G70 Vs X1800 Efficiency

[russo121]

http://www.driverheaven.net/articles/efficiency/

Sure series X1800 are a looser all the way.

Edit: I don't know if already been posted.... Anyway moderator feel free to delete

 

[Skrying]

Interesting, but this tells me something else though. The extra 8 pipelines for the 7800GTX doesnt seem to be doing jack.

 

[DemoCoder]

Or is it that the extra Mhz of a fully clocked R520 is doing jack? You can't come to the conclusion that extra pipelines are doing nothing unless you compare apples to apples. Compare a 16-pipe GTX to a 24-pipe GTX both running the same benchmark.

 

[arjan de lumens]

Hmmm.. did they make sure to turn off the angle independent AF that R520 offers? The performance hit caused by the angle independent AF may very well explain at least some of the game results.

 

[jb]

Dont forget drivers may play a part of this...as they are only doing shadermark, and 3dmark which were more PS2.0 apps...so it would be interesting to when more "PS3.0" test are ran.... Still kudos to the GTX

 

[Pete]

to quote myself

It's an interesting test, though I would've liked to have seen more games benchmarked. Very interesting to see by how much R520 loses in Doom 3, clock for clock.

I have to wonder if the X1800 doesn't lose a little more in the comparison because it's being downclocked so much, whereas the 7800GTX is actually being overclocked slightly. For instance, the XT's memory originally runs at 1500MHz. Does it run with higher latency (like, say, CAS 2.5 vs. CAS2, or 2T vs. 1T) than the 7800GTX's 1200MHz memory? Does this latency remain at lower clocks? Also, I think they forgot that the G70 has its vertex shaders clocked higher than its pixel shaders when they ran the 3DM tests:

Vertex shader figures are also interesting, as both cards have 8 vertex shader units and are clocked at 450/1000 we can really see which architecture processes vertex shaders more efficiently. In this test the result goes to the R520 when simple vertex shaders are used. However when more complex shaders are used the G70 just outperforms R520.

I also wonder if they took the G70's clock plateaus into account (the G70 apparently only OCs in specific increments, not by single Hz as with older cards. So, for instance, even though they set the G70 to 450MHz, did it actually run at that speed? (The 3DM MT fillrate number seems to indicate it did.)

Nice to see such a test is possible, though. Yeah, it's not really practical, but it's an interesting theoretical exercise.

 

[Joe DeFuria]

The whole theory behind these benchmarks is pretty flawed. I mean, It's sort of interesting from a theoretical standpoint...but from a product standpoint?

Part of the design architecture of R520 is to allow for higher clocks...trades transistors / pipes for clock speed.

The G70 design went the other way. (Wider pipes, lower speed.)

So when nVidia ships a card that clocks as fast (and is able to handle the higher speed memory), let me know. At the same time, let me know when ATI implements 24 ROPs and texture units...Until then, I'll just scratch my head.

 

[OpenGL guy]

http://www.driverheaven.net/articles/efficiency/

Sure series X1800 are a looser all the way.

Edit: I don't know if already been posted.... Anyway moderator feel free to delete

Completely uninteresting. What is interesting is comparing R520 to R480 at the same clocks. Comparing chips from different manufacturers at the same clock speed is not interesting.

 

[Blastman]

Does t he X1800 support DST? If not they are running different render paths in 3Dmark5. The 6800 takes a ~ 11% drop in 3Dmark5 scores when not running the DST path.

 

[wireframe]

Completely uninteresting. What is interesting is comparing R520 to R480 at the same clocks. Comparing chips from different manufacturers at the same clock speed is not interesting.

Why on earth would it be more interesting to compare R520 to R480 than to G70? If ATi claims the R5xx is a new architecture then surely it is different enough from both R4xx and G7x to warrant such comparisons.

Perhaps I am missing something, but this sounds like you want to only compare the R5xx to something it can "win" against.

Oh, and I was extremely surprised to see this come out of Driverheaven. Seems to me that usually they have very inflated numbers in favor of ATi. I thought they would have burned the paper with results like these and instead the published them on the Web for everyone to see . Good for them. :smile:

 

[Moloch]

Completely uninteresting. What is interesting is comparing R520 to R480 at the same clocks. Comparing chips from different manufacturers at the same clock speed is not interesting.

QFT.
I'm not interesrest in two completely different design aspects from different companies, I'm interested in how either companies video cards compares to the previous generation clock for clock.

Why on earth would it be more interesting to compare R520 to R480 than to G70? If ATi claims the R5xx is a new architecture then surely it is different enough from both R4xx and G7x to warrant such comparisons.

Perhaps I am missing something, but this sounds like you want to only compare the R5xx to something it can "win" against.

Oh, and I was extremely surprised to see this come out of Driverheaven. Seems to me that usually they have very inflated numbers in favor of ATi. I thought they would have burned the paper with results like these and instead the published them on the Web for everyone to see . Good for them. :smile:

It's interesting because ati quotes effieciency gains and so comparing it to the last gen product should be interesting.
Just like to me It's uninteresting to compare a P4 clock for clock to a A64, I'm more interested in minior updates they do, like presscott vs northwood, where we see intel trading even more clock for clock performance for pure clock speeds, and we see amd striving for even more clock for clock performance comparing their 90nm parts to their 130nm parts.
When you know a company has a different goal in it's products compares to a competetors, it's not that interesting because it's obvious.
I find it more interesting if the trade offs worked out.
In ati's case it did.
In intels case, not so much.

 

[Humus]

Interesting, but this tells me something else though. The extra 8 pipelines for the 7800GTX doesnt seem to be doing jack.

Perhaps the option in RivaTuner doesn't work so it's actually still running with all pipes?

