G70 Vs X1800 Efficiency

[Arun]

Well, I'd say this COULD be a fair comparaison, depending on how high NVIDIA manages to clock the G7x on 90nm. If it's roughly identical, then this is a relevant architectural discussion. If it isn't, then it isn't as OpenGL Guy/Humus stated, since all it proves is that ATI's architecture compensates for that with an architecture that is more finely tuned towards higher clockspeeds. Of course, you'd ideally have to take margins into consideration too, but then it gets a fair bit harder to evaluate properly.

Uttar

 

[OpenGL guy]

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.

When R5xx was designed, how could it have been designed to be more efficient than a product that didn't exist (i.e. the G70)? I thought it was clear that the R520 was designed to be more efficient than previous ATI parts.

Edit: Note that I am not saying that R520 is more or less efficient than G70, I am merely stating that when it was designed ATI could not have targetted G70 efficiency.

 

[Humus]

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.

Sure, but comparing to G70 isn't particularly useful.

 

[Mintmaster]

Sure series X1800 are a looser all the way.

Not sure if you're being sarcastic, but I don't think so at all.

For Doom3, naturally, due to NV's double-Z and interference with ATI's HiZ.

For CS:VST, the difference is pretty much what NVidia gained in its most recent drivers. Clock for clock and pipe for pipe, I don't think ATI ever had an advantage in CS:VST. However, it is odd that NVidia only loses ~20% when disabling a third of its pipes.

The shadermark stuff surprises me the most. Looks like there is some sort of cap on the first 16 shaders, because no other site shows such uniform results. However, there are no worries for the X1800XT here.

It would have been more interesting if they:
a) Tested more current titles with advanced shaders, like FarCry or Splinter Cell: Chaos Theory
b) Threw in the X850XT and 6800U

One thing that interests me, however, is that ATI could have stuck closer to the R300 architecture one more time, and with 320M transistors could have made a pixel shading monster of a chip. But I guess that's happening with R580 anyway, if the rumours are correct...

 

[trinibwoy]

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.

You're ignoring the fact that they disabled 8 pipes on the G70 as well? One thing that I missed earlier and that people seem to still be missing is that the only difference between these tests and reality is the memory bandwidth being equalled out.

Retail:
GTX: 3.2% fillrate advantage
XT: 25% bandwidth advantage

DH Tests:
GTX: 0% fillrate advantage
XT: 0% bandwidth advantage

 

[CMAN]

Could the R520 efficiency improvements only kick in at certain frequencies (ie: it could scale a lot better than a G70)? We need someone to test the G70 at higher frequencies to test that out. In addition, running the core and memory at different ratios than stock could also negate optimizations made with the ratios in mind. (At least I think that could)

 

[hughJ]

there's like a 25% difference in (3dmark) single texture fillrate between r520->g70?

(edit: in the DH tests.. 3432.2 MTexels/s vs 4288.1 MTexels/s)

 

[CMAN]

You're ignoring the fact that they disabled 8 pipes on the G70 as well? One thing that I missed earlier and that people seem to still be missing is that the only difference between these tests and reality is the memory bandwidth being equalled out.

Retail:
GTX: 3.2% fillrate advantage
XT: 25% bandwidth advantage

DH Tests:
GTX: 0% fillrate advantage
XT: 0% bandwidth advantage

Has anyone been able to overclock the memory on the GTX to 1500 MHz? In addition, we need Nvidia to release a card with 512 MB RAM to compare to the XT.

 

[pino]

You're ignoring the fact that they disabled 8 pipes on the G70 as well? One thing that I missed earlier and that people seem to still be missing is that the only difference between these tests and reality is the memory bandwidth being equalled out.

Retail:
GTX: 3.2% fillrate advantage
XT: 25% bandwidth advantage

DH Tests:
GTX: 0% fillrate advantage
XT: 0% bandwidth advantage

i agree, i think it could be just more interesting to see a comparison with only same memory speed and using retail clocks and pipes settings.

 

[^eMpTy^]

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?

Basically what makes this interesting is the advantage ATi currently has is the 90nm process these chips are built on. G70 will do 500Mhz on 110nm and will likely push towards 600Mhz when they switch to 90nm next year.

So basically once both companies are at 24 pipes and 90nm, the more efficient design will win...and right now that's G70.

Also, G70 can compete with the X1800XT while running at much lower clocks, much lower temperatures, and using a single slot cooler. In my opinion, that's a superior design for the same reason the Athlons are better than the P4's, they produce the same performance using less power and generating less heat.

 

[Bouncing Zabaglione Bros.]

You're ignoring the fact that they disabled 8 pipes on the G70 as well? One thing that I missed earlier and that people seem to still be missing is that the only difference between these tests and reality is the memory bandwidth being equalled out.

But so what? The R520 can and does run at a higher clock where the G70 can't. Even if you accept the G70 is more efficient, R520 makes that up with it's extra speed - speed the G70 can't match. Taking away the speed advantage of one architecture to compare to another that can never equal that speed is heavily biased.

The cards are what they are meant to be. Comparing them by changing what one of them actually is does not make a valid or useful comparison in the real world.

You might as well upclock the G70 using nitrogen cooling and claim it's a useful metric when compared to a standard R520.

