Slashdot Mirror
Opteron Benchmarked Against Xeon
jbmnuke writes "Tom's Hardware has posted a review of AMD's Opteron v. Intels Xeon." Nothing gets the blood pumping like a whole new generation of CPUs to compare numbers to, right? Update: 04/22 12:35 GMT by H : And there's the official benchmarks as well, with more coming - like Linux Magazine and Newsforge
It doesn't get my blood pumping, I can't afford such things (or cool them)! *Pats Duron 1000*
Nothing gets the adrenaline pumping like the flood of trolls this sort of comparison should inspire.
but it reminds me of a benchmark performed between desktop x86's and a sun machine. Given the different architectures, it really didn't make sense. However, the benchmark was supposed to show price::performance. Is this what Pabst is trying to convey? I don't take much stock in benchmarks anyway, as I would rather get my hands on it and try to break it.
So they benchmarked Opteron's 64-bit operations against Xeon's 64-bit operations? ;-)
Next Review: Apples vs Oranges, Which has more of an Orange taste?
Looking for hardware (Currently need: Large Etch-a-Sketch) Have one? See my journal!
So you do not believe that Intel got where they got today becuase of competition and pressure? You sincerely believe that Intel wouldn't sit back on their lazy ass and inflate prices, if there were no copmetition?
Naivity ensues obviosuly.
Not Buzzword 2.0 compliant. Please speak english.
Wooo Haaa!
;-)?
The era of commodity 64 bits is here, hopefully to stay!
I feel like celebrating, probably as much as for Mozilla 1.0. (What, I need to get a life for what
Save your wrists today - switch to Dvorak
Suddenly, I feel this is old news... It came out five hours ago for gossake!
_ ____Very good____Good
Nonetheless here is the condensed version:
_____________Server_______Workstation
Opteron_
Xeon_________Good_________Very Good
Didn't I see this in an Evangelion episode?
-Tolerate my intolerance
I was right, I got this link off another website (amdforums.com), when I noticed it was slowing to a crawl I immediately thought "Slashdot". What do you know! first story :p
History will be kind to me, for I intend to write it - Sir Winston Churchill
"Nothing gets the blood pumping like a whole new generation of CPUs to compare numbers to, right?"
I find sex better, whatever flicks your switch I guess...
Check the Spec benchmarks here.
SpecFP_rate, 2CPUs:
Itanium2 1GHz: 30.7
Opteron 1.8GHz: 26.7
SpecFP_rate, 4CPUs:
Itanium2 1GHz: 49.3
Opteron 1.8GHz: 49.2
Here we see the beauty of AMDs integrated memory contoller. Despite that 1GHz Itanium2 is a $4000 chip and has 3MB of cache, doubling the number of CPUs increase performance only by 60% because Itanium2 uses shared bus.
Opteron gets impressive 84% improvement because
memory bandwidth increases as more CPUs are added.
In SpecInt Opteron is much more faster than more expensive Itanium2.
The dual Xeon has 512 MB RAM.
The dual Opteron has 2 GB RAM.
Pretty sloppy, if you ask me.
We do not live in the 21st century. We live in the 20 second century.
Nothing gets the blood pumping like a whole new generation of CPUs to compare numbers to, right?
Seeing a naked girl is really going to blow your mind.
Tom seems to be blocking referrals from slashdot, so copy and paste this to make it view the article: http://www.tomshardware.com/cpu/20030422/index.htm l
when are we going to see something featuring currently manufactured product?
...vividly encapsulates that post-Watergate/pre-punk/coked-up moment when you could trust no one, least of all yourself.
Comment removed based on user account deletion
Well, these benchmarks are supposedly suggesting that Opterons excell at server-type operations, while workstation performance is lacking. However, if you check their benchmark setups, there seems to be another way of looking at this: isn't is so that Opterons simply run better on Linux rather than Windows?
"Man in the Moon and other weird things" - wfmh.org.pl/thorgal/Moon/
or is that Opteron, one HUGE processor?
Pictures on computers can not possibly be to scale as we have different screen sizes and resolution, for instance, if you are looking at it on a project then it probably is you
Conclusion: you
If you read a speed reading book, does it take you less time to read the second half?
not anymore. my advertising contract with them ran out. :(
The World's Worst Webcomic!
Slashdotted after 20 comments, that was quicker than usual.
Considering how P4 3.06Ghz actually runs at 3Ghz and does much better at .....
/. readers seem to as well. Naive of me perhaps to think that /. readers would be more clued up, but hey.
Tell me, why is the fact it actually runs at 3Ghz important?
MHz is not a useful measure of performance.
Jesus. No wonder AMD implemented their 'marketing MHz' rating system - the average guy on the street thinks that's how you measure perfomance of CPUs, and even some
I'd love to see the MHz rating be completely scrapped from how we rate CPUs in stores. Yes, it's useful to see that an AMD 2000+ is faster than a 1800+, but it's not so great when comparing with Intel chips. The trouble is that since AMDs are better at some things, and Intels better at others, a number of figures would have to be provided to make a fair and useful comparison. Too many numbers though I'm sure might confuse people, so I guess we'll be stuck with the MHz wars for a while yet.
