price hit is becoming too steep for the Intel option.

Re:Does anyone buy performance anymore?

[dtjohnson]

Sure, the flash memory business was a loser. But, according to the article, the cpu sales were only $750 million for Q1 2005. AMD's market share for Q1 2005 was 16.9 percent of a worldwide market of approximately 120 million units. That means that AMD's average selling price was only about $37. Obviously, AMD is selling mostly the ultra-cheap Semprons and only a relative tiny handful of the $500+ units. We can rattle on about how great that new powerful chip is that just came out but no one ever seems to actually buy them in the kind of volume that might allow an innovative company like AMD to survive in business.

We can go on here about how great AMD is doing and about how Intel is struggling but it is not the truth. Intel and Dell smile every time they read that as they set new revenue and profit records every quarter. Why does the truth matter? Because AMD is making some extremely good products that deserve to be selling far better than they have been. We should be asking WHY they are not selling rather than just pretending that it isn't so. AMD is in a business where huge capital investments are needed every year. AMD's access to capital will dry up if they can't show that they can make substantial profits from sales of cpus when they have VERY competitive products. The lender's will ask 'if not now, when?' and there will not be a good answer. The end result is that Intel is successfully choking off AMD's air supply even while we are here talking about how good the latest AMD chip is.

Re:market for this?

[Anarke_Incarnate]

Macs did not have dual cores in 2002. They don't have dual cores NOW. They did have dual CPU machines, but PCs have had that for a tad longer.

Re:market for this?

[coolGuyZak]

Most people who go crazy over these new technologies are either wanting it for pure bragging rights, or simply aren't aware of how little it will actually do for them... or both in all liklihood.

While bragging rights do constitute one of my goals in going dual-core, there are other motivations at work as well. (Granted, I may not be a representative of the market proper). I like to upgrade my systems once every 2-4 years, and hold out until several strides are made in computer design. The goal when building computers this way is that the amortized cost is less than that of smaller incremental upgrades.

It requires keeping a good eye on the tech industry, and understanding when certain hardware will break into your budget. You then plan around those releases. For instance: I planned my last upgrade (about 2 months ago) to be compatible with Dual-core. I went with a socket 939 and a low-end Athlon 64, expecting to get the dual core when it comes out.

Something that should be noted is that I knew when I bought it that I would have an OS that could take full advantage of the tech. I am preparing Gentoo on an Athlon64. When the X2 comes out, I'll have a fully functional, 64-bit OS. All I have to do is boot into windows, perform the bios update, and drop in the new proc.

As for which proc I'll end up buying: definitely going for the 4400+. The 1MB cache is worth the $50-60, especially considering my purchase practices.

From what I know of OC'ing, PC extremists use it as a hobby. Just like you or I would spend hours carefully tweaking linux (etc) to be the best it can be, they tweak their system. Personally, I like linux tweaking much better; It's easier on the wallet.

Re:market for this?

[coolGuyZak]

The only concern is that RIGHT NOW I don't think Windows will do The Right Things

If you mean NUMA, you're right. They didn't have it till 64-bit edition. (And they call themselves innovative? ;)

Re:Redsigning your applications.

[LWATCDR]

"Multiprocessor machines (along with AsymmetricMultiProcessing) have been around a *long* time. They're just now making it into the commodity desktop world."

Yes I know. Heck even the Commodore Vic20 and 64 used extra CPUs to handle the disks. There where a few hacks to use the 1541 drive's cpus to provide some extra processing.
Yep the PCs are going to start looking like the Mainframes and supercomputers of yore.

Power measurement

[wiredlogic]

Their power measurement method is suspect. Switching power supplies can become more efficient when they supply more current. This effect would mask any increased power consumption by the processor(s) when measuring of the AC input.

Re:Power measurement

[the eric conspiracy]

Switching power supplies can become more efficient when they supply more current.

Maybe so, but what you pay for is wall socket power.

Re:Redsigning your applications.