 

[DemoCoder]

Wasn't OPPAINTER or some friends of his able to run the G70 air-cooled at 500-550Mhz? I remember seeing posts to this effect the night before the G70 launch. I think there is definately some headroom in the G70, so I would not neccessarily say that it is not able to achieve higher clocks (albeit probably not as high as R520). For an apples to apples comparison, we'll need a G70 on 90nm.

 

[scificube]

However, it is odd that NVidia only loses ~20% when disabling a third of its pipes.

You may want to consider the core clock was increased by 30MHz to 450Mhz.

Also when thinking about clocks one may want to think about RSX. Nvidia claims the part will be clocked at 550Mhz and most assume the architecture is based off the G70 while some go farther in thinking it's G70 with some icing. If this is true there are some things one could consider. When G70 is moved to the 90nm process clocks in the 500Mhz+ range do not seem out of the question. Also given the architecture's good IPC it should be a very good performer at such speeds. The good IPC of the architecture (altough I based this off of NV40 vs. R420) tied with an uncommonly high clock speed for Nvidia HW is one the reasons I have tried to avoid underestimating RSX.

Anyway...if Nvidia can succeed in getting G70 up to these speeds it would seem it too could make a monster shader part with more pipes or more ALU's per pipe to compete with R580. Then again I haven't kept up much so I've no idea if Nvidia has a different architecture completely lined up to duke it out with R580.

 

[trinibwoy]

But so what? The R520 can and does run at a higher clock where the G70 can't. Even if you accept the G70 is more efficient, R520 makes that up with it's extra speed - speed the G70 can't match. Taking away the speed advantage of one architecture to compare to another that can never equal that speed is heavily biased.

What's your point? The G70 can't hit 600Mhz+ just like the R520 doesn't have more than 16 pipes. You're conveniently ignoring what I pointed out earlier, that the only siginificant difference in theoretical numbers between DH's setup and retail configurations is the memory bandwidth advantage of the XT.

 

[jb]

But when you mean "efficent" dont you want to put some bounds on that?

What about PS1.x performance?
What about PS2.x performance?
What about PS3.x performance?
What about AA performance?
What about AF performance?
What about HDR performance?
What about theoretical vrs realwold fill rate numbers?
What about vertex processing?

It would be very easy to cherry pick test to show one or the other was "more efficent" based on what you were testing... just food for thougth...

 

[colinisation]

Basically what makes this interesting is the advantage ATi currently has is the 90nm process these chips are built on. G70 will do 500Mhz on 110nm and will likely push towards 600Mhz when they switch to 90nm next year.

So basically once both companies are at 24 pipes and 90nm, the more efficient design will win...and right now that's G70.

Also, G70 can compete with the X1800XT while running at much lower clocks, much lower temperatures, and using a single slot cooler. In my opinion, that's a superior design for the same reason the Athlons are better than the P4's, they produce the same performance using less power and generating less heat.

I think you (vastly?) overstate the amount of clock speed headroom the nvidia has with G70. Was it not also the same Driverheaven which posted overclocking results for the R520 yesterday and if I recall correctly they were pretty good as well. SO ATI too may have some head room to release another product as well.

As for the validity of the article I believe it is quite a worhwhile endeavour from a technical point of view obviously it is of little use as a yard stick by which one bases his purchasing decisions. I was actually hoping a site would do something like this. (also interested in variations in performance under different memory speeds)

Aceshardware did a clock for clock comparison of the P4 and Athlon.

 

[PurplePigeon]

I'm probably just grumpy, stupid, or ignorant, but I don't like the way DH uses the term "efficiency".

In my mind, a high level of efficiency suggests that the resources of the GPU are not idle very much. However, it seems to me that DH is actually measuring how much work each GPU performs each cycle. Which is fine, but it isn't "efficiency".

 

[trinibwoy]

I think you (vastly?) overstate the amount of clock speed headroom the nvidia has with G70. Was it not also the same Driverheaven which posted overclocking results for the R520 yesterday and if I recall correctly they were pretty good as well. SO ATI too may have some head room to release another product as well.

That's all true. But don't forget to factor in the current power consumption and cooling configuration of the GTX. Nvidia has way more headroom in this regard. The XT in its current state is already bigger, hotter and louder than the GTX.

 

[jb]

That's all true. But don't forget to factor in the current power consumption and cooling configuration of the GTX. Nvidia has way more headroom in this regard. The XT in its current state is already bigger, hotter and louder than the GTX.

I would not be so sure on that..going to 90 will give them sum head room on yeilds and allow for faster clocks, but at faster clocks you get more heat and more power drain...

 

[Maintank]

I think you (vastly?) overstate the amount of clock speed headroom the nvidia has with G70

I think it is safe to say Nvidia may realize a 10% headroom with a process shrink to 90nm?
That pushes their stock card upto ~470Mhz with companies like EVGA getting near 540 Mhz if they hold to their same %s. Remember EVGA doesnt seem to have a problem shipping a 490Mhz air cooled G70 part right now.

In all reality that is a very good clockspeed for a 24 pipe part.
You have to remember 2 years ago Nvidia was shipping a 4 pipe 450Mhz part.