However, there's little doubt that they are meant to be compared to pentiums, and you raise an interesting point. Even stranger would be - what happened if intel adopted the same scheme? Then they'd both basically be making up numbers!
No, I don't want a free iPod
I know AMD is planning a "home" version this September, but are there any boardmakers planning regular boards for this chip?
Opteron is a server/workstation CPU, thus 244 marking make sense. You usually do some research before purchasing $2000 workstation :)
The German version of the review seems to be quite a lot faster now than the English one: http://www.de.tomshardware.com/cpu/20030422/index. html
while true; do eject; eject -t; done
The thing that got me was the roughened up brushed metal look. Reminded me of the hull of some near futuristic space ship from something like Aliens.
Or perhaps you are saying they should use Pentium III processors [that] provide a better "Internet Experience"
I'm sorry if I haven't offended anyone
Sehr gut, aber wieviele slashdot Benutzer sprechen Deutsch?
</probably bad german>
Why does P4 with 2x64bit memory-bus get so much better results than opteron with 2x64bit memory-bus? One would think that since the mem-controller is integrated on the Opteron, it would get better results. Also, since each CPU has it's own memory-bank but they can still use other CPU's memory as well, the bandwidth should go up as number of CPU's increase. But still, P4 has more bandwidth than 2x Opterons! How can that be? IS there something wrong with the chip Tom benchmarked?
Lesbian Nazi Hookers Abducted by UFOs and Forced Into Weight Loss Programs - -all next week on Town Talk.
Apples.
I am TheRaven on Soylent News
Wow, I think this review actually took down Toms. I can't even get to the site. "Contacting www.tomshardware.com..."
well then, I'm sold.
at first I was thinking I wasn't sure if I needed the 64bit space... but if it is near futuristic looking, then I guess I have to get it. performance is secondary to looks for me when it comes to my CPU.
There are some odd things afoot now, in the Villa Straylight.
Let the images do the talking. ;)
And there is always Babelfish that can help too.
while true; do eject; eject -t; done
"The average person business user..."
If you had a brain, you'd look outside your narrow little view of the world and realize that the Opteron is NOT designed for the "average person"!!!
The average person will never need a semi-truck. Should we stop making semi-trucks?!
The average person will never need a satellite. Should we stop making satellites?
The average person will never need a submarine. Should we stop making submarines?!
Please do us a favor and THINK before you post!
P.S. One more thing, what the heck is a "fasical waste of resource"?!
If someone says he and his monkey have nothing to hide, they almost certainly do.
How about these?
Stick Men
Why don't they post comparative benchmark results with a 16mhz 386 (with a microscope to see them), so we can see how much we moved ahead in the last 10 years?
they're just trying to save bandwidth by letting users download the movie first.. this way, less users will download it twice (either by visiting the page twice, or by clicking refresh)
So Centrino running at 1.6 Ghz but outperforming the 2.8 Ghz Pentium 4 is invalid?
How about Itanium at 1.2 Ghz outperforming the Pentium 4 at 3.06?
Or how about the 3.0 Ghz Pentium 4 beating the 3.06 Pentium 4 in every benchmark?
Yeah, you are right, Centrino, Itanium and the 3.0 Ghz Pentium 4 are all P.O.S. They are all officially dead.
If voting were effective, it would be illegal by now.
I think he mistyped "farcical".
I don't know what a semi-truck is, but all of your other counterpoints are meaningless. He's right, all the things that are wrong with our computers NOW will not be ameliorated by faster CPUs, more RAM and more bloated software. It often occurs to me that my Palm Tungsten T is QUITE A BIT more powerful than my Commodore 64, my Mac Classic, my old 386DX, possibly even than my Amiga 500 and my Mac Colour Classic, yet that extra power doesn't seem to make it much quicker or easier to do stuff with. My PowerMac G4 dual 500 doesn't seem to run my Excel spreadsheets any faster than my old Quadra 800 either - despite it being probably 100 x faster and despite it having 1024MB of RAM where my Quadra had 80MB.
Mac OSX's "Genie" mnimising effect is probably the best example of pointless CPU utilisation I've ever seen - with the possible exception of MS Office.
That was classic intercourse!
Keep in mind this opteron only uses 40 Watts.
love is just extroverted narcissism
Once again, you're thinking about YOUR needs and not about the needs of others. Is your brain SO small that you can't even imagine a situation outside your own little world?!
You use your computer for runing Excel spreadsheats. That's fine. But the Opteron is designed to run servers. If you don't see a need for the Opteron, DON'T BUY IT! Those who do need it certainly will.
