Strange trend with GPU and CPU combo

[I_no]

" Your eyes are not deceiving you. After 100+ clean OS installs, countless video card, motherboard, memory and driver combinations, we have results that are not only repeatable, but appear to be valid. We also tracked in-game performance with FRAPS and had similar results. Put simply, unless we have something odd going on with driver optimizations, a BIOS bug, or a glitch in the OS, our NV cards perform better on the AMD platform than they do on the Intel platform. The pattern reverses itself when we utilize the AMD video cards. These results also repeat themselves in other games like H.A.W.X. and Left 4 Dead but not in Crysis Warhead or Dawn of War II. So, besides the gaming situation, we also see a similar pattern in AutoCad 2010 and other 3D rendering applications where GPU acceleration is utilized, it is just not as pronounced. "

(From Anandtech. Source http://anandtech.com/mb/showdoc.aspx?i=3639&p=3 )

Can anyone here explain this phenomenon to me

[OpTiX]

nice find!

[informal]

Phenom II is very balanced CPU with excellent I/O and memory subsystems-HyperTransport and IMC.The combination of these two and their hw implementation could possibly be the key in the results they observed(CPU-GPU communication).

[I_no]

http://www.tomshardware.com/reviews/...80,2156-2.html
Found this at Tom's, showing the same thing.

[I_no]

Phenom II is very balanced CPU with excellent I/O and memory subsystems-HyperTransport and IMC.The combination of these two and their hw implementation could possibly be the key in the results they observed(CPU-GPU communication).

Then why are the roles reversed with Radeons if its simply IO and CPU-GPU communication ?

[keithlm]

Then why are the roles reversed with Radeons if its simply IO and CPU-GPU communication ?

Drivers?

[EnJoY]

Wow...this is strange. That's not a single fps here or there either, that is significant.

Simon?

Gary?

What's up?

[informal]

Then why are the roles reversed with Radeons if its simply IO and CPU-GPU communication ?

Like keithlm said ,probably drivers.
One can also use NV based board for AMD system and test it with Radeons and NV cards to see if chipset is the culprit.

[Flambo]

Maybe now we can understand when nVidia said i7 isn't worth it for games back in April ...

[keithlm]

Like keithlm said ,probably drivers.
One can also use NV based board for AMD system and test it with Radeons and NV cards to see if chipset is the culprit.

I should expound:

In the last 6 months or so there have been one or maybe two times ATI has claimed they were adding performance improvements for multiple CPU. But it definitely seems to be a work in progress.

NEXT QUESTION: Does this trend also pertain to Crossfire vs SLI?

[wuttz]

when an amd cpu scores better than an intel cpu, they call it strange. when an intel cpu scores better than an amd cpu, then the benchmark is credible and the incessant intel leg humping starts.

cant we accept the benchmark results as it is? that an amd cpu, despite all benchmarks being compiled with intel optimization flags, still trounced the inferior architecture? i know using vendor neutral and open source benchmarks, amd processors are better, clock for clock.

this just proves the bias that exists at those benchmarketing sites.

[Zucker2k]

Then why are the roles reversed with Radeons if its simply IO and CPU-GPU communication ?

Good question! Poor answers tho, unless you want to believe that AMD writes bad drivers for its own platforms, etc. For now, the phenomenon can only be explained as follows: in bottle-necked scenarios "... Nvidia's card cannot translate the Core i7's microarchitecture into the same performance advantage, giving AMD's Phenom II-series chips the advantage...." - Tom's Hardware (quoted from link above).

[PoppaGeek]

From the article:

It is items like this that make you lose hair and delay articles. Neither of which I can afford to have happen. However, we have several suppliers assisting us with the problem (if it is a problem) and hope to have an answer shortly. These results also repeat themselves in other games like H.A.W.X. and Left 4 Dead but not in Crysis Warhead or Dawn of War II. So, besides the gaming situation, we also see a similar pattern in AutoCad 2010 and other 3D rendering applications where GPU acceleration is utilized, it is just not as pronounced.

Now it is a problem when a Phenom tops the chart.

I would have thought a ATI card on a ATI chipset would do a bit better than an Nvida GPU just because all the drivers are from one place. Guess not.

[soundood]

nice to see this in solid numbers, cause i have sort of guessed this for a while, thanks

[demonkevy666]

Like keithlm said ,probably drivers.
One can also use NV based board for AMD system and test it with Radeons and NV cards to see if chipset is the culprit.

it happens in 3dmark 06 ati card score lower cpu points.
sm2.0 and sm3.0 are both different too.