[MikeBabcock]

That statement is a red herring every time its brought up.

Most people who post it don't realize that your CPU is context switching dozens of times per second when idle in your OS already. Simply letting two cores handle different interrupts is a benefit for system responsiveness.

How often is your CPU wanting to do more than one thing at a time? All the time in an OS like Linux or Windows.

If you're running Linux, run vmstat and check the context switches per second.

If you install a second CPU, you may not see a 2x performance increase, but you wouldn't if you doubled your CPU speed either.

You *will* however see a much more responsive machine, because of how the system handles load better.

Re:market for this?

[pjrc]

AMD on the other hand has always started out chips on the enthusiast / enterprise

s/always/recently/

Clearly, you've just not been around very long, or not paying attention, or have only short-term memory.

It's only been in recent years that AMD has bested Intel, performance-wise. For many, many years, AMD could release a new chip with good performance similar and then Intel would beat them with another new chip.

There's a long, long history of AMD selling their chips at approximately half the price. Certainly through all of the 90's (486, pentium 1/2/3), AMD chips were substantially cheaper than buying Intel.

During much of this time, AMD's chips also had a strong reputation to run very hot. Intel had a reputation for running cool and being easy to overclock. It was Intel that introduced the multiplier locks to prevent overclocking, which apparantly became quite a problem outside the USA where unscrupulous companies would sand down the tops of the chips (back then they were usually ceramic on top) and print a faster speed and resell them as such.

It wasn't even all that long ago when the infamous celeron 300A, which was multiplier locked, could overclock to 450 MHz (then, nearly the fastest chip they sold) by overclocking the front side bus by 50%. At the time, AMD's chips were far behind, and they were running hot with very little overclocking margin, just to try closing the substantial perforance gap.

Even back in the early Pentium days, even before AMD came out with a comperable chip, the 90 MHz pentium appeared in a new, smaller geometry process that made it run about as cool as the 486 66's.

Intel has indeed been in the lead, technologically, for a very long time... ever since they stopped licensing IP from Intel. For a bit of really ancient history, long ago, some large well known companies had a strong policy of never using any components that were not available from a second source. AMD's business model 20+ years ago was to license designs and be that second source.

Even a number of articles mention how the tables have turned recently, and speculate whether Intel will regain the honor of top performance.

I'm not affiliated with Intel, and in fact the PC I'm using to write this comment runs an AMD chip. When I upgrade, it'll probably be AMD again. Recently, AMD appears to have made some really smart architectual decisions that have put them in the lead, technology-wise.

But to believe such has always been the case, or even been a trend that's anything more than recent, is to ignore or be utterly ignorant of the very long history of Intel dominating the PC / x86 market with the best chips.

good point

[drxray]

Absolutely right. But they also get 230 fps for Doom 3, so they've probably just got those graphs the wrong way round.

What would have been even more interesting...

[vnangia]

...is to have had one or two Apple Macs in the mix as well. I know that a number of tests have "equivalency" problems, but the rest would have been useful to see, if for no reason other than to compare state of the art on AMD/Intel with state of the art on PowerPC.

Usual disclaimers: not trying to start a flame war, just pointing out something else that could be useful.

Re:What would have been even more interesting...

[SirTalon42]

The problem with comparing different architectures is generally the same program doesn't exist on both, and if it does both versions may not be the same quality (one could be programmed by the best developers that arch has to offer, while the other is a 20 minute hack job). There are really too many variables to compare a specific part of an architecture.

I'm not saying its impossible to compare 2 different architectures, I'm just saying its not practical to compare 1 part of 2 architectures and expect to get results that mean anything (though Macs are getting closer and closer to the x86 architecture as time goes on).

Re:What would have been even more interesting...

[vnangia]

Agreed, but in this case the variables would have been the motherboard (which already varies between AMD and Intel), OS (can't really help it) and chip. It would've been possible to test the same video card, same HDD, same memory, and I doubt there would've been that much difference from power supply, optical drive, things like that. But nonetheless, your point is well taken.

