First Looks at Athlon 64 4000+ & FX-55

CrzyP writes "AnandTech.com has benchmarked the new "Athlon 64 4000+ and the FX-55" in various areas including business application performance, audio/video, gaming, and much more in this first look at AMD's newest 64bit chips. Just after AMD's announcement, AnandTech posted this article to help consumers choose between Intel and AMD."

Comments

Also here

[elid]

on Extremetech

Spread the love

[Hedon]

Why should Anand get all the attention?
Feel free to also check http://www.hardocp.com/article.html?art=Njc1

I wonder

[robslimo]

With Intel having recently backed off on the effort to push clock rates ever higher, is there a plateau in sight for AMD? Will we not see anything between 5 to 10 GHz with today's techniques?

Maybe it'll take optical computing to spur the next clock push.

Re:I wonder

[iggymanz]

clock speed isn't the only way to do more work in less time, and get better balance with the current bottlenecks getting to memory & i/o -- time for some new architectures, instead of the same old stuff with smaller transistors and higher clock speeds

Re:I wonder

[Martin Blank]

Indeed. AMD's designs didn't scale as well speed-wise, so they had to get more creative to get better performance out of their chips. Improving their branch prediction, boosting FPU speeds, enlarging the cache, and improving the data bus were all methods that Intel would go back to from time to time, but usually with some reluctance.

Re:I wonder

[swordboy]

With Intel having recently backed off on the effort to push clock rates ever higher, is there a plateau in sight for AMD?

There's some information to be realized:

AMD uses IBM's Silicon on Insulator (SOI) technology. This reduces power consumption by a very large degree. It is rumored that Intel tried to license the technology but, IBM and their fondness for cross-licensing, wanted too much (probably an x86 license). So Intel has been pushing out chips with standard silicon fabrication techniques at the expense of tremendous power consumption.

My guess is that Intel is coming up with a "massively parallel" architecture that can be applied to mainframes all the way down to handhelds simply by reducing the number of cores on a chip. The cores, will probably be very small and flexible. A mainframe might have a few thousand while a handheld might have a few dozen. They've certainly been hinting at a change in architecture for some time.

And then there was the "Windows Elements" that was supposed to come out with the P5. I'm not sure why that didn't get more press. I'm guessing that it is a version of Windows that will run in local storage on these processors (i.e. - the processor will have enough on-chip storage to hold "Windows Elements").

Re:I wonder

[ViolentGreen]

I think this is a lot of the reason AMD is starting to pull ahead of intel. Intel was able to just increase the clock speed to make their chips "faster" than AMDs. AMD had to look for other methods to increase performance and perhaps they learned a bit along the way.

Re:I wonder

[Anonymous Coward]

"130nm process, max 1Ghz"
???

130nm Northwood 3.2Ghz
90nm Prescott 3.8Ghz
65nm ???

and tom's hardware

[he1icine]

another review on tom's hardware

http://www.tomshardware.com/cpu/20041019/index.h tm l

Re:and tom's hardware

[Sivar]

Surprisingly, that review seemed to be unbiased and made few if any rediculous claims based on miniscule performance differences. If they keep this up, I might start reading them again.

Anand does not know much about SC manufacturing

[bstadil]

CPU manufacturing is all about yields, if AMD can make more chips that work by increasing the die size by adding a larger cache instead of upping the clock speed, then that's the route AMD will take.

This is actually the last resort, as the cost of wafer real-estate versus speed increase is low. You rarely do this for raw speed rather for special needs like Servers and the like.

The increase in the speed for a workstation is probably one speed grade at a cost increase of 30% or so.

There is two good articles over on TheInquirer about Intels road map and why they have to go the Increase the cache route for 2005. Worth a read. Part One and Part Two

Nforce4 also coming in a couple weeks..

[Anonymous Coward]

This is a major overhaul of the aging nforce3 chipset.. Check it out.

Expect a flurry of new advances by the end of the year.

I am ready to buy a new Linux system and am pulling hair out trying to make the best choice. Due to Linux compatibility issues (and mixed experiences with nforce2), I cannot really consider nforce4 so it will be Via for me. Though Nvidia will likely get the nod for graphics.

The 90nm chips are a mixed bag at the moment.

Re:Nforce4 also coming in a couple weeks..

[freelunch]

VIA has more problems than I can count.

Here we are not talking about the MB manufacturer but rather the chipset at the heart of the MB.

For example, the MSI K8N Neo MB exists in a version with the NVIDIA nForce3 chipset and in a version with the Via K8TPro chipset. Just as ASUS uses both chipsets. Etc.

The gotcha in all of this is buying for Linux. All of these new boards and chipsets coming out is Just Great. Reviews that focus on Linux are a huge step forward. But if your primary focus is Linux, being in the early group of folks to give a new product the go can be a Real Bear..