If someone says he and his monkey have nothing to hide, they almost certainly do.
the point wasn't about the Opteron per se, but about the almost irrelevant gains in productivity and utility we see from our ever-stronger CPUs. And you know as well as I that AMD isn't really interested in selling 1000 Opterons when they could be selling 100000 Athlon 64s. Opteron is just a slow start for them.
Is the introduction of the Opteron going to help out my ever growing inbox spam problem? Didn't think so.
That was classic intercourse!
Opterons are not for desktops/workstations but they are very powerful despite quite low frequency they are running (~1.6GHz). Desktop versions of Hammer chips will have smaller cache thus it will be much easier to manufacture chips around 2.5-3GHz. How will Xeon 3GHz compare then? Wait now judge later.
My words exactly!
I thought guys like you and me were all but extinct?
-H
Here
The review is very good and contains lot of real-world Linux benchmarks.
Notice that the DB benchmarks were done under Linux. Please point me in the direction of the nearest Microsoft SQL Server XP2003.Net Home Edition that is available for Linux.
War is one of the most horrible things a human can be exposed to. And one of the worlds largest industries.
eh?
if that IS the case then it says more about their website than anything else - if they wanted to force a download, they didn't have to go to the lengths of ZIPing.
That was classic intercourse!
Ace's hardware as an in-depth review as well, and it isn't slashdotted.
http://www.aceshardware.com/read.jsp?id=55000251
Probably because the Opteron isn't supposed to compete with the Athlon MP. It would be like AMD comparing the Athlon with the K6. The Opteron IS supposed to compete with the Itanic II, and the Xeon.
I want my rights back. I was actually using them when our government stole them after 9/11.
- A 486DX 50mhz system with 16mb of ram is well enough for these purposes
You haven't used MSIE or Mozilla lately, have you?Joe Sixpack believes MHz = speed.
;)
Ï really hope those in charge of purchasing servers know better, or they'd be replacing their Xeon/Itanium with a PIV anyway.
Now, the desktop is another story, but I guess we'll have to wait till September for that. Don't be surprised if the marketing dep. make it indirectly seem as if 64bit = 2*32bit, so it must be twice as good though
Kjella
Live today, because you never know what tomorrow brings
If you had a brain, you'd look outside your narrow little view of the world and realize that the Opteron is NOT designed for the "average person"!!!
Right, and the 386 is designed only for use in servers (as were the 486 and Pentium). The posters general point is correct, in that all CPUs in the current generation and from this point forward are hugely inefficient in terms of power usage, heat dissipation, and packaging. On top of that, they're not designed to do anything particularly well. They're huge, do-everything processors. This is why a graphics processor running at 1/10 the clock speed of a CPU can outprocess a high-end P4 by an order of magnitude.
If you think that the Opteron is a weird exception, and that upcoming processors are going to be much more energy efficient, then you're wrong. The problem is only getting worse. Transmeta's Astro provides a bit of hope at least.
Well the fact that it's totally untrue might have something to do with it.
Dissolve... Resolve... Evolve...
Finally been able to read the whole article, damn 404's, I think the opteron is very reasonable as a workstation its not bleeding edge compared to the dual xeon rig, but it all comes down to the price, if amd can sell this chip slightly cheaper than the xeon then its definately gonna sell extremely well, my only worry is the yield per wafer, this is really gonna have a huge affect on the price i dont know if they can afford to price it cheaper than the xeon, im confused at where this is being marketed, is it a direct competitor to the itanium(2) or xeon?
I'm just a little dissapointed with the whole ddr2 situation, i find it interesting that some mobo maufacurers have already worked out how to disable the on-chip memory controller, will using a northbridge memory controller have even larger latencies as a side affect of that? i suppose its just gonna depend on how long it takes amd to react and change the controller.
I think the smartest thing they can do with this chip is upgrade the controller to ddr2 and move to a 9ìm production processes, but is this gonna happen anyway because of the fab venture with ibm?
If the memory controller is built into the processor, does that not mean that in order to update to a better memory technology, that you would have to redesign the processor, or at the very least buy a new processor? I can see the performance enhancements of the on chip controller, and dedicated memory, but will this not make for expensive upgrades?
CPU time =instruction count x clock cycles per instuction x clock cycle time
SPEC's have been proven not to be able to measure the speed of "just the CPU" as they say they do.
Sorry, but there are lies, damn lies, and benchmarks. Measures of individual parts of a system might be good for determining bottlenecks, but they aren't good for measuring real-world performance.
(-1, Raw and Uncut is the only way to read)
A counterargument then:
Assuming the price isn't inflated by either AMD or Intel, the Opteron uses less materials = less waste (wastage isn't a word anymore than smartness is a word).
That's debatable. The prices are probably inflated.
However, given the performance of the chips, it will require less Opterons to do the same job as the Xeons which means less "dumpage", less waste, and less contamination.
Better performance = better environment, in comparison to the competition.
The message on the other side of this sig is false.
