R600vsG80 Benchmark Comparisons From Any Game Welcome!

[sammy sung]

Yea maxblade i know,for some reason there's often bigger differences between the g80 and r600 in some rewiews than it is in real life.In this case i know from previous benches that Obie overclocks his card quite a bit and therefore is on par with the gtx in this game.I have no clue as to why the 2900 performs so badly in firingsquads rewiew,it is DX10 however so its probably a driver issue.

[mtl]

X-Bit is notoriously pro ATI in their attitude and language to the point that even if results are accurate I find them annoying. Many times they write if 2900 loses its a driver issue and if Nvidia loses its because ATI's better, etc, etc.

[sammy sung]

X-Bit is notoriously pro ATI in their attitude and language to the point that even if results are accurate I find them annoying. Many times they write if 2900 loses its a driver issue and if Nvidia loses its because ATI's better, etc, etc.

Hehe yea i noticed that too ,as it is the case with Lost planet that was developed on nvidia hardware its well known that call of juarez is atis baby sort of and that its developers have sofar been unwilling to optimize for nvidia hardware.So its a definate driver issue there with the G80 cards but they dont even mention it,theyre a bit too biased to be credible imo.

[Xion X2]

Hehe yea i noticed that too ,as it is the case with Lost planet that was developed on nvidia hardware its well known that call of juarez is atis baby sort of and that its developers have sofar been unwilling to optimize for nvidia hardware.So its a definate driver issue there with the G80 cards but they dont even mention it,theyre a bit too biased to be credible imo.
Well, that's not entirely true.

Call of Juarez DX9 was a TWINMBP(sp?) title and tied strongly w/ Nvidia developer relations. The developers decided to go a different route w/ the DX10 version by using shader-resolve AA (like R600 does by default) which is a mandatory DX10.1 spec, and this hurt G80 a bit. G80 has dedicated ROPs on the back-end to handle AA and doesn't perform as well using shader-resolve.

This pissed Nvidia off quite a bit, and Nvidia fans started shouting from the rooftops that Techland was favoring ATI products on purpose, but no one's been able to prove it. Shader-resolve AA is said to increase IQ when used in combination with certain HDR effects--so that right there is plenty reason to attempt it since COJ has some really nice HDR in it. And I don't think the performance hit is necessarily something that Nvidia could turn around with driver optimizations; the card is optimized more for the ROPs to handle AA, not the shaders. That's why it's performing so much better w/ AA enabled on current-gen titles. No games to my knowledge, except CoJ DX10, are currently using shader-resolve; all of them are optimized to use dedicated ROPs which G80 has and R600 doesn't.

[Obie]

The unreal 3 engine uses deferred rendering (http://www.pcgameshardware.de/?article_id=602522).

[sammy sung]

Well, that's not entirely true.

Call of Juarez DX9 was a TWINMBP(?) title and tied strongly w/ Nvidia developer relations. The developers decided to go a different route w/ the DX10 version by using shader-resolve AA (like R600 does by default) which is a mandatory DX10.1 spec, and this hurt G80 a bit. G80 has dedicated ROPs on the back-end to handle AA and doesn't perform as well using shader-resolve.

This pissed Nvidia off quite a bit, and Nvidia fans started shouting from the rooftops that Techland was favoring ATI products on purpose, but no one's been able to prove it. Shader-resolve AA is said to increase IQ when used in combination with certain HDR effects--so that right there is plenty reason to attempt it since COJ has some really nice HDR in it. And I don't think the performance hit is necessarily something that Nvidia could turn around with driver optimizations; the card is optimized more for the ROPs to handle AA, not the shaders. That's why it's performing so much better w/ AA enabled on current-gen titles. No games to my knowledge, except CoJ DX10, are currently using shader-resolve; all of them are optimized to use dedicated ROPs which G80 has and R600 doesn't.

Well looking at x-bit's benches COJ benches like crap on all hardware so they took a wrong turn somewhere ,25 - 40 fps sucks regardless wich hardware its running on.Saw a documentary about Bioshock today and the us-based developers raved about all the performance optimizations their aussie branch had done to the code,enabling them to cram more detail in there without having to sacrifice so much fps.I hope that is a sign of things to come,that the UE3 engine wich so many games will use will allow a nice level of detail and still be smooth.