Right now, we're about to see a whole new generation of faster MBs come out. At the moment, I still haven't figured out which MB/Athlon 64 combo is ideal and I've been wanting to place the order for a few days. As much as I want to wait for XXX to release YYY, I have to keep telling myself that my Linux requirement means it is the support date and not the release date that matters most.

Power density

[lagartijo]

The power density at nm process (watts per square inch) has reached the nuclear reactors. See page 8. http://cnscenter.future.co.kr/resource/rsc-center/ presentation/intel/spring2003/S03USCQNS67_OS.pdf It's intel's but I assume it is the same for AMD.

More Links

[JaF893]

http://www.hothardware.com/viewarticle.cfm?article id=592&cid=1
http://techreport.com/reviews/2004q4/athlon64-fx55 /index.x?pg=1
http://www.bit-tech.net/review/364/
http://www.short-media.com/review.php?r=266
http://www.pcstats.com/articleview.cfm?articleID=1 666
http://www.tbreak.com/reviews/article.php?id=331
http://www.amdreview.com/reviews.php?rev=fx-55-400 0
http://www.techwarelabs.com/reviews/processors/amd 4000_fx55/
http://www.hardocp.com/article.html?art=Njc1
http://www.xbitlabs.com/articles/cpu/display/athlo n64-fx55.html
http://www.sudhian.com/showdocs.cfm?aid=614

amd bias?

[uniqueCondition]

if you're going to stand your ground with wintel and attack reviews from tomshardware & co. then i have to ask what you take issue with.

did you disagree with the test system?
the benchmarks used?

i've read tomshardware for years and have found them objective and informative. While their results disagree with your emotion you shouldn't make baseless remarks

Duh

[igzat]

I think the choice is clear regardless of this article. Intel announced last week that they are giving up a 4Ghz Pentium CPU, and even the 3.8 Ghz model is very scarce. Where as AMD's Athlon 3800+ can be easily found, With the announcement of the 4000+ CPU, AMD has a clear lead over intel, and will until the Dual-CPU wars begin sometime next year. I think now is a good time to own AMD stock. Their marketshare is going to slowly increase over the next 12 months. I'm not taking sides here, just stating the obvious.

Re:Like my boss said...

[Anonymous Coward]

really??

we lost a purchasing person at corperate because he bought intel.

we asked for some SGI workstations for a specific project. the nimrod decided he could save us $$$$thousands by buying Intel Based Dells instead.

He was fired.

Re:Impartial? hah.

[harrkev]

I certainly don't see that.

Well, in a way I do. They certainly are biased towards AMD, but only in the sense that AMD generally offers better value for the money, and is the only way to get 64 bits (right now). I bet if given the choice between a thousand dollars or a kick in the head, they would be biased toward the money ;)

All Intel has given us lately is a new extra-fragile socket, and PCI express (but good luck finding a PCI express vid card). In short, ho-hum. In a year or two, PCI express will be a good thing, but I will pass on it for right now.

Re:Impartial? hah.

[ergo98]

"AnandTech and Tom's Hardware are so AMD-biased"

Given the current political climate there seems to be a lot of ignorance over the difference between "bias" and "fact". Indeed often people confuse a lack of bias with a counter bias.

For instance, if indeed AMD is the superior consumer chip, perhaps offering measurably better performance/value, then it is entirely reasonable that a site would say such, and it isn't a "bias" to pronounce the AMD the superior choice of the current candidates. Similarly if George W. Bush dines on kittens for dinner, and someone reports it, that doesn't mean that they are biased against Bush.

Apples to apples?

[Quixote]

I'm reading the review right now (I know, I should burn my /. membership card), and the first thing that jumped out at me were the difference in memory specs between the AMD setup and the P4 setup:

AMD: 2 x 512MB OCZ PC3200 EL Dual Channel DIMMs 2-2-2-10
Intel P4: 2 x 512MB Crucial DDR-II 533 Dual Channel DIMMs 3-3-3-12

Why not keep the rest of the components exactly the same, so we can have a _real_ comparison?

I'm no Intel fanboy (or an AMD fanboy, for that matter), but when you're doing such benchmarking, some attention to details would help.

Re:Apples to apples?

[Quixote]

Oops, my bad: I didn't notice the "DDR-II" in the specs.

Re:Impartial? hah.

[leathered]

AnandTech and Tom's Hardware are so AMD-biased, I don't even bother reading them any more, especialy when it's related to the latest AMD or Intel chips.

Oh please, without any evidence or even anecdotes to back this up you're just making yourself sound like yet another disgruntled fanboy when reviews aren't going your way.

Almost all hardware sites agree that at the present, not only are AMD's chips providing the best performance, they also provide the most bang for the buck. That now even goes for media encoding, an area previously dominated by Intel. Who knows, no doubt the see-saw will swing back in Intel's favour in the future, but now it's clear that AMD will hold the lead throughout 2005.