This is entirely correct. The distribution of SuSe linux and the applications used were designed to use the x86-64 bit mode, and get the full benefit of the new instruction set and large registers. If he was able to try some windows benchmarks designed for x86-64, he might have gotten more favorable numbers. Alas, betas of these microsoft products are nearly impossible to obtain, although they have been promised.
Where the opteron will shine is when people start doing things like testing OpenSSL using the new instruction set in the encryption core. I notice in the latest tarballs, there is x86-64 optimized bignum code, which is used elsewhere to implement various parts of the crypto library. I am eager to see the numbers on that.
Fuck Beta. Fuck Dice
Jarvik brand coolant pumps, Hellfire thermal paste, copper tubing with simulated brimstone anodized finish. And as for the cosmetic aspects of the case-modding, the thematic possibilities are endless. Start with this: Horns!!!
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
Classic bad/intentionaslly misleading graphics on the AMD site. The graphs typically cut off 80% of the lefthand side of the chart values, making a 10% difference look like a 100% difference to the causual user. If you design a graph like this you can arbitraily make the difference look as large as you like -- completely meaningless.
It doesn't get my blood pumping, I can't afford such things...
Actually, you can afford it; I submitted this a couple days ago, but I guess it didn't make the cut:
Probably both AMD and Intel will compare future chips to some cheap P4 and say that their new chip has 6.4 GHz of equivalent performance.
Like in the days of yore, when new computer performance was measured in terms of the DEC VAX 11/780.
Or, in the mid 1990s, SPECfp95 was close to 1.0 for a Sun SPARCStation 10.
I'd be curious what the new chips do in terms of the old benchmarks. The numbers would probably be outrageously high.
I'm glad that AMD is bringing out the Opteron. Competition in the CPU market is good news for consumers and it's been a lot of sad news to see some promising high performance chips like the Alpha get canned.
"Provided by the management for your protection."
The "review" at Toms is crap. Review at Aceshardware is alot better. Opteron beats Xeon in several benchmarks, and in some cases, by a huge margin.
Yes, by the time Athlon64 is introduced, Intel will have faster chips. But AMD-chips will be faster as well.
Lesbian Nazi Hookers Abducted by UFOs and Forced Into Weight Loss Programs - -all next week on Town Talk.
Guess you haven't used a 486DX 50 recently. I have. IE 3.0 is the only browser that was there; it bombed on about 1/3 of the pages I tried to view. But, it did play solitare fine.
Eat Lamb, 1 million coyotes can't be wrong
Has anyone else noticed that on AMD's benchmarks all the configurations are different on all of the machines. Like some are running Windows XP Pro, RedHat 7.3, SuSE Linux Enterprise Server 8, Windows Server 2003 Enterprise Edition. Makes it kind of hard to draw a clear picture. Conspiracy? :P
... more like 2 MILLION TB (2 Exabytes) of addressable RAM, in theory...
$ dc
2 64 ^ # 2**64 bits
1024 / # kilobits
1024 / # megabits
1024 / # gigabits
1024 / # terabits
8 / # bytes
p
2097152
q
$
Its much better at finding server-centric applications to benchmark:
Ace's Hardware Review
Why didn't they benchmark an Operton vs. Itanium?
Whats the point of bencharmking it against a Xeon which is still just a 32-bit CPU?
Mac OS X and Windows XP working side by side to fight back the night.
please mod parent down.
he is grossly wrong.
amd speed ratings have nothing to do with intel.
not food for thought considering you don't know what you are talking about.
I wish he did do some more appropriate tests, though. By far, the one app that I spend the most time waiting for on my desktop, is gcc and I would love to have seen comparative timings for that. Time gcc building itself or the kernel or a whole Gentoo system (I both love and hate Gentoo ;-) or something, make -j 3.
It would also be interesting to see comparative timings for mencoder, though it might be hard to justify that as a benchmark when you have a multiprocessor system.
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
What the fuck are you talking about ? It takes no more cycles than it did to do 32-bit pointer math, because that's part of what being a 64 bit CPU means.
The Opteron will be available in a desktop version. The Xeon is not, and will not be.
That means there is a HUGEEEEEEE difference in price.
Also, every review except for Tom's shows the Opteron beating Xeons in more workstation tests --- Tom didn't enable the second memory channel or use DDR400. Both of which limited performance.
How long do you think it would take the marketing departments to resort to scientific notation?
One of the reasons that I became a lawyer was to avoid ever having to hire one. -SPYvSPY
1994 - Apple launches a new line of personal computers, called the Power Macintosh, which uses a 64 bit RISC microprocessor developed in alliance with IBM and Motorola.
How small a thought it takes to fill a whole life
I can just see it. Corporation 1: The new Processor 4e7+. Corporation 2: The new Processor 4f7+! It must be better!
I've had this sig for three days.