Re:What would have been even more interesting...

[mako1138]

I believe it's also been shown that there are real differences in applications between platforms, especially the standard, Photoshop.

But I agree it would be interesting.

Re:Sounds nice, BUT....

[johnw]

My mind is boggled by questions like this. Are there really people out there who still use their computers for just one thing at a time?

The machine I'm typing this on (just a simple diskless workstation) currently has 75 different processes running. The server it's connected to has 145. With a dual core processor in either of them, the number of processes able to run simultaneously would be increased by 100%.

The idea of running just one application on your box went out more than 10 years ago. Wake up and smell the coffee.

(If nothing else, all those blasted Flash animations can run without chewing up CPU cycles I would rather use for something else.)

John

Re:Redsigning your applications.

[platypus]

And

""computers are state machines. Threads are for people who can't program state machines""
Alan Cox

Not enough comparisons ...

[Alain Williams]

1. against a whopping 17 competitors

Yes: but they are all Intel compatible ... what about some interesting comparisons like: IBM's power PC, Sun's SPARC, IBM mainframe, ...

Re:Not enough comparisons ...

[Fizzl]

what about some interesting comparisons like: IBM's power PC, Sun's SPARC, IBM mainframe,

Yes, today on Slashdot: Double blind taste test; apples to oranges. Which one tastes like banana?

Re:Not enough comparisons ...

[BFaucet]

It's difficult to make a fair comparision because programs are usually designed/optimized for one architecture.

Re:Not enough comparisons ...

[hyperm0g]

I agree, and while we're at it, can we get some apples in there, and would a few oranges kill them either?

Re:Not enough comparisons ...

[Ninja Programmer]

Yes: but they are all Intel compatible ... what about some interesting comparisons like: IBM's power PC, Sun's SPARC, IBM mainframe, ...

Because those processors will score 0 on 3DMark, SiSoft Sandra, and just about any other video games used in the comparison. Any more stupid questions?

Re:Not enough comparisons ...

[What me a Coward]

Why does everybody gotta keep going on about these Apples and Oranges? Whats wrong with a few pears, peaches and maybe a few Bananas?

Does it always have to be Apples and Oranges? :D :P ;) :D

Re:Redsigning your applications.

[Jace of Fuse!]

Keep in mind that while software does need to keep duel-core/processor design in mind to gain the best advantages, a lot of advantages can also be gained by using APIs whose newest versions themselves support such configurations.

By redesigning core APIs of an OS to support dual-core or dual processors then any software using that API may see an improvement when applicable.

As an example, I imagine it would be possible for future versions of DirectX to be heavily multithreaded and thus many games for Windows that use DirectX might gain some of those advantages.

Conclusion, Intel has problems

[trintron]

I think we can safely say that Intel has problems. AMD is currently beating them all the way. Including former P4-friendly tests. HyperThreading doesn't help anymore.

Re:Redsigning your applications.

[platypus]

What a lot of people don't realize is that threading per se doesn't at all guarantee faster performance on SMP systems.

When your threads running on different CPUs are working on a shared dataset that causes numerous issues concerning locking, cache alignment, cache coherency etc., you can get severe perfomance drops (in fact, you can get negative scalability). In fact, if you read in comp.programming.threads, you'll find many posts about people finding out that it's much easier to write good performing multi-processed applications with communicating over ipc that multi threaded applications.

Re:Redsigning your applications.

[masklinn]

Best spell checkers are supposed to include a grammar checker as well.
It's one of the few MSWord features that still aren't mirrored in OOo, and even though grammar checkers are much harder to create than spell checkers, I found Word 2003's to be quite nice and work quite well (the green wavy thing is butt ugly though)(and I don't mean "nice lil 19yo chick's butt ugly" but "55yo 250 pounds greasy hairy male's butt ugly")

ExtremeTech too

[MStiles]

ExtremeTech has an excellent story as well. They call it the best desktop processor ever.