Maybe some of you should read the article...

[adiposity]

The 4000+ isn't clocked any higher than the 3800+, it's just got a bigger cache. It's basically an FX-53; in fact, that's exactly what it is, sans the name. It would seem AMD is plateauing as well, but perhaps 90nm will get them out of the jam later on.

However, this is a wise move by AMD even if the rating isn't justified (hint: the benchmarks say it's not). Intel will never have a 4GHz CPU, and idiots who don't understand performance will see the 4000+ and want it because it breaks the 4000 barrier. It could backfire, but probably not, because even though 4000+ isn't justified, it's still faster than any of Intel's chips on 90% of applications.

-Dan

Re:forget the new CPU's

[AliasTheRoot]

If you had read the article you would have noticed it prominently linked to a partner article on the Nvidia nForce 4 chipset, in fact they went as far as using an nForce 4 benchmark system:

Socket-939 Athlon 64 CPUs
2 x 512MB OCZ PC3200 EL Dual Channel DIMMs 2-2-2-10
NVIDIA nForce4 Reference Motherboard
ATI Radeon X800 XT PCI Express

Power Consumption - how much is waste heat?

[Bryan Ischo]

This very thorough article also includes a comparison of power usage of the various processors during idle and busy states. The numbers look HUGE - the 90 nm Athlon 64 3500+ does the best at 86 watts at idle, with the Intel P4 560 (3.6 Ghz) doing the worst at 124 watts. While under a workload, the range is 114 watts to 210 watts.

At first I couldn't believe my eyes - how can heat sinks keep up with these figures? But then I realized that only some of that wattage is being converted to waste heat - some of it is actually doing the useful work of the processor.

Just curious - does anyone have any idea what the likely waste heat dissipation, in watts, would be for these processors, given the total power consumption figures in the article?

Re:Power Consumption - how much is waste heat?

[Bryan Ischo]

So basically, the amount of energy it takes to perform calculations is tiny? If processor A performs 2 billion arithmetic operations per second, and it is able to perform each operation just as efficiently as processor B which only performs 1 billion operations per second, the I would expect processor A to use twice as much energy performing its calculations as processor B.

But what you're saying is that the amount of energy being wasted as heat for both processors A and B is 99%, so the extra power used by processor A in its calculations won't be noticeable compared to processor B (assuming that the only extra power used by processor B is that used to perform calculations).

486s comsumed what, 10 - 20 watts? And they performed something like 1/100 or fewer as many arithmetic operations per second as today's processors? So they used 1/5 the power but performed 1/100 the amount of useful work. I guess that today's processors actually convert more of their input power to useful work (calculations) than processors of the past did.

Re:Power Consumption - how much is waste heat?

[mczak]

you misinterpreted that power consumption graphs. This is total system power consumption, not only the cpu!
So if it says 100W, that is 100W measured at AC! Since psus are only 65-80% efficient, that means the system (without including psu loss) is only using 75W. If you keep in mind this includes hd, graphic card, mem, chipset,..., this doesn't leave that much for the cpu actually. Measuring system power also makes the differences in cpu power consumption look much smaller than it is in reality obviously.
And others have mentioned it already, ALL power is transformed to heat.

Re:Power Consumption - how much is waste heat?

[NerveGas]

How can heat sinks keep up? If you've seen the size of heat sinks that come with these processers, you'll understand.

I built an P4/LGA system for a guy last week. The heat sink that came standard with the CPU really impressed me - it's the kind of heat sink you would have expected to see hardcore overclockers paying $60 for two years ago. Very large and well-designed!

Times used to be when heat sinks weighed one or two ounces, and came with 40mm fans. Then came the 60mm fans. Now, 80mm fans and two-pound aren't at all uncommon, with some models using 92mm fans, and some weighing three pounds or more. Copper is being used for more and more of the heat sink. Better heat conduction, more surface area, and more air. It's not rocket science. : )

Plus, on the new P4's, the chips are able to run at much higher temperatures than previous generations. The greater temperature differential between the chip and the heat sink, the faster you can get the heat to conduct.

steve

snicker

[DeathByDuke]

all these accusations of Tom's Hardware being AMD biased makes me laugh. Just be happy for once we dont have to moan about Intel biased websites anymore ;)

It's ALL "waste" heat.

[raygundan]

All of the energy going into the processor is going to come out as heat. It's similar to what would happen if you put a lightbulb in a box, and then measured how much heat was being produced outside.

Some of the power going to the lightbulb makes waste heat directly, and some of it makes light. But since it's all closed up in a box, all of the light ends up making heat, too.

So yes, some of the power going into the processor does useful work. But from the point of view outside the processor at the heatsink, even the useful work creates heat.