Programmers can be much more productive today because they don't have to waist as much time getting simple things done and if it is at the cost of some speed and bloat, I am just fine with that. If a program takes a month to slap together using (by your definition) inefficient tools but doing it the "right way" (by your definition) would take nine months- which way is better? In the past when machines were slow and short on resources, the extra eight months might have been the right solution. Thankfully today, in most cases, the one month solution is the better option and as a programmer (and user), I am happy with that. Faster computers allows for programmers and users be more prductive.
I miss the Karma Whores.
Read the review ..
;)
AMD did not implement the full 64-bit virtual address (neither does Itanium2). The Opteron has *only* 48-bit virtual address and 40-bit physical address. That means it can address upto 256TB of virtual space and 1TB physical space.
And yeah, 256TB ought be enough for everyone
Ok, so the Opteron can nativly run 32-bit apps faster in Itanium can emulate them. Speaking onf not reading, did you notice that Opteron is not really compition for the Itanium 2, it's aimed at where Xeon is.......
Why am I replying to an AC that clearly didn't read the article?
Semi's, Satellites, Submarines, all are irrelevant in this context. Opterons will be aimed at workstations and those of us old enough can probably remember times when 386's, 486's and Pentiums were not for the average user. Well, time changes technological expectations. Once the price drops sufficiently it becomes the norm for the average user. Though AMD has a consumer model on the roadmap for September (which, quite honestly must be very worrying to Intel), you might expect in a couple years (which AMD is looking at) the current crop will (a) be out of date (b) their successors will probably be on the shelf at Office Depot.
Personally, I was quite stunned to see the Opteron 240 priced at $283. I paid $349, back in December, for an XP 2600/333, so that squarely plants it in my ballpark of affordability. Mobo availibility and memory prices may have more to say on this, but if schmucks like me slap one together to do A/V or high end games on, today, you can pretty much bet I won't be so cutting edge in 6 months. Particularly where the hyper-aggresive tech companies in Taiwan are concerned.
A feeling of having made the same mistake before: Deja Foobar
Commodity 64.. did anyone except myself read this as Commodore 64???
Learn from the mistakes of others. There isn't enough time to make them all yourself.
Intel is already making up the numbers.
They can peg the clock wherever they like,
and just introduce wait states. In fact,
they have effectively already done this,
but call them pipeline stages.
A modern CPU is a hairy beast, and it has so
many physical metrics, with such a tenuous
relationship to application performance,
that you could pin just about any number you
like on it. Why stop at clocks?
People who are intelligent enough to butter
toast on the top use benchmarks anyhow.
-I like my women like I like my tea: green-
or was the Opteron also running the OS in compatability mode?
Lets wait and see what AMDs numbers say. I know there testing group and they have good things to say.
In the World of Covact
It's not the Opteron that can't use the extra registers, it's the software. Machine code is compiled to refer to specific registers by name, so if the software wasn't compiled to refer to the extra registers, of course it's not going to be able to use them.
x86 CPUs usually have a second layer of registers. The microcode uses these registers to execute multiple parts of the same program simultaneously. These registers aren't referenced directly by the machine code instructions, so they number in use is independant of the number of x86 registers available.
What's this Submit thingy do?
If you look at the second page of Tom's configurations, you can see that he used the onboard Trident chip to conduct workstation benchmarks. Trident?!! This is really inexcusable. You may as well run Windows SQL Server on a Pentium M and compare it against the Itanium--it's not even apples to oranges any more, it's more like apples compared to lettuce.
I know that Linux has NUMA capabilities...
Is the support that Linux has for Hammer good enough? How good is the SMP support for it?
We'll also be testing Opteron workstations as soon as they're completed. I expect the first reviews will be using the nForce3.
I think the point these folks are attempting to make is:
We've been doing the desktop dance for the last few years, IPCs fluctuating, but performance and power usage always rising.
Yet, a decade ago, Intel was strictly against this kind of practice. Intel chips for years were packaged in a big ceramic heat spreader, and could be run without a sink. There were no multipliers, memory ran at processor click speed, so there was little performance skew. Performance increased linearly with CPU clock speed. Even with the 486, Intel never released a chip with more than 2x multiplier, and purposefully never released a chip that required an active fan cooling the heatsink. It was only after the Pentium was released that they gave in to competition and pushed clock speeds, power disappation and multipliers.
What ever happened to the concept of the efficient computer? Sure, we've tackled the multiplier issues using multiple layers of efficient cache, paired with features like prefetch.
But what about power consumption?
RIGHT THIS MOMENT, we have laptop computers using the Centrino chip, AND THEY SPANK the P4 clock-for-clock.
RIGHT THIS MOMENT, we have laptops ion development using the extremely low power ATI Mobile DirectX 9 solution.
Both these chipsets are designed to maximize performance for power, and have the capability to shut down unused parts of the chip for maximum power efficiency.
AND NEITHER OF THEM ARE AVAILABLE FOR DESKTOP. NOR will they ever be. Nobody cares that desktop power consumption has quadrupled since the 486, and the processor's power consumption alone has increased by almost 10x.