But I agree with OpenGL guy, this is not a particularly useful test. If we'd clock down a P4 to Athlon level, and find they run extremely slow in comparison, does that mean the P4 architecture is vastly inferior to the Athlon?

 

[Moloch]

Perhaps the option in RivaTuner doesn't work so it's actually still running with all pipes?

But I agree with OpenGL guy, this is not a particularly useful test. If we'd clock down a P4 to Athlon level, and find they run extremely slow in comparison, does that mean the P4 architecture is vastly inferior to the Athlon?

In this case yes..
When one factors in performance per watt it's obvious.
However P4 does very well in every day tasks and media encoding.
The dual core battle shifted the balance back to amd for the most part though.

 

[Humus]

Does t he X1800 support DST? If not they are running different render paths in 3Dmark5. The 6800 takes a ~ 11% drop in 3Dmark5 scores when not running the DST path.

Yes it does. Not sure if it's exposed in the same way though. Reading nVidia's programming guide it seems to be different.

 

[Nite_Hawk]

Perhaps the option in RivaTuner doesn't work so it's actually still running with all pipes?

But I agree with OpenGL guy, this is not a particularly useful test. If we'd clock down a P4 to Athlon level, and find they run extremely slow in comparison, does that mean the P4 architecture is vastly inferior to the Athlon?

You might want to be careful with that analogy. Some people do consider the P4 an inferior architecture primarily because the (precieved) design goal is an increase in frequency (not necessarily in increase in speed). This probably would not be the case if the chips would run cooler and scale to higher frequencies.

Nite_Hawk

 

[Moloch]

You might want to be careful with that analogy. Some people do consider the P4 an inferior architecture primarily because the (precieved) design goal is an increase in frequency (not necessarily in increase in speed). This probably would not be the case if the chips would run cooler and scale to higher frequencies.

Nite_Hawk

++
P4 failed and ati has been doing pretty good except with some driver effciency problems in OGL.
Of course though with a GPU any increase in clock speed improves your fillrate provided you have enough bandwidth to support the increased fillrate and of course enough pipes.

 

[suryad]

You might want to be careful with that analogy. Some people do consider the P4 an inferior architecture primarily because the (precieved) design goal is an increase in frequency (not necessarily in increase in speed). This probably would not be the case if the chips would run cooler and scale to higher frequencies.

Nite_Hawk

Agree 100%. In fact wasnt there an article out where they overclocked the P4 to 5.2 ghz and it barely was able to surpass the 2.6 ghz Athlon FX? So yeah the p4 is an inferior architecture.

While the article is interesting, I fail to draw a definite conclusion as to which cards are superior at the moment. I think ATI has something cooking with the R600. This looks to me like its an engineering experiment to just compete with Nvidia and not surpass it. I could be wrong though. Who knows?

 

[Bouncing Zabaglione Bros.]

The whole theory behind these benchmarks is pretty flawed. I mean, It's sort of interesting from a theoretical standpoint...but from a product standpoint?

Part of the design architecture of R520 is to allow for higher clocks...trades transistors / pipes for clock speed.

The G70 design went the other way. (Wider pipes, lower speed.)

So when nVidia ships a card that clocks as fast (and is able to handle the higher speed memory), let me know. At the same time, let me know when ATI implements 24 ROPs and texture units...Until then, I'll just scratch my head.

My thoughts exactly. It's like comparing a family saloon to a Formula 1 car and saying the saloon is more efficient once you've downclocked the racing car to make the comparison "fair". Fact is the F1 car is designed to rev to 19000 rpm and do 250 mph - which is what makes it what it is.

R520 is designed to be clock faster, not to run at a lower speed. I'm sure if Nvidia could make the G70 run faster, with it's extra pipes, then they would. Hamstringing the part capable of higher speed down to the limit of the lower speed part is not a useful or fair comparison. It's a totally artificial comparison that is never going to happen in the real world.

 

[Humus]

You might want to be careful with that analogy. Some people do consider the P4 an inferior architecture primarily because the (precieved) design goal is an increase in frequency (not necessarily in increase in speed). This probably would not be the case if the chips would run cooler and scale to higher frequencies.

Nite_Hawk

Some people consider the Athlon an inferior architecture too. The point is that one was made to run at high frequencies, but with lower IPC, while the other was made to run at higher IPC, but at lower clocks. In the end when you benchmark the products against each other you find that they have their strengths and weaknesses, but they both perform overall more or less equal. I wouldn't call any of the architectures failed or inferior. In both cases they made rational decisions, and achieved similar performance in the end.

Sure, it's interesting to know the technical reasons for why a chip performs the way it does, but direct comparisons like this between totally different chips has little value. I have not seen anyone do the same on the CPU side, so I'm a bit surprised someone found this useful to do with GPUs.

 

[DemoCoder]

It is a fair comparison to compare the R520 high clock part to an R4xx clocked similarly for the purposes of evaluating ATI's efficiency claims in their PR. It is also fair to compare efficiencies per clock between two cards if "efficiency per clock" is being used as a PR weapon. Basically, these comparisons are fair game if "efficiency" becomes a stick with which to beat one IHV over the head. Certainly, there is nothing wrong with experimental investigation of PR claims.

Humus, the problem with your analogy is that while true, the reality is, Intel found out that going for high clocks was an architectural dead end. In the end, they couldn't scale their performance any higher, and the architecture they used to achieve high clocks also sabotaged power consumption.

Back when the NV3x came out, some people were saying "It doesn't matter if they need to use a dustbuster and high clocks to achieve their performance, all that matters is the performance result", but it turned out to matter. Given the voltage issues with the R520, and the delays ot market, I'm not sure one can say that the "high clock" design doesn't matter.