I should expound:

In the last 6 months or so there have been one or maybe two times ATI has claimed they were adding performance improvements for multiple CPU. But it definitely seems to be a work in progress.

NEXT QUESTION: Does this trend also pertain to Crossfire vs SLI?

it's like 1-5 % though

when an amd cpu scores better than an intel cpu, they call it strange. when an intel cpu scores better than an amd cpu, then the benchmark is credible and the incessant intel leg humping starts.

cant we accept the benchmark results as it is? that an amd cpu, despite all benchmarks being compiled with intel optimization flags, still trounced the inferior architecture? i know using vendor neutral and open source benchmarks, amd processors are better, clock for clock.

this just proves the bias that exists at those benchmarketing sites.

lol

Good question! Poor answers tho, unless you want to believe that AMD writes bad drivers for its own platforms, etc. For now, the phenomenon can only be explained as follows: in bottle-necked scenarios "... Nvidia's card cannot translate the Core i7's microarchitecture into the same performance advantage, giving AMD's Phenom II-series chips the advantage...." - Tom's Hardware (quoted from link above).

maybe it's cause the data from L1 cache is smaller in I7 then in phenom II.
I do like test the boards both idea that informal said.

From the article:


Now it is a problem when a Phenom tops the chart.

I would have thought a ATI card on a ATI chipset would do a bit better than an Nvida GPU just because all the drivers are from one place. Guess not.

LOL x2!

[knopflerbruce]

This isn't weird at all, just because i7 can run 8 threads with alot of computing power doesn't mean it's fastest in every situation. Deneb rocks in games, you just saw it This is no bug or "problem" or whatever people may think it is.

Facepalm @ those who are shocked by the fact that Phenom II actually is more than good enough by today's standards

[informal]

Funny how when Phenoms rock under linux (GCC compiled apps) then OS version is the culprit and then when Phenom rocks in games(delivering much higher minimum fps with NV cards which drivers are well optimized for fast multicore chips) then there is a "problem" with the NV cards or the system or something else . The fault/problem is clearly the Deneb core,its very existence

[wez]

This isn't weird at all, just because i7 can run 8 threads with alot of computing power doesn't mean it's fastest in every situation. Deneb rocks in games, you just saw it This is no bug or "problem" or whatever people may think it is.

Facepalm @ those who are shocked by the fact that Phenom II actually is more than good enough by today's standards

Well that wasn’t the “issue” either, the mystery lies with the performance of the said CPU+GPU combo, which doesn’t really make any sense

[Zucker2k]

Funny how when Phenoms rock under linux (GCC compiled apps) then OS version is the culprit and then when Phenom rocks in games(delivering much higher minimum fps with NV cards which drivers are well optimized for fast multicore chips) then there is a "problem" with the NV cards or the system or something else . The fault/problem is clearly the Deneb core,its very existence

Let's be a little objective here:

1. If the deneb core is this superior why is this anomally only evident with these two game titles?

2. Why does phenom ii ONLY "rock" when the gpu becomes bottle-necked? Why even allow the gpu to bottleneck in a test that is mainly aimed at showing cpu performance in gaming, and real world gamers do everything to avoid bottle-necking the gpu for best gaming results?

3. Why is phenom only very competitive on a marginal alpha OS which is clearly bugged and unoptimized?

4. Why would an enthusiast ignore all mainstream evidence in favor of fringe (read rare) results where the weaker platform gains an edge, especially when it involves third-party hardware whose developers have admitted the results are an anomally?

5. If the deneb core is clearly superior, why doesn't it dominate all tests at the same clocks, but actually comes last in all tests at the same clocks?

Yes, this is the AMD section, but how about a little objectivity? One can't just pretend that this section of the forum is a closed world where all manners of untruth can fly under the radar.

[Hornet331]

im more interested in the reply that NV will give... gimping performance on purpose, to hurt intel?... now that would be a epic (fail) move by NV

Good thing i never bought a NV card for the last 6 years.

Its also kind ironic to see that you get best performance with an AMD card on a intel system, while NV gives better performance on AMD....

edit:
lol just realized the toms hardware article is form february... gg NV for not addressing this even in the slightest way for half a year...

[Zucker2k]

im more interested in the reply that NV will give... gimping performance on purpose, to hurt intel?... now that would be a epic (fail) move by NV

Good thing i never bought a NV card for the last 6 years.