[Xion X2]

Well looking at x-bit's benches COJ benches like crap on all hardware so they took a wrong turn somewhere ,25 - 40 fps sucks regardless wich hardware its running on.Saw a documentary about Bioshock today and the us-based developers raved about all the performance optimizations their aussie branch had done to the code,enabling them to cram more detail in there without having to sacrifice so much fps.I hope that is a sign of things to come,that the UE3 engine wich so many games will use will allow a nice level of detail and still be smooth.
Yeah, but most DirectX10 games to this point run like garbage on all current hardware. The problem's definitely not exclusive to CoJ.

There's a much smaller difference in visuals between BioShock DX9/10 compared to Call of Juarez DX9/10, also. The textures and lighting in Call of Juarez DX10 are completely re-worked, along with the water. The only thing noticeably different in BioShock are the smoke/fog edges and sharpness of shadows.

http://www.firingsquad.com/hardware/bioshock_directx10_performance/page4.asp

I just migrated to Vista this past week and am going to run a series of benches shortly with a bunch of games from my library.

[Redeemed]

Yeah, but all DirectX10 games to this point run like garbage on all current hardware. The problem's definitely not exclusive to CoJ.

There's a much smaller difference in visuals between BioShock DX9/10 and Call of Juarez DX9/10, also. The textures and lighting in Call of Juarez DX10 are completely re-worked, along with the water. The only thing noticeably different in BioShock are the smoke/fog edges and sharpness of shadows.

http://www.firingsquad.com/hardware/bioshock_directx10_performance/page4.asp

I just migrated to Vista this past week and am going to run a series of benches shortly with a bunch of games from my library.

Wow- haven't seen you post in a while... then again I haven't been in these threads for a while. :p

I haven't gone back to re-read this thread... so forgive me if you've already answered this but... how well does your 2900 run WiC (the SP Demo)?

[Xion X2]

Wow- haven't seen you post in a while... then again I haven't been in these threads for a while. :p

I haven't gone back to re-read this thread... so forgive me if you've already answered this but... how well does your 2900 run WiC (the SP Demo)?

I haven't tried it yet, but I may download it and give it a whirl. I'm generally not that enthused about strategy games.

I have to say that Vista absolutely pwns XP, though. There is just no comparison.. I love it. :cool:

[Redeemed]

I haven't tried it yet, but I may download it and give it a whirl. I'm generally not that enthused about strategy games.

I have to say that Vista absolutely pwns XP, though. There is just no comparison.. I love it. :cool:

Okay, I'm asking about WiC simply cause it runs like crap on my rig... at best. :( Yet it seems I recall jAkUp and others with rigs not significantly better then mine playing it and claiming performance was awesome (in Vista as well).

Soo... I'm thinking it is just this Opty not being able to keep up... which then makes me wish I had a mobo for this E6600 I have sitting next to me. :p

Anyhow, just curious what you get in that game. If your fps is only slightly better than mine then I guess I wouldn't have too much to worry about.

As for Vista... I won't make a comment- I shall merely " ;) ". :p

[Razor1]

Well, that's not entirely true.

Call of Juarez DX9 was a TWINMBP(?) title and tied strongly w/ Nvidia developer relations. The developers decided to go a different route w/ the DX10 version by using shader-resolve AA (like R600 does by default) which is a mandatory DX10.1 spec, and this hurt G80 a bit. G80 has dedicated ROPs on the back-end to handle AA and doesn't perform as well using shader-resolve.

This pissed Nvidia off quite a bit, and Nvidia fans started shouting from the rooftops that Techland was favoring ATI products on purpose, but no one's been able to prove it. Shader-resolve AA is said to increase IQ when used in combination with certain HDR effects--so that right there is plenty reason to attempt it since COJ has some really nice HDR in it. And I don't think the performance hit is necessarily something that Nvidia could turn around with driver optimizations; the card is optimized more for the ROPs to handle AA, not the shaders. That's why it's performing so much better w/ AA enabled on current-gen titles. No games to my knowledge, except CoJ DX10, are currently using shader-resolve; all of them are optimized to use dedicated ROPs which G80 has and R600 doesn't.