Re: new fab process

[katorga]

AMD's new 90nm fab process with SOI and other technologies licensed from IBM allow them to drasticly reduce the power consumption of a single AMD64 core (venice core) to roughly half that of previous 130nm fabs. That allows them to fit two cores into the power envelope of the preview 130nm single cores.

wth

[oskard]

Does a bear pope in the woods?

Re:wth

[What me a Coward]

When i read that in the review the only thing that came to mind other than uhm yeah ok lets not spellcheck was I don't know "does a pope bear in the woods" :P

Re:Redsigning your applications.

[Anonymous Coward]

I must say something about SMP "smoothness".

I have a dual 1-GigaHertz Pentium III that I've used for 5 years. Just recently I bought a single AMD Athlon 64 3500+ (939). Guess which system is more responsive?

The AMD 64 system is faster in a single application, like a game. But doing ANYTHING ELSE is much nicer on the SMP machine. (I.e., regular system use... browsing, e-mail checking, etc. etc.)

Building the WORLD in FreeBSD took 4 times longer on the dual machine... but YOU COULD DO SOMETHING ELSE AT THE SAME TIME. The single CPU machine is so much faster... but the system is flakey if you try to do anything else while doing that build.

I'll be getting a dual core as soon as I can.

I submit that one reason my SMP machine lasted so long before I needed to get a faster CPU was because it was SMP... if it was a single CPU machine... I'd probably have upgraded several years before out of frustration.

(Yes... the nice fast single CPU machine is often frustrating... after living with the SMP smoothness.)

--Keith Morris

Re:Redsigning your applications.

[shutdown -p now]

Think e.g. ripping music/video in background.

Yes

[ArbitraryConstant]

Intel has vast resources. Even with their penchant for backing the wrong technological horse time after time (after time), they're not simply going to go away because they will have a lot of money and they still make a lot of money.

The whole NetBurst architechture has reached the end of its life. You can see this because Intel has stopped selling the chips by their clock speed. Their alternative, the Pentium M, is a match for the Athlon64 in many ways, and if they re-engineer it as a dekstop chip without quite as many power constraints it's quite possible they can pull ahead again. It will be a rough year or two for them, but they're not going anywhere.

Re:Redsigning your applications.

[CyBlue]

This will be my first negative post ever on slashdot. I don't really need to say anything because I think everyone else has covered it adequately. Since when does your typical geek only run one application at once? If you're running anything (even if its all single-thread apps) like Folding@home, SETI, ripping DVDs, TV Tuner card, etc then DUAL-core will definitely beneifit you.

Re:market for this?

[John Courtland]

I actually still have a functional AMD 386 clone. Was 40MHz, about half-ish the price fo the Intel 386DX and was faster.

Re:Redsigning your applications.

[w42w42]

If a peripheral is hurt, then it's only a flash wound. If the processor is hurt at its core, that'd be fatal, wouldn't it?

Re:market for this?

[hackstraw]

Just my two cents...

That is all people care about. People will switch to dual-core chips when the price is right. Abnormal people (scientists, gamers, etc) will switch when their apps will gain significant performance from them.

Is this new to people? I mean, cars like BMWs, Mercedes, Ferraris, etc are typically better than Toyotas, Fords, and Chevys, but only car freaks and the wealthy drive the more expensive ones, whereas the rest of us buy the cheaper ones.

Re:Supported by Linux - Yes/No?

[w42w42]

I'm assuming it's seen as just a std dual core setup. As to benchmarks though, I'd love to see some Linux ones if anyone has a set, some db and web server stuff would be excellent.

Who cares about 1 application?

[TheLittleJetson]

You're quite right. Performance in single-threaded apps will suck. Does this matter? Your average system will have a bunch of threads running, why not split them between the 2 processors (or cores).

We have multi-tasking OS's for quite some time now. Multiple processors isn't just about running one application faster, it's about running all your applications faster collectively.

Re:market for this?

