Tom's Hardware Forums » CPU & Components » CPUs » Anandtech Review => AMD's Quad FX: Technically Quad Core
 

Anandtech Review => AMD's Quad FX: Technically Quad Core




Word :   Username :  
 
 Page : 1 2
Next
Author
 Thread : Anandtech Review => AMD's Quad FX: Technically Quad Core
 
Profile: journeyman
More Information

Last message on previous page:
Lets just say AMD goes to 65mn and smacks a 8MB cache directly ondie just like Intel has done in the past to try and squeeze performance out of thier P4's i wionder what this will do for performance, besides its not good that people expected AMD to do what they did with the Athlon 64 bit again.
Just a thought

Related Product

Register or log in to remove.

Profile: nimble knuckle
More Information

I guess someone's got to put up some defense for QuadFX, if for no other reason than to have it...

http://www.legitreviews.com/article/425/1/

They tested everything with NUMA support (via XP Pro 64bit), and the Quad FX rig shows quite respectable results, beating the QX in a number of benchmarks. Most other benchmarkers failed to use an OS that is properly capable of using a dual-socket configuration with split memory. XP64 and Vista, on the other hand, are far more apt in this respect, and thus the performance increased drastically in almost every benchmark over XP Pro tests.

Since the people buying these systems are enthusiasts and in most cases will know these things, saying that no one uses XP64 and that Vista is not released to the mainstream market yet isn't the most watertight way to try and call the NUMA issue...well...a major issue. In about two months Vista will be out, and despite what many people claim, most people buying a new performance rig will be putting Vista on it, and Vista will perform better on the QuadFX platform than the 32-bit variety of XP.

As usual, the differing architectures have their own areas of exceptional performance over the competitor, and it seems K8 still has a little bite left in it in some of the benchmarks, though it's clear that it's about reached the end of its run and AMD is having issues pushing the clock speed higher (or can, but the energy and cooling requirements are just too ridiculous). QUICK maturation of 65nm is needed for AMD to scale the clock speed up, so they don't have to rely solely on architectural improvements.

With results like these, it really seems to me that anyone that expects Intel to be able to beat AMD's "K8L" systems by the end of next year are kidding themselves. If four K8 cores can almost come to parity with four Intel, what does it seem like four (if AMD releases the dual-core K8s for the QuadFX platform) or eight "K8L" (it won't be the named Barcelona core for the QuadFX platform, I believe, as that is reserved more especially for server only purposes, but it will be a derivative, clearly) will be able to do? The Intel bias here is really staggering, as the Barcelona improvements aren't that small.

The only significant drawbacks I can find with buying a QuadFX system over a QX6700 system are the power requirements, price (which essentially falls back on the need for two 2GB dual-channel RAM kits and the expensive motherboard, which there is only one of for the QuadFX platform at this time. The FX-74 processors aren't any more expensive than a QX6700, and with the FX-70 pair running at $599, one could build a "quad-core" system with $599 spent on the processors), and the high level of noise. If someone told me they were going to buy a QuadFX system now with FX-70s (rather than FX-74s) with the intent of upgrading to whatever Barcelona derivative finds its way to the QuadFX name, I would have to warn them about the downfalls, but wouldn't really push them away from it. If nothing else, it'd be nice knowing someone who could do it to see firsthand how well the new chips perform in a performance HT1.0 system (as opposed to HT3.0), which I might feel a little bad about, but wouldn't lose sleep about :lol: .

Bottom line: Intel is still unquestionably superior at the moment, but the QuadFX isn't as lame-duck as people are making it out to be.

edit - edited for spelling errors.

m25
Profile: Faithful Poster
More Information

Quote :

Lets just say AMD goes to 65mn and smacks a 8MB cache directly ondie just like Intel has done in the past to try and squeeze performance out of thier P4's i wionder what this will do for performance, besides its not good that people expected AMD to do what they did with the Athlon 64 bit again.
Just a thought


Benchmarks show that a K8 core moving from 512k to 1024k L2 benefits almost nothing, too few performance to justify the added cost and correctly, AMD will equip it;s future chips (FX and Opterons apart) with 512K or less.

m25
Profile: Faithful Poster
More Information

Quote :

case in point,the elusive sempron x2 3600 benches,nearly identicle to full fledged 64 chips at that clock.

thats rumored to change with k8l
.


