NV35 Reviews
- Thread starter BenSkywalker
- Start date
Very impressed with the Doom3 Benchmark results over at [H]. Amazing that those kind of frame rates are possible with something as intensive as D3. As the article states, it wasn't that long ago that > 25fps in Unreal 1 was impressive 
Also nice to see the NV35 drivers defaulting to Quality, and nV suggesting reviewers really push the details up wherever they can. A marked difference to the NV30 launch, no? It would be a shame if, now they've got the hardware, they hack the drivers significantly - Could really be a black mark against an otherwise excellent start.
Finally, a very impressive O/C on the memory is possible, as shown over at nVnews. Whether that will be the same in the retail cards is an unknown, but great stuff nonetheless.
Also nice to see the NV35 drivers defaulting to Quality, and nV suggesting reviewers really push the details up wherever they can. A marked difference to the NV30 launch, no? It would be a shame if, now they've got the hardware, they hack the drivers significantly - Could really be a black mark against an otherwise excellent start.
Finally, a very impressive O/C on the memory is possible, as shown over at nVnews. Whether that will be the same in the retail cards is an unknown, but great stuff nonetheless.
BenSkywalker
Regular
Why do you say that? Given what we have seen from the NV3X core I have been figuring the Doom3 would be best case scenario for the boards relative to the competition. Their improper pixel pie layout seems perfectly suited for a game like Doom3.
Well, when Doom 3 is using an extension that is optimal for the architecture, and nVidia has been giving it their focus besides for drivers, it isn't surprising the NV35 leads, as that posts SarcasmMark rating indicates you probably were thinking, Russ. It also has some things going for it in addition to that...it is 8 pipelines for stencil ops, and it has lots more bandwidth. <- Why I think these results are valid and not like the 3dmark situation.
Nice "shootout" by HardOCP 
Rev.
Nice "shootout" by HardOCP
Rev.Joe DeFuria
Legend
RussSchultz said:
Not stange at all. Carmack is using specific NV extensions that probably almost no one else is using for nVidia hardware. The performance differences to me (assuming that fixed ATI Cat 3.4 drivers "only" have the same performance as the 3.2 drivers), are in the neighborhood of the core clock and memory bandwidth advantages of the 5900 vs. 9800.
In other words, the benchmarks show to me, how one needs to use native extensions on nVidia hardware if you want to get the performance that you "should" get, considering it's specifications relative to ATI.
It would be interesting to see the 9800 overclocked to 5900 speeds, to see the actual clok-for-clock comparisons.
At the request of nVidia?
Joe DeFuria
Legend
Heh...seem's there's an echo in here.
Its just odd that it gets a 25-70% lead over the 9800, when the clock speeds are only 18% different. The memory speeds are about 20% different.
I had in my mind that the NV3x was 'archictectually' equal to the R3xx series in special cases (z only fills), but not overall. Going by this assumption, the NV solution solutions should be, at best, beating the ATI solution by the difference in clock rate, but most likely lower since we're not always in the proper gear for the NV solution.
Perhaps there's more to this ultrashadow stuff than a fancy marketting name?
I had in my mind that the NV3x was 'archictectually' equal to the R3xx series in special cases (z only fills), but not overall. Going by this assumption, the NV solution solutions should be, at best, beating the ATI solution by the difference in clock rate, but most likely lower since we're not always in the proper gear for the NV solution.
Perhaps there's more to this ultrashadow stuff than a fancy marketting name?
Extremetech:
They seem to have been fooled into thinking that nVidia hasn't taken extreme liberties in 3dmark03 performance figures, and can only come up with the idea that the NV35's advantages versus the 9800 depends on more shader usage (i.e., "future games"), and not bandwidth demands. What a rosy picture for misinformation this presents as people hear from other sources how the NV35 is "2x as fast as the NV30 for shaders" and think that validates this conclusion.
Would have been nice if the GT2 and GT3 improvements wrt the NV30 were clarified, but I didn't even notice any mention of driver versions used for the cards. There was a notable vacuum on the image analysis front as well.
Notable quote, indicative of nVidia's successful misinformation: "...but if you're buying for the DX9 future, it appears that the NV35 gives you some advantage." <-This apparently based on (what are supposed to be) PS 2.0 results, and ignoring the GT 4 results.