Hmm its a Dx10 spec not a Dx10.1 spec, and its manditory that the hardware be capable of doing it, both the g80 and r600 can do it, Xbit labs review used beta Dx10 performance driver for the 2900, and old WHQL drivers (3 months old) for the 8800's, when newer WHQL (162 were available since a couple of weeks ago actually even before then). I could see where that was a little messed up.

Shader AA resolve will only look better if there was a custom filter, and since COJ doesn't really give options to go between the shader based AA and ROP based AA there is no way of descerning what its doing. Not a big deal anyways, the hit on both cards ends up similiar with the latest drivers from both camps. What does that say, neither of these cards really have any advantage over the other when doing this operation in the shader units in this patricular game.

What did I say about Bioshock? Still think I'm looking into a crystal ball, or maybe its a magic 8 ball, at least that would have better odds then a crystal ball ;)?

[Xion X2]

BioShock is a little concerning, but I really don't see any graphical features in that game that should be making R600 take such a hit. From what I can tell, and from what's been said in reviews, texture detail between DX10 and DX9 is relatively the same and similar to what was on Xbox 360. The major differences are in shadow sharpness and and fog/smoke edges which usually aren't on-screen most of the time. Considering it plays much better in DX9 than DX10, it leads me to believe drivers could be holding it back quite a bit.

It plays pretty good on my setup in DX10 but not great. I get minimums of 35fps and maximums around 75fps w/ everything maxed and my card @ 850/950 and CPU @ 3.6Ghz.

Razor, if all of your passionate claims over the last month or two about R600 start to come true, then trust me, I'll be the first in line to shake your hand and say "you were right." It's never been, nor will it ever be, just about taking a hardened stance w/ me just because I happen to own something. I'm just skeptical of those I don't know who claim they know exactly what's going to play out 3 to 6 months into the future.

I think if you were in my position you would probably feel the same way.

[Redeemed]

BioShock plays like a dream on my setup. Haven't run through it with FRAPS on though... might do that tonight. But I did have literally everything cranked as high as I could and never once did it become "unplayable" or even appear to hitch up or even stutter. Granted, I wasn't using AA (well, had it set to 8xAA in the driver CP but it didn't really seem to do anything) so that might explain the stellar performance but still... :p

[Razor1]

BioShock is a little concerning, but I really don't see any graphical features in that game that should be making R600 take such a hit. From what I can tell, and from what's been said in reviews, texture detail between DX10 and DX9 is relatively the same and similar to what was on Xbox 360. The major differences are in shadow sharpness and and fog/smoke edges which usually aren't on-screen most of the time. Considering it plays much better in DX9 than DX10, it leads me to believe drivers could be holding it back quite a bit.

It plays pretty good on my setup in DX10 but not great. I get minimums of 35fps and maximums around 75fps w/ everything maxed and my card @ 850/950 and CPU @ 3.6Ghz.

Razor, if all of your passionate claims over the last month or two about R600 start to come true, then trust me, I'll be the first in line to shake your hand and say "you were right." It's never been, nor will it ever be, just about taking a hardened stance w/ me just because I happen to own something. I'm just skeptical of those I don't know who claim they know exactly what's going to play out 3 to 6 months into the future.

I think if you were in my position you would probably feel the same way.

Well if I would feel the same way about someone that I didn't know either, but from my point of view, I've been right for the most part (yes I have been wrong too, but usually because I'm over optimistic about a product) the past 3 generations of graphics cards (hardware features wise, and possible performance upon launch), that on it self should tell you I'm not pulling stuff out of my ass ;).