I love the 3600+, had they thought about it before they'd have put that cache amoount on all the athlos and got some decent cost cut maintaining a healthy performance.
BTW, what are these rumours about?

m25
Profile: Faithful Poster
More Information

Well, if K8L benefits from larger cache, than it means it can use the extra information on it and thus, means it's efficiency and IPC might have grown quite a bit. It has totally redesigned SSE units and internal busses widened and still many people insist it's performance will be considerably inferior to C2D (most of them talf about actual K8 numbers).

Profile: Forum Resident
More Information

Quote :

I guess someone's got to put up some defense for QuadFX, if for no other reason than to have it...

http://www.legitreviews.com/article/425/1/

They tested everything with NUMA support (via XP Pro 64bit), and the Quad FX rig shows quite respectable results, beating the QX in a number of benchmarks. Most other benchmarkers failed to use an OS that is properly capable of using a dual-socket configuration with split memory. XP64 and Vista, on the other hand, are far more apt in this respect, and thus the performance increased drastically in almost every benchmark over XP Pro tests.

Since the people buying these systems are enthusiasts and in most cases will know these things, saying that no one uses XP64 and that Vista is not released to the mainstream market yet isn't the most watertight way to try and call the NUMA issue...well...a major issue. In about two months Vista will be out, and despite what many people claim, most people buying a new performance rig will be putting Vista on it, and Vista will perform better on the QuadFX platform than the 32-bit variety of XP.

As usual, the differing architectures have their own areas of exceptional performance over the competitor, and it seems K8 still has a little bite left in it in some of the benchmarks, though it's clear that it's about reached the end of its run and AMD is having issues pushing the clock speed higher (or can, but the energy and cooling requirements are just too ridiculous). QUICK maturation of 65nm is needed for AMD to scale the clock speed up, so they don't have to rely solely on architectural improvements.

With results like these, it really seems to me that anyone that expects Intel to be able to beat AMD's "K8L" systems by the end of next year are kidding themselves. If four K8 cores can almost come to parity with four Intel, what does it seem like four (if AMD releases the dual-core K8s for the QuadFX platform) or eight "K8L" (it won't be the named Barcelona core for the QuadFX platform, I believe, as that is reserved more especially for server only purposes, but it will be a derivative, clearly) will be able to do? The Intel bias here is really staggering, as the Barcelona improvements aren't that small.

The only significant drawbacks I can find with buying a QuadFX system over a QX6700 system are the power requirements, price (which essentially falls back on the need for two 2GB dual-channel RAM kits and the expensive motherboard, which there is only one of for the QuadFX platform at this time. The FX-74 processors aren't any more expensive than a QX6700, and with the FX-70 pair running at $599, one could build a "quad-core" system with $599 spent on the processors), and the high level of noise. If someone told me they were going to buy a QuadFX system now with FX-70s (rather than FX-74s) with the intent of upgrading to whatever Barcelona derivative finds its way to the QuadFX name, I would have to warn them about the downfalls, but wouldn't really push them away from it. If nothing else, it'd be nice knowing someone who could do it to see firsthand how well the new chips perform in a performance HT1.0 system (as opposed to HT3.0), which I might feel a little bad about, but wouldn't lose sleep about :lol: .

Bottom line: Intel is still unquestionably superior at the moment, but the QuadFX isn't as lame-duck as people are making it out to be.

edit - edited for spelling errors.




To interject, the site link shows that neither of these platforms with 8800 SLI is easy on power.


http://www.legitreviews.com/images/reviews/425/power_idle.gif

http://www.legitreviews.com/images/reviews/425/power_load.gif

As is shown, both of these systems are amazingly power-hungry though C2Q tends to tame it slightly being on a 65nm process. These may end up as the first FX EEs by the time Agena FX ships. 700W load. 8O

C2Q at nearly 600W load. 8O


ANd amazingly enough the power difference is almost exactly the difference of 1 FX74. Well, I guess that isn't amazing.

Profile: Forum Veteran
More Information

Quote :

with early bragging about barcelona,i think we are in for a great chip.

it will be then that we can dissect the core.Amd doesnt brag,and if tey started pumping hype like intel with prescott,it would hurt them immeasurably.