[pjbass]

Their (Intel) whole business plan has been based off of their vast number of production plants and relatively cheap process of putting hordes of chips on the market

....

AMD on the other hand has always started out chips on the enthusiast / enterprise market because they simply don't have the fabrication capacity that Intel does.

Do you even know what Intel's first dual-core processor is? It's the Smithfield product, or, in market terms, the Pentium Processor Extreme Edition 840. I'd be willing to argue this isn't targeted at the general desktop world as much as it's targeted at the enthusiast market. The next chip that is releasing (Presler I believe) is aimed at the Xeon, again, not exactly an end-user part.

I think what is more accurate is once Intel has matured a product (to an extent), they use their manufacturing juggernaut to flood the market with their parts. Whereas, AMD cannot, since they don't have the fab capacity, so once their part does become mature (to an extent), they're still producing a seemingly low volume. Therefore, they must target higher ASP's (servers, workstations) in order to make money.

Here's info on the Intel release.

Re:market for this?

[AgNO3]

I don't know about windows but on the Mac single thread games do get a boost from the second cpu because all the open al calls go to the second processor. Maybe windows does something like this?? I don't know but it would mean that games would get a boost from the second cpu already.

You are completely right!

[DimGeo]

I think the grandparent didn't mean to say that all multitasking code was bugfree, but you are completely right that most synchronization problems surface only on dual-core machines. Amen to that! That's one of the main reasons I can't wait to get my hands on one of those new CPUs - to test my own code! By the way, I think some Redmond-centric OS-es had better implement a feature allowing certain programs to run on one CPU only, thus making some bug workarounds possible, until the deadlocking / messed up programs are fixed.

I think I can buy some reasonably priced old dual-cpu servers at 1000 mhz or something, but nah, I need the gaming performance as well - you know, people don't work 24/7, we also need recreation and resting - and the best resting is, indeed, not sleep, but computer games! :)

Re:You are completely right!

[Big_Breaker]

That is exactly what I meant. Synchronization is a tricky issue but I'd consider it far more rare than the problem of cycle hogging single-threaded apps.

I'm not professional programmer and my hobby style projects have all been single threaded. That said, if my multi-threaded app performed poorly on a dual processor system I'd consider the good single processor performance to be a fluke more than anything. Threads cease to be threads when you need to worry about their low level interactions. The abstraction and communication protocol should be all that is necessary for hardware, OS and app to work nicely with threads.

In most programming environments I'd expect the operating system to handle the allocation of threads to processors and for the secondary threads to have an asynchronos relationship with the master thread anyway. I'm sure the details can get tricky - they usually do.

Re:You are completely right!

[maraist]

That is exactly what I meant. Synchronization is a tricky issue but I'd consider it far more rare than the problem of cycle hogging single-threaded apps.

Bugs might be more rare, but "hacks" to allow multi-threading by having universal locks are a serious issue. I know that perl 5.5 has a few global locks which REALLY impeeded multi-threading. This is of course because perl wasn't originally designed for MT and people were experimenting with different possible migration paths. Java was pretty good at allowing for MT, but early garbage collection had the same general universal-lock effect. I've been impressed with various MT memory management schemes I've seen in glibc and by SUN (magazine cells), but that's somewhat of a digression.

The thing is that it's easy to write a piece of code to be MT-safe; just never use globals, and when you do ;synchronize the hell out of it. THe most common aspect of java these days (EJB's or servlets) are all inherently MT, and the design paradigms generally make it easy to work with; you hardly ever have to think about MT (unless you're being evil and creating static variables (e.g. globals)).

But the hard thing is writing some optimized for speed, because you're VERY prone to take short-cuts.. You have to have the entire variable-space mapped out in your head so you know all the special conditions, side-effects, and use-cases. You're likely to cut any corners you don't expressly see a need to keep "elegant". Thankfully such optimized code tends to be restricted to a very small piece of functionality, lest the whole project be doomed to managerial anarchy. But such pieces of code are serious issues for MT, especially since their goal is high performance. Pray you need never be tasked to do this. If you can divide and conquer by threads, then you can likely divide and conquer by independent pieces of hardware and thus literally throw money at the problem whilst using a more elegant programming paradigm.