They seem to have been fooled into thinking that nVidia hasn't taken extreme liberties in 3dmark03 performance figures, and can only come up with the idea that the NV35's advantages versus the 9800 depends on more shader usage (i.e., "future games"), and not bandwidth demands. What a rosy picture for misinformation this presents as people hear from other sources how the NV35 is "2x as fast as the NV30 for shaders" and think that validates this conclusion.
Would have been nice if the GT2 and GT3 improvements wrt the NV30 were clarified, but I didn't even notice any mention of driver versions used for the cards. There was a notable vacuum on the image analysis front as well.
Notable quote, indicative of nVidia's successful misinformation: "...but if you're buying for the DX9 future, it appears that the NV35 gives you some advantage." <-This apparently based on (what are supposed to be) PS 2.0 results, and ignoring the GT 4 results.
BenSkywalker
Regular
It worked at [ H ] too, the shader specific benches they ran were all pretty much twice as fast as the 5800U they benched a bit ago, roughly even up with the R9800Pro.
Joe DeFuria
Legend
RussSchultz said:
There's also a difference in memory (128 MB due to Cat 3.2 driver vs. 256 MB on the 5900,) That could come into play at the higher resolutions with AA, where those larger performance differences emerge. We also don't know about the actual precision being rendered by both cards...NV35 might be using fixed point rather than Fp16. (I don't think it's ever been officially stated what mode the NV30 path uses?)
That is more true about the NV30 than the NV35, IMO. the NV35 seems to have tweked / fixed the shader performance so that it's more in line with expectations, vs. the the NV30. I would say that "overall" the NV35 is a pretty even match (clock for clock) with the R350.
One other comment, that maybe [H] could clarify: can it be verified the actual aniso mode that was used by both cards? Was it using default modes in the driver panel settings, or were there filtering settings made within Doom3?
Overall, the NV35 looks like a substantially improved part over the NV30. Much better performance, better aniso (based on the reviews), and atleast "decent" AA. A couple of issues though:
1) I assume there is still no gamma correction in the AA implementation?
2) The fan appears to still slow down and speed up depending on if you are in 3D or not. This isn't a terribly good design given MacOSX's use of opengl in the desktop, and the impending longhorn release.
3) PS 2.0 still seems to be slower than the 9800pro given the results at [H], though much better than the NV30.
4) This is the top end $500 card. I'd like to see what the 9800pro 128MB competetor performs like (will it be the same clock as the $500 card?)
Nite_Hawk
1) I assume there is still no gamma correction in the AA implementation?
2) The fan appears to still slow down and speed up depending on if you are in 3D or not. This isn't a terribly good design given MacOSX's use of opengl in the desktop, and the impending longhorn release.
3) PS 2.0 still seems to be slower than the 9800pro given the results at [H], though much better than the NV30.
4) This is the top end $500 card. I'd like to see what the 9800pro 128MB competetor performs like (will it be the same clock as the $500 card?)
Nite_Hawk
http://www.hexus.net/review.php?review=554&page=7
===================================================
We're not exactly sure what's going on with the FX's antialiasing. It seems to completely blur the whole image. Sure, it gets rid of the stair effect at the bottom of the image, but at what cost? Looking at the whole picture, it seems as if there is a reasonable degree of blurring throughout. Quake 3, for example, doesn't show this kind of blurring with anti-aliasing used. We'll have to take a far closer look at the FX5900 Ultra's image quality over a variety of games. The limited time scope of this review didn't permit it. Anisotropic filtering appears good, antialiasing is something that needs looking at.
===================================================
http://www.hexus.net/review.php?review=554&page=21
UT 2003 Max Quality. (Pretty Comparable image quality)
1024x768
9800Pro = 130.45
5900U = 67.4
1280x960
9800Pro = 93.44
5900U = 42.85
anyone else see this?
===================================================
We're not exactly sure what's going on with the FX's antialiasing. It seems to completely blur the whole image. Sure, it gets rid of the stair effect at the bottom of the image, but at what cost? Looking at the whole picture, it seems as if there is a reasonable degree of blurring throughout. Quake 3, for example, doesn't show this kind of blurring with anti-aliasing used. We'll have to take a far closer look at the FX5900 Ultra's image quality over a variety of games. The limited time scope of this review didn't permit it. Anisotropic filtering appears good, antialiasing is something that needs looking at.
===================================================
http://www.hexus.net/review.php?review=554&page=21
UT 2003 Max Quality. (Pretty Comparable image quality)
1024x768
9800Pro = 130.45
5900U = 67.4
1280x960
9800Pro = 93.44
5900U = 42.85
anyone else see this?