I'd have to disagree about that. They were pretty cocky while K8 was beating down Netburst. Remember who issued the dual-core challenge? :wink:

Profile: Forum Resident
More Information

Quote :

Point taken, those 8800 while screamers are really sucking down the electrons.



I guess the next nVidia shrink won't be far off, especialy since they're sayign a FASTER G80 is in the works. Hopefully R600 @ 80nm will help. I mean, buying DX9 cards right now is crazy.
I did hear that nVidia ready to do 65nm also as UMC and TSMC have gotten to 45nm SRAM.
A 65nm G80 will definitely cut power a lot.

Profile: addict
More Information

http://www.tgdaily.com/2006/09/25/amd_idf-2006/
AMD sends "Multi-Core Processing for Dummies" to IDF

http://digitimes.com/bits_chips/a20060313PR201.html
"We've seen, at least in a couple of areas, Intel acknowledge that AMD was right."
"Then their new-generation micro-architecture (NGMA), is, quite frankly, a quick fix on the front-side bus."

And in general, it's never a pretty sight when AMD talks about Intel in interviews :D

Oh yeah..
www.amd.com/duel

Quote :

Intel did not Participate in the Dual-Core Duel
10.Tried to follow their own roadmap to get to the duel
9.Decided to take the "front-side bus" to the duel; got stuck in a bottleneck
8.The "Intel Inside" stickers they used to package the cores together keep melting
7.Too busy rearranging the deck chairs on the Itanic
6."Hey, we don't expect anyone to actually buy these things!“
5.Didn't want to compete when they realized that the duel would involve actual "rules" of fair competition
4.They couldn't get a permit from the fire department to emit thatmuch heat
3.No systems available yet --protective clothing used by manufacturers only safe for up to 149 watts
2. Dell told them they weren't allowed to participate
And the number one reason Intel didn’t accept the dual-core duel:
1. Moore's Law has been replaced by "Paul's Paradox": the number of canceled products per year at Intel will double every year after the introduction of the AMD Opteron™processor.


:lol:

Profile: nimble knuckle
More Information

I would have to say AMD was bragging, but at the time, they needed to, since they were trying to pull themselves up from the underdog position (which, I would say, they did with incredible effectiveness).

Profile: addict
More Information

I don't think bragging gets you anywhere, it just makes you look silly. Apple's been ragging on Windows for forever.. Look where they are :lol:

Profile: nimble knuckle
More Information

Quote :

I don't think bragging gets you anywhere, it just makes you look silly. Apple's been ragging on Windows for forever.. Look where they are :lol:



Point taken! :)

Profile: nimble knuckle
More Information

Quote :

I would have to say AMD was bragging, but at the time, they needed to, since they were trying to pull themselves up from the underdog position (which, I would say, they did with incredible effectiveness).



well there is a difference between bragging and saying it for what it is.i think AMD is stating the truth,but it comes off as a boast because we cat see the benches.

I think its real.I dont care if i am proved wrong,because if AMD is talking out their butts;they will loose more customers back to intel.
And they know it.

I agree only to a point. In terms of marketing, they HAVE to say that their new processors rock, even if they don't.

But I agree, they probably will.

Profile: nimble knuckle
More Information

Quote :

this was just a "preview release" so to speak.Just a luxury look at things to come.with its issues AMD is not expecting to sell alot of these,and the purpose was to put their next great idea in the hands of us whiny enthusiasts regardless of polish ,just to shut us up.

of course it has enraged us in lieu of am2 and c2d :lol: :lol: :lol:
and I think that was expected as well.When the time comes the platform will rock.Until the it gives intel nothing interms of a reason to up the anty,it gives enthusiasts something to whine about,and buys them alot more time .



QuadFX is more of a stopgap, with the somewhat-intended consequence of being a platform that, in the future, could entirely dominate the high-end market.

Profile: nimble knuckle
More Information

Quote :

I have read other your 'calculations' lately and I tell you one thing; the greatest mistake in this field is taking a calculator and making divisions to estimate percentage gains. Despite the name, K8L is qualitatively different from K8 and o many aspects, substantially different from Core2, so you'd better not assume K8L is K8 bec