Bioshock performance doesn't seem that concerning, the only thing is I do expect it to have improved performance over the x1950, thats a wierd anomoly, but seems to be more of an issue with a bug. Now the firingsquad benchmarks seem a bit low, for all cards, which is probably the level they are benching, wish they would giive us that info. Anyhow bioshock does put more stress on texture filtering them any game before it % wise to other games, I'm pretty sure it hits fillrates hard too because of the type renderer the UE3 engine is. The texture detail is quite a bit higher in this game then games of the past. I don't understand why ATi decided with the 16 TMU's, pretty pixels are made in two ways, first per pixel accuracy will give cinimatic like art and more ALU computation power for better lighting features. We need both, I mentioned this in another post don't remember if it was here or not, but lets take X-men 3, I know the guys that did Mystique's effects and some others in the X-men series, the textures they used were 1 4096x 2048 per model. Unreal and Crysis are getting there 2 2048x2048 textures add up to that, but of course an offline renderer can offer much better lighting, now what kind of lighting models can we do in realtime right now, definitly not what we can do in offline renders yet, maybe another 2 or 3 gens before we can do that. The r600 has 35% more shader power (and that is no where near enough to do GI, it looks nice in a demo but games today and in the future will always be texture based to most degree, it gives artists the flexibility to make what they want, if we were to go to raytracing, artwork will have to be done by programmers for the most part, which sh*t I don't want to see that) and lower in everything else compared to the the GTX, and about equal to the GTS. Well that kinda tells us why we see the r600 sometimes faster then the GTS and mostly equal to the GTS. Its nothing complex to understand, theoretically in some situations the 2900 should be faster, but most of the time, it will be equal. New games will push higher ALU needs but they also need more in the texturing department, texture sizes are also increasing.

[Xion X2]

Well if I would feel the same way about someone that I didn't know either, but from my point of view, I've been right for the most part (yes I have been wrong too, but usually because I'm over optimistic about a product) the past 3 generations of graphics cards (hardware features wise, and possible performance upon launch), that on it self should tell you I'm not pulling stuff out of my ass ;).

You're assuming that I've known you and spoken with you over the past 3 generations of cards, which I haven't. You have never came to my attention until the last month or two.

However, on the other hand, I have owned and dealt with ATI for years and they haven't let me down yet.

Which one would you trust? A company you've been doing business with for years with great success or some random person you just met on the internet who won't even disclose his identity to you? :)

Concerning texturing, if that turns out to be the case, then you'll see me upgrading to another card. If not an ATI solution which resolves the issue then perhaps an Nvidia one--well after launch when the first batch of beta-testers have gone through the bad products. I'm not yet of that mindset, however. I was expecting BioShock to give me a glimpse of what I can expect out of DirectX10 performance, but that isn't the case after finding out what little difference there is between DX9/DX10 versions of the game.

Crysis looks to be the first true glimpse of that. And mark it down--if performance on that game suffers badly on my current setup then I'll look elsewhere for a solution. That'd be the first sign, to me, that ATI may have misjudged this gen.

[sammy sung]

Ah i like this,you 2 are having a civil damnnear friendly convo without harsh words or anything :thumbsup: :p :)

Btw,i'm not gonna bench at default cpu speed anymore,this new mobo even overclocks and i wanna explore that tbh.It will do 3.2 ghz on default vcore but im settling for 3ghz.Running cool and stable,32c idle 44 c load with my ram at 800mhz 4-4-4-12 2t.

[fivefeet8]

The developers decided to go a different route w/ the DX10 version by using shader-resolve AA (like R600 does by default) which is a mandatory DX10.1 spec, and this hurt G80 a bit. G80 has dedicated ROPs on the back-end to handle AA and doesn't perform as well using shader-resolve.


The G80 actually handles shader AA resolve better than the r600 because they lose less performance from enabling it. From my understanding, shader AA resolve shouldn't be that much more demanding since it's only doing the resolve filter not the entire AA. There's something else that's pushing down AA performance on the R600 from what I've heard at beyond3d and elsewhere.


Shader-resolve AA is said to increase IQ when used in combination with certain HDR effects--so that right there is plenty reason to attempt it since COJ has some really nice HDR in it.

Since COJ doesn't give us the option to use either Shader based resolve or ROP resolve, it's hard to prove "increased IQ". In terms of meaningful differences anyways. Shader based AA resolve is there to alleviate compatibility problems with current ROP based AA resolve on certain types of rendering like FP HDR. But there are games now and upcoming that use FP HDR and work fine with regular ROP based AA resolve.


And I don't think the performance hit is necessarily something that Nvidia could turn around with driver optimizations; the card is optimized more for the ROPs to handle AA, not the shaders.


Dedicated hardware will always do what they do faster than using generalized hardware to do the same thing. It's not a matter of hardware being optimized for ROP AA resolve. In any case, the shaders can be optimized to do whatever be it AA resolve or pixel/vertex/geometry simply because it's their nature being programmable. There is no fundamentally different way either hardware(g8x/r6xx) is doing shader AA resolve in that one can't optimize their shaders for it while the other can.