We don't have to have windtunnel systems. We don't have to settle for VIA Epia boards with the horsepower of a 386. The sad fact is the perfect balance already exists, but you'll never see it except in an overpriced notebook.
Man is the animal that laughs.
And occasionally whores for Karma.
Why was memory performance poor compared to Intel?
Tomshardware really sucked this time.
Going to check out Acehardware test now.
how could they fuck up like that?
Who the hell cares about winxp and software specially optimized for p4s'.
Intel's chips are simply horribly inefficient, which is why hyperthreading works. One assumes that with a tight enough SMP kernel a hyperthreading P4 would approach the performance of a thoroughbred with the same PR as its clock rate.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
How would you propose to measure the speed of "just the cpu"? Performance is dependent on many factors like memory bandwidth/latency and compiler optimization. You can't just isolate the CPU, unless you choose your benchmarks to fit entirely in the caches which is obviously not at all realistic.
SPEC has been a respected benchmark used for 14 years - probably more since I can only recall SPEC89.
Nice effort to compare x86 processors, but what I would really love would be a comparision to RISC (UltraSPARC, POWER, ARM) processors and VLIW (EPIC: IPF) ones. After all, x86 is supposed to go away and x86-64 will have to battle VLIW and RISC, not x86-32.
Leandro Guimarães Faria Corcete DUTRA
DA, DBA, SysAdmin, Data Modeller
GNU Project, Debian GNU/Lin
Naaah, they'd just somehow relate it to Penis size, so only insecure guys who want bigger numbers would buy the latest models.
You'd have:
Intel "But if I shaved it would look longer..."
AMD "Feel the girth Baby!"
Intel "Ours now gets real hot."
AMD "You'll need a fan in your underpants for this one!"
Intel "Ron Jeremy uses this!"
AMD "Annabelle Cheong uses ours! No! Wait! Damn!"
dave
Of course you did ...
What are you on? Intel has in fact depened the pipe to get an architecture that scales to higher speeds, but except in invalid branches and cache stalls this works pretty darn well (it's nothing like wait states), and a lot of the impact from cache stalls can be minimized by having a second set of instructions flying in the hyperthreading unit (assuming it doesn't need data that will be staled out of cache so that it too stalls). The fact is that average case Intel's aproach works pretty well, best case it flies (see media apps), and worst case it can really be slammed, not bad tradeoffs on average. Just because AMD did not choose to go down the same road does not mean that it was an invalid design decision, in fact since most of the applications that actually stress a modern cpu have a similar profile to the media apps it seems to be a pretty good decision overall.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
most companies do zero or nearly zero research before ordering a $4K dollar workstation, they just ask what can they get for $4k that will run $CAD_OR_ENGINEERING_APP of choice. Trust me I have ordered probably a million worth of workstations over the last 3 years or so and other than ram amount and disk subsystem size I almost never ask what cpu is in it, I just order whatever HP or Dell is in the pricerange the manager has in mind and is certified on their divisions applications. We did order some oddball systems for one group that tends to order a large number of systems (asic synthesis and routing are major cpu users) just to see what we could do on price/performance but that's only because they tend to order 20+ workstations at a time for their work farms and it made sense to buy like 3-4 alternatives before getting a large order.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Am I the only one who thinks that ceramic packaging looks cooler than the OPGA (fiberglass like)?
I want my rights back. I was actually using them when our government stole them after 9/11.
Well, MHz's aren't everything are they. Maybe they should start putting BogoMIPS ratings on their ads.. ;)
What time is it/will be over there? Check with my iPhone app!
Duron 1000 would be a good name for a rock band.
</DAVE-BARRY>
Fascism starts when the efficiency of the government becomes more important than the rights of the people.
Are you by any chance talking about the heat spreader? The heat spreader isn't the size of the die. The die is 193mm^2, and the barton die size is 101mm2, which means the Opteron is about 2 times as big as the barton die.
The heat spreader makes it look huge, like it covers the entire packaging.
I want my rights back. I was actually using them when our government stole them after 9/11.
The Opteron die (the actually processor part, not the packaging around it, which includes the pins) is something like 193 or 194sqmm. Here's a quick list of other processor sizes:
Athlon Classic: 184sqmm
"Willamette" Pentium 4: ~217sqmm
"Cascades" PIII Xeon: > 300sqmm
Pentium Pro: 306sqmm(?)
"McKinley" Itanium II: 421sqmm
Pentium Classic: 294sqmm
As with the Opteron, these die size reflect the first introduction of a processor. In successive versions, process shrinks drastically decrease the area of the die. This is pretty normal.
As it is, the Opteron -- the *server* version of its family, is only slightly larger than AMD's previous generation's *consumer* level intro processor (the original Athlon). The consumer variant of the K8 core, the "regular joe" analogue of the Opteron, will likely have less L2 cache and fewer HyperTransport connections, so it's probably a given that this, the "Athlon 64" (I believe that this is what they're planning to call it), will start out at substantially less than the Athlon Classic's initial 184sqmm.