Re:You are completely right!

[Big_Breaker]

You bring up a good point: Programming languages and their interpretors, where applicable, need to be MT safe and they often aren't.

A large matrix calculation is the perfect place for two processors to work on a problem, but if you ned to copy memory back and forth it loses a lot of efficiency. If you pass a pointer to save time you have created a sync problem.

list of Athlon 64 X2 dual core reviews

[ruiner5000]

Just one is never enough. Spread the love people. I've overclocked it to 2.7GHz by the way.

AMDZone.com Tech Report Sudhian Hexus Hot Hardware Anandtech xbit xbit PCWorld Trusted Reviews

Re:Sounds nice, BUT....

[ArbitraryConstant]

If you do anything that is capable of using more than one flow-of-control, you stand to benefit.

For developers, or (shudder) Gentoo users, compile times will go way down. Graphics people will benefit because those applications are already optimized for dual-CPU or hyperthreading systems most of the time.

It's a chip targetted at the "prosumer". The people that stand to benefit already know they will benefit, and don't need to be sold on the concept. If you don't stand to benefit, you'll be able to save a lot of money getting the single-core version.

Re:Redsigning your applications.

[J.R. Random]

Duel core processors are for gamers -- the two cores fight each other, so the gamer can run Excel and get his work done.

My spelling and grammer combined with the fact that I have college degree, proves a problem with the educaion system

You're right about that.

Re:Redsigning your applications.

[Aldric]

You're talking about desktops. Most big server apps are multithreaded.

Re:Redsigning your applications.

[Cromac]

Dvorak, he'll also sympathize about it using 100% of the CPU as well.

Wrong

[GunFodder]

I think home computer use depends more on what the machine is capable of than anything else. All current major OSes are already multithreaded and many home users already have systems that are powerful enough for multitasking.

Home users now work with lots of media files that require encoding, decoding, filters, and other compute-intensive tasks. These are mostly "trivially parallelizable" computations; they perform repetitive operations on lots of discrete data. I believe many game-related operations fall into the same category. We will see more optimized software now that dual-core chips are shipping.

Home users run multiple apps simultaneously as well. They expect to use their mail, browser and office productivity tools at the same time. And they have a host of background processes running at all times; virus scanners, wizards, trojans, etc.

Java

[sean.geek.nz]

For simple (ie largely single-threaded) apps the dual core chips will reduce further the speed difference between app and C and C++ and apps written in Java (or other high-level languages).

Because automated garbage collection is the big performance difference: and for modern runtime environments automated garbage collection runs in a parallel thread.

bottom tier white boxers actually

[DABANSHEE]

As far as I've noticed, no one uses AMD except local operator owned PC outlets & people like me who build boxes for myself & others that twist my arm.

All the brandname PCs still virtually always have Intel inside, mores the pity.

Re:Redsigning your applications.

[Destoo]

Rolls for save.. Critical success.

Would that be a fatal exception?

Vacuum Colling Reaction explained this post

[newpath4comVersion2]

Easy explanation! The signal speed is creating a cooling vacuum coming in behind. Sort of like the way the Flux Capacitor works. No Thanks Necessary. The signal transmission speed is then being forced into an even faster speed by the buildup of the vacuum. Sort of like how a negative + a negative equals a Positive...

Re:market for this?

[Money for Nothin']

Try running a Java or C# app sometime. You'll see the what's so great about SMP then! :)

(please, no flames, I don't care about your language holy war... (besides, I like Java as a language. It's the slow-running, memory-hogging VM I hate.))

Re:Redsigning your applications.

[jellomizer]

Un excuse me Redundant Moderator This was one of the first posts. And this topic was once on top of the list for a while. Perhaps you should focus on moderating positivly. then moderating stupidly.