No games to my knowledge, except CoJ DX10, are currently using shader-resolve; all of them are optimized to use dedicated ROPs which G80 has and R600 doesn't.

Shader Model 4 calls for shader based AA resolve to be mandatory, but it doesn't do away with ROP resolve as that's still an option. Directx10.1 also doesn't. We're not likely going to see many games use only shader AA resolve like COJ. Atleast not until dedicated ROP based AA resolve is done away with in a future API spec. Not having dedicated hardware to do AA resolve only hurts the R600 because of this.

[Xion X2]

Since COJ doesn't give us the option to use either Shader based resolve or ROP resolve, it's hard to prove "increased IQ". In terms of meaningful differences anyways. Shader based AA resolve is there to alleviate compatibility problems with current ROP based AA resolve on certain types of rendering like HDR.

I think that "compatibility problems" would probably fall under "increased IQ" would it not? That was exactly what I had in mind when I made that statement.

Dedicated hardware will always do what they do faster than using generalized hardware to do the same thing. It's not a matter of hardware being optimized for ROP AA resolve.

You're splitting hairs here and taking a defensive stance which, honestly, I'm starting to tire of. Lately it seems you're quick to twist any statement I make around into something negative said about Nvidia cards, when it really isn't.

The only thing I was trying to say there was that G80 runs faster when its ROPs are handling the AA than when it's running shader resolve. As you said yourself, dedicated hardware always works better than generalized hardware for such a thing.

In any case, the shaders can be optimized to do whatever be it AA resolve or pixel/vertex/geometry simply because it's their nature being programmable. There is no fundamentally different way either hardware(g8x/r6xx) is doing shader AA resolve in that one can't optimize their shaders for it while the other can.

Never said anything of the sort. But I do think, and I implied this earlier with regards to the ROP/shader resolve discussion, that forcing G80 to do shader resolve brings it more down to the level of R600 when speaking in terms of overall performance.

This is an interesting perspective to view from, because if shader resolve becomes more popular in the future then it could decrease the sizeable gap we see in performance in some games when AA is enabled and thus bring the two cards much closer.

[fivefeet8]

I think that "compatibility problems" would probably fall under "increased IQ" would it not? That was exactly what I had in mind when I made that statement.

Not neccessarily. Supersampling AA is more compabitible with more forms of rendering than MSAA, but does it neccessarily give better IQ? It all depends on how the developer uses it and how it compares to ROP AA resolve in terms of edge anti-aliasing and performance.


Never said anything of the sort.



And I don't think the performance hit is necessarily something that Nvidia could turn around with driver optimizations; the card is optimized more for the ROPs to handle AA, not the shaders.


You said that Nvidia can't optimized their performance with shader AA resolve through drivers. I said since shaders are programmable, they can. I also explained why because shader AA resolve is done fundamentally the same on both cards and both can optimize drivers thereby optimizing the shaders for it. It's not because G8x cards are optimized for ROP AA resolve, it's simply because they have dedicated hardware for it.

[Xion X2]

You said that Nvidia can't optimized their performance with shader AA resolve through drivers.

No, I didn't. Since you recently coined the phrase "not necessarily," then I'm guessing you know what it means. So please go back and see where I used the term in reference to the theory of sammy sung's that driver optimization would completely turn things around with the performance hit G80 took by going from the ROPs to the shaders in handling AA.

I never said they "can't optimize it." My reply had to do with a lot more than just optimization; it had to do with the magnitude of performance increase one would see from that "optimization."

[fivefeet8]

So please go back and see where I used the term in reference to the theory of sammy sung's that driver optimization would completely turn things around with the performance hit G80 took by going from the ROPs to the shaders in handling AA.

I believe Sammy's post was more general than specific to shader AA resolve. There were more than only moving to Shader AA resolve in COJ dx10. Performance actually decreased on G8x cards after the patch even without using AA.


I never said they "can't optimize it." My reply had to do with a lot more than just optimization; it had to do with the magnitude of performance increase one would see from that "optimization."

Ah, I see. But then the r600 also faces a similar situation since both seem to have a hard time running shader AA resolve.