Granted, it'll have to, because Intel's consumer-level processors have die sizes under 150sqmm, and Intel is working with the benefit of larger production wafers.
-JC
What do you have against lettuce?
"...today consumers have been conditioned to think of beer when they see a bullfrog..."
In fact, they have effectively already done this, but call them pipeline stages.
r s/Deep%20Pipes.pdf
Intel did not just randomly increase the pipeline depth for kicks- it was a very specific design decision based on a lot of research. For example:
http://systems.cs.colorado.edu/ISCA2002/FinalPape
This study concludes that the P4 processor could be improved by increasing the pipeline depth. They theorize that the optimum pipeline depth for the processor is around 50 stages- more than double than its current 20 stage pipeline.
The way I see it, increasing the pipeline depth allows you to increase the clock speed. Sure you aren't doing as much per clock, but the increases in the clock speed more than make up for this (we have seen this in the P4 vs the Athlon over the past year). And to top it off, the higher clock frequencies are much easier to market.
Time will tell if AMD can use Opteron/Hammer and their SOI processes overcome thier problems scaling the clock. Otherwise Opteron's triumph might be pretty short lived.
"The defense of freedom requires the advance of freedom" - George W Bush
Ok, so he did Linux/MySQL instead of getting a license for 2K AS + MS SQL, a more common combination. The question still stands.
Itanic does not emulate x86. There is a little tiny x86 core stuck on the die of the Itanic. It is a horribly inefficient and slow x86 core, but it is the real thing.
Notice, that Intel wants you to think the hammers are competition only for Xeons. But Xeon's don't do 64-bit and Itanics do. Thing is, an Itanic costs 5-10x more than a a hammer (and uses 2-3 times the watts). So, they may not be competition on a flops/hz level, but on a flops/dollar or integer-ops/dollar or even flops/watt, the hammer is very competitive. And Intel knows it, they just don't want you to.
Check it out. Your next workstation?
As copyright owner of this comment, I authorize everyone to defeat any technological measure which limits access to it.
From the first page of AMD's technical datasheet released today:
Target CPU core power: 80.6W
Target maximum thermal power: 84.7W
I wish it was only 40 watts! Where did you find this number - was it an actual measurement? If so, was it at 100% cpu load, idle, or midway?
I thought this was a little low (esp. compared to my athlons), and didn't actually look it up until I ran across this EETimes article that claims 89 Watts for the AMD64 240.
HIV Crosses Species Barrier... into Muppets
The Opteron will be available in a desktop version. The Xeon is not, and will not be.
How so? The Opteron is to the Athlon64 ("the desktop version") as the current-generation Xeon is to the Pentium 4. In fact, the Pentium 4 is much closer to the Xeon than the Athlon64 is to the Opteron.
Tom didn't enable the second memory channel
The Athlon64 will not have a second memory channel regardless...
or use DDR400
Can't blame him for that. AMD does not officially support DDR400.
Uttering logically derived and empirically supported truths to the disciples of the orthodox establishment.
In theory, with 64 bits, you could address 16 exabytes. Each 64 bit address would refer to a byte, not a bit, so you don't divide by 8.
Thank you. Thank you. Please no applause; just throw money
AMD is not dead.
Sun on the other hand has terminal cancer.
Now that AMD has a scalable server processor solution, system vendors will be quick to integrate it into server systems. Sun will become irrelevant.
See here.
Patrick Doyle
I mod down every jackass who puts his moderation policy in his sig. Oh, wait a sec....
Intel's chips are simply horribly inefficient, which is why hyperthreading works.
This does not reflect the efficiency of Intel's chip - it reflects the fact that the most aggressively superscalar chip anyone could hope to build would still be lucky to execute even three instructions per clock on average.
Modern microprocessors are optimized to handle the peak load case - where control and data dependencies go on vacation and you have a string of instructions that you can burn through that also happen to perfectly match the number and type of functional units you have available.
In the real world, this rarely occurs. If you're very lucky, you'll have a tight loop that's reliably predicted operating on data that's mostly in L1, and keep _some_ of the functional units busy _most_ of the time.
The rest of the chip sits idle, waiting for the load profile to change.
This happens on *any* general-purpose processor. Intel, AMD, and everyone else suffer from it.
Symmetric Multi-Threading was proposed quite some time ago to cope with this. You'd have multiple instruction fetch units on the chip and run multiple processes. As these have totally independent instruction streams and mostly-independent workloads, they can be scheduled at the same time without conflict. This deluge of unrelated instructions lets the scheduler fill all slots and issue instructions to all functional units that *any* process has demand for. This drastically boosts utilization, with far fewer units being idle, at a cost of increased memory bandwidth load (mostly).
Intel got to market with this first, and called it "hyperthreading". IBM went the CMP route and didn't mess with SMT. Sun announced a future processor with both CMP and SMT features.
The problem occurs for everyone, and techniques to address it are well-known and being applied.