Its also kind ironic to see that you get best performance with an AMD card on a intel system, while NV gives better performance on AMD....

I thought about that possibility, but gave up the idea for the same reason you mention. It'll be suicidal, and too risky imo.

[wuttz]

^(sucker2k)your post is based on the premise that the dragon platform is a weaker design because it lags in most benchmarks, without taking into account that these benchmarks are heavily optimized with intel flags.

if you want unbiased and objective results, bench with an open source code and a vendor neutral operating system.

as far as im concerned, youre the one who lacks objectivity when the results of the benchmark favor amd.

[Zucker2k]

^(sucker2k)your post is based on the premise that the dragon platform is a weaker design because it lags in most benchmarks, without taking into account that these benchmarks are heavily optimized with intel flags.

if you want unbiased and objective results, bench with an open source code and a vendor neutral operating system.

as far as im concerned, youre the one who lacks objectivity when the results of the benchmark favor amd.

No need for a conspiracy theory....

What OS are you typing on right now? Since when did fringe become mainstream? No need to turn the world upside down for the sake of hardware.

The solution to question of you conspiracy theorists is simple: compare the hard numbers on that linux platform to ANY on the windows platform and you'd realize your results. If you feel your hardware runs better on linux, go for it.

[informal]

Zucker you have to read the article first

These results also repeat themselves in other games like H.A.W.X. and Left 4 Dead but not in Crysis Warhead or Dawn of War II. So, besides the gaming situation, we also see a similar pattern in AutoCad 2010 and other 3D rendering applications where GPU acceleration is utilized, it is just not as pronounced.

You see there is similar behavior in 2 more titles and plethora of apps which use GPU acceleration which rules out the "game" code.

Phenoms don not only rock when GPU is bottleneck,they do well all around in gaming. If GPU is the bottleneck how come Phenom II coupled with NV card has SO MUCH higher minimum fps? The NV GPU is not the bottleneck in AT tests,but something else is in intel platform while used with NV cards(Pcie implementation,chipset,I/O or something else may be the culprit for sub par performance with NV cards).

Linux is not marginal OS at all.It may be in your head,but in real world it is used a lot.Many DT users have dual boot(myself included) of Win XX and Linux. The version they used is clearly not bugged and unoptimized since phoronix use it for a while and in their opinion is one of the most stable and reliable thus far. If this wasn't the case,rest assured they wouldn't use it.

"Weaker platform" meaning lacking Turbo boost to OC itself? Yeah by that definition it is a weaker platform. Thank God that many experienced DT users auto tune their systems which with BE CPUs is a matter of 2 variables in BIOS or AOD-multiplier and Vcore. Walk in the park and is provides great fun. This weaker platform provides 6% less performance on average than i7-870 @ 2.93Ghz with turbo and SMT (on stock settings- rated with workloads from office, multimedia and games) :
http://www.computerbase.de/artikel/h...formancerating


Deneb core is not overall "superior" ,that's is not the issue.The things is that the picture is not black and white as one can see from the above review from plethora of applications.Deneb is very competitive across multiple workloads on the OS platform for which developers mostly code with intel in mind(using icc which as shown by Agner at Aces discriminates AMD cores with "GenuineIntel" check at compile time and assigns them less then best/optimal compile switches).

You can play your intel cop game,but the facts won't change. Deneb is very well performing core and AMD is not going anywhere.Deneb could use its own Turbo feature ,just for "benchmarketing" purposes,but I suspect AMD will incorporate one in near future(maybe as soon as the improved 32nm shrink version comes on the market).

[VoodooProphetII]

However, the difference between the ATI - AMD and ATI - INTEL combination is far less then the difference between NV – AMD and NV – INTEL combinations.

Whatever the reason, I’ll take any system that will play the games I want to play.

What I’ve picked up moving from GTX280 to borrowed 4870X2 and now GTX 295 is that my system do feel a lot more responsive with less garbage running in the background on both XP SP3 and Vista 32bit Home Premium. I’ve tried ATI and things worked, but the system just felt a lot more sluggish when compared to the system with Nvidia graphics.

The system I am using at present is a Phenom II 955 @ 3.6ghz 2 x 1Gb DDR3 @ 1600mhz 7 7 7 21 T1 and MSI 790FX G70 mobo.

But I do find it strange that if the tables were turned there would have been no hoohaa about this.