[Xion X2]

I believe Sammy's post was more general than specific to shader AA resolve.

From his perspective it probably was. I'm not sure if he's aware of the technicalities of the issue and why CoJ in DX10 is running equal or a little better on 2900XT right now. That's why I was willing to explain it and bring to light that it may be a hardware-specific issue.

You and I have both had this discussion before, I believe, and know that the AA path Techland has chosen is affecting things.

Performance actually decreased on G8x cards after the patch even without using AA.

Really? I wasn't aware of this. Haven't heard it discussed at all, actually.

Ah, I see. But then the r600 also faces a similar situation since both seem to have a hard time running shader AA resolve.

We have not been comparing shader-resolve vs shader-resolve, however. In most games, G80 has had an outlier w/ its ROPs handling the AA which is a big reason it's coming out on top in most of the benchmarks. If it no longer has that outlier in specific titles in the future, then the performance gap between the cards, at least respective to what we've seen with CoJ, is destined to close quite a bit. And we've already established that it makes sense from a hardware perspective.

[Razor1]

Really? I wasn't aware of this. Haven't heard it discussed at all, actually.

Yeah think it was specific to the parrallex mapping shader too


We have not been comparing shader-resolve vs shader-resolve, however. In most games, G80 has had an outlier w/ its ROPs handling the AA which is a big reason it's coming out on top in most of the benchmarks. If it no longer has that outlier in specific titles in the future, then the performance gap between the cards, at least respective to what we've seen with CoJ, is destined to close quite a bit. And we've already established that it makes sense from a hardware perspective

It doesn't seem this generation hardware will be do shader based AA with any good performance numbers, most likely there is a latancy involved that was overlooked or missed for this generation. Maybe it takes longer for the data to get the ROPS to do the final output. Just a guess but its possible. If AA in the future gen cards are going to take this kind of hit 40%~ thats just too much of hit to go for marginally improved image quality.

[fivefeet8]

Really? I wasn't aware of this. Haven't heard it discussed at all, actually.

It's been discussed, but not at much length. The COJ dx10 benchmark is not that comparative to any Dx10 games out today nor the near future. It's use of shader AA resolve leaves much to be desired from a performance standpoint. The superior IQ that it's suppose to be getting from it is hard to prove as no other games uses it and it doesn't have an option for comparison.


We have not been comparing shader-resolve vs shader-resolve, however.


Even if we were able to, we can't because the R6xx lacks dedicated hardware for a point of comparison, the outcome would still be the same from a driver optimization standpoint.


And we've already established that it makes sense from a hardware perspective.

It makes more sense to have both dedicated and shader AA resolve in hardware right now.

[Xion X2]

Even if we were able to, we can't because the R6xx lacks dedicated hardware for a point of comparison, the outcome would still be the same from a driver optimization standpoint.

You're going off on one of your usual tangents here. Please try and stay on topic or at least half-way follow what I'm saying.

What I am saying is that none of these benchmarks have been comparing two cards handling AA resolve in the shaders. G80 has not had to do it in any game but CoJ in DirectX10, so it has had the advantage of dedicated hardware this entire time while R600 hasn't.

Going back to what started this line of discussion, your comment:
But then the r600 also faces a similar situation since both seem to have a hard time running shader AA resolve.

You're acting like this "similar situation" is something new for R600 when it's not. It's been handling AA in the shaders the entire time. We know exactly how it performs when doing so, and if CoJ is any indication, we have a small glimpse of projected performance difference that we can expect if G80 is forced to do it, too.

We certainly know, at least, that the significant AA advantage that G80 holds would be seriously diminished.

It makes more sense to have both dedicated and shader AA resolve in hardware right now.

And here's another knee-jerk reaction. That is not what I was claiming[that lack of ROPs makes sense from a design perspective], and you should know this by now because we just discussed it a few posts ago. You and I had already agreed upon dedicated hardware being more effective at AA than generalizing it in the shaders just a few posts back.

My comment had to do with our line of discussion earlier about that. Thus, if developers decide to go the route of shader-resolve for HDR compatability reasons, G80s ROPs become effectively useless and it's brought down to the AA performance level of R600.

Sheesh you are one defensive cat. It is truly difficult to carry on a civil discussion with you.