Claiming that Intel was forced to implement it due to inefficient processors is very silly.
That's pure bullshit. x87 floating point has always had 80bit precision in hardware, even when the 80x87 FPU was a co-processor on a separate chip from the x86 CPU (back in the days of 286, 386SX, and 486SX). "64bit CPU" can mean several different things, including address size (i.e. sizeof(char *) == 8), or integer register size (i.e. sizeof(int) == 8), but never size of floating point registers. Everyone has at least 64bit (IEEE double precision) floating point registers. Some have larger registers, such as x87, where sizeof(long double) == 10 (80 bits), or Sun SPARC's 128bit hardware floating point.
The real reasons for 64bit machines having better FP performance than x86 is probably only due to the fact that x86 sucks at FP, because it doesn't have enough registers. Not to mention the fact that until everyone started using SSE for even non-parallel FP ops, it was saddled with the stack-based x87. 64bit arches can't be backward compatible with ia32 (while running 64bit code), so they simply use good floating point design. More registers helps because you can do software pipelining, and have more intermediate results kept around, which is good for calculating big formulae where there are lots of common sub-expressions.
(BTW, some floating point hardware only runs at full speed in non-IEEE compliant modes, even if they have 64bit wide FP registers. What I said about everyone having IEEE double precision hardware is thus not strictly true. The differences are usually in FP exceptions (i.e. whether you can tell exactly which instruction caused a divide-by-zero) and rounding modes. This only matters for numerical algorithms whose stability depends on e.g. rounding toward zero, instead of always rounding down. This is what gcc -ffast-math or -mno-ieee-fp affects, and why games, and pretty much anything except Octave use that.)
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cordes ,
> They are too abstract and unreliable to influence smart buyers.
Unless you're actually choosing a computer based on how fast it can run crafty (the chess engine), gcc, perl, gzip, or bzip2. Those programs are part of the SPEC2000 benchmark suite. It is interesting to note that Athlons tend to run crafty relatively faster than the other benchmarks, probably because it's harder to predict branches or something along the lines of chess being unpredictable and thus more likely to screw up the P4 than the Athlon. (For a P4 and an Athlon with equal SPECINT2000 scores, the Athlon has a higher crafty score than the P4.) My dad ended up getting a P4 laptop anyway, and it's faster than any of the other computers we have, so we're pretty happy with it, so that just goes to show that CPU speed isn't everything.
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cordes ,
> yeaahh right, too bad : it's not a whole new generation of CPU
What else is in the same generation as Opteron, then? Nothing from AMD, which is what he meant.
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cordes ,
...with no NUMA support. Means it wasn't using the memory controllers optimally (only one channel used instead of two).
Tom's review was laughable at best.
A few months ago I was looking at buying a new computer, and I tried to find out anything about whole-life environmental impact of computing gear. i.e. including toxic waste produced during manufacturing, and ease of disposal/recycling. It was pretty hard to find anything useful about specific products. I want to know which hard drive was produced in the most environmentally friendly manner, even within the same company. The only web pages I could find were about laws in different countries; I don't think I even found anything about any specific manufacturers. One interesting site I found was the Silicon Valley Toxics Coalition. Other than that, there was some stuff from greenpeace IIRC, and some gov't web sites about international treaties.
Does anyone know how to buy an environmentally friendly computer?
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cordes ,
So, it's almost twice as expensive as the Xeon (according to PriceWatch), but it is not nearly twice as fast. Now I see.
--sdem
Those TPC-C Price Performance numbers are awesome but has anybody noticed that the rig runs $227,000???????
Damn... talk about Enterprise pricing!!!!!
Dissolve... Resolve... Evolve...
The tom's hardware artical says that 64-bit is faster because it "can natively calculate the important 64-bit floating point format". This is a rather shows up badly their limited knowledge of CPUs. The FPU has already been 64-bit for quite some time. I wonder how much of the rest of the artical is acurate. For instance does increasing the amount of registers from 8 to 16 really increase performance? Or does this (more likely in my opinion) make things more convient for people who write software in assembler?
I love the smell of CPU wars in the morning. Smells like... smells like coffee break.
True, the core's (basically) the same, but I think introducing a new processor mode and instruction set to go with it counts for something. All the integer units had to be 64bit now, and it probably took some changes to get everything to fit with twice as wide signal paths.
> !(whole new generation) != same generation
I usually think about generations as integers, and thus they're either equal or non-equal. If you're going to use fuzzy logic-style comparisons, then yeah, I guess there's too much that's still the same to say "whole new gen", but the new stuff does warrant a description of "new gen". OTOH, who says you have to fundamentally change the way the core works (like going from Alpha 21164 (in-order superscalar) to Alpha 21264 (OO execution)) to call it a new generation? I don't really care, since I want to know the tech details anyway, instead of judging by who's calling it a "whole new" or just "new" generation!
#define X(x,y) x##y
Peter Cordes ; e-mail: X(peter@cordes ,