NV35 - already working, over twice as fast as NV30?
- Thread starter PrzemKo
- Start date
This 250/500 MHz terminology in use for DDR still causes no end of confusion. To me, it seems clear that it was 250 MHz DDR... aka "500MHz" by convenient choice when a bigger number serves (i.e., "marketspeak").
Anyways, I can easily think of situations where Quake III is bandwidth limited ...namely, when bandwidth is lower. In such a situation, a true 8 x 1 fixed function pixel pipeline would have no trouble outperforming a 4 x 2 design with half the bandwidth with the characteristics exhibited. The details of the comparison seem to serve to obscure how mundane such an occurence is.
I'm more interested to see comparisons of shader performance, but that is mostly because I considered such fixed function performance increase a given, not because it is insignificent to consumers just because it is unsurprising.
I'm not sure that the nv35 is necessarily going to clock higher than the nv30, however...then again, I'm not of the opinion that it has to in order to be a much better part for consumers than the nv30, even if shader improvement shows no significant improvement per clock, though I hope the latter is not the case.
Anyways, I can easily think of situations where Quake III is bandwidth limited ...namely, when bandwidth is lower. In such a situation, a true 8 x 1 fixed function pixel pipeline would have no trouble outperforming a 4 x 2 design with half the bandwidth with the characteristics exhibited. The details of the comparison seem to serve to obscure how mundane such an occurence is.
I'm more interested to see comparisons of shader performance, but that is mostly because I considered such fixed function performance increase a given, not because it is insignificent to consumers just because it is unsurprising.
I'm not sure that the nv35 is necessarily going to clock higher than the nv30, however...then again, I'm not of the opinion that it has to in order to be a much better part for consumers than the nv30, even if shader improvement shows no significant improvement per clock, though I hope the latter is not the case.
Bigus Dickus
Regular
Were they running a map where only single texturing was used?
Hmm.....
Hmm.....
what is beyond me, is that no one has drawn THIS conclusion from the NV35 benchmarks yet.
I think for being clocked at 250/250 it's performance is still dissapointing.
I come to this conclusion based on a few facts.
1) They said 42.72 drivers in "performance" mode
-this means that it is running forced bilinear filtering rather than trilinear as shown by [H]'s review of those drivers.
-factoring in above, I'd say that if they were to enable trilinear than the 111 FPS would drop considerably to anywhere in the 45-60FPS range
-According to Anandtech (http://www.anandtech.com/video/showdoc.html?i=1779&p=16 the Radeon 9700 (275/270, not clocked TOO much higher than the 250/250 of the NV35) gets 90.2FPS with 4XAA/8XAF in quality mode.
Can someone tell me where my logic is off? Or is the Nv35 just some more smoke and mirrors that Nvidia is puffing in the eyes of consumers?
comments?
I think for being clocked at 250/250 it's performance is still dissapointing.
I come to this conclusion based on a few facts.
1) They said 42.72 drivers in "performance" mode
-this means that it is running forced bilinear filtering rather than trilinear as shown by [H]'s review of those drivers.
-factoring in above, I'd say that if they were to enable trilinear than the 111 FPS would drop considerably to anywhere in the 45-60FPS range
-According to Anandtech (http://www.anandtech.com/video/showdoc.html?i=1779&p=16 the Radeon 9700 (275/270, not clocked TOO much higher than the 250/250 of the NV35) gets 90.2FPS with 4XAA/8XAF in quality mode.
Can someone tell me where my logic is off? Or is the Nv35 just some more smoke and mirrors that Nvidia is puffing in the eyes of consumers?
comments?
Johnny Rotten
Regular
galperi1 said:
I dont think I've EVER seen trilinear eat up 50% of your framerate.
Johnny Rotten said:
Go look at [H] review of the 5600 Ultra when they switch from Aggressive (bilinear) to Application (Trilinear) with 4XAA and 8XAF enabled
Ut2003 | "Application" | "Aggressive"
1280 X 1024 |12.6FPS | 39.6FPS
1024 X 768 |19.4FPS | 66.6FPS
As you can see, with Trilinear enable, the 5600 Ultra (based off the NV30 core) gets roughly 30% of the original bilinear performance. Thus making my 50% look very optimistic
I think he reached that conclusion as they said:Xmas said:
Personally I don't see how you can have a "performance" driver which doesn't sacrifice something to achieve that performance, and with nVidia's tendancy to ditch IQ at the drop of a hat recently, galperi1's conclusions are not unreasonable imo..
They could set whatever they want in-game (and I believe they say max settings), but if they've chosen "balanced" in the drivers then they're not doing trilinear.
I don't know why they put that comment there. Maybe because it's one of those '3DMark optimized' drivers (it's really the only reason I can see). On the other hand, this driver supports 8xS and 16x antialiasing in OpenGL, which do not really fit the "performance" label.grrrpoop said:
Yes, it is very likely that they used 'balanced' setting (which, btw, is renamed to 'quality' in those new drivers). It's default, I think.
Maybe someone who has a GFFX card could test how much of a difference there is in Q3 between application and balanced?
Xmas said:
Let's look at some facts then:
1) Nvidia told review sites to use the 42.72 "performance" drivers for the FX family reviews
2) They also told review sites to benchmark AF situations using the "aggressive" which forces bilinear filtering
Why would they stray from their strategy thus far. Obviously using those settings will give them the BEST numbers (ala 111FPS)
That is how I came to that conclusion. They will use any means necessary to boost their FPS as shown in the drastic difference between agressive and balanced/application settings in [H]'s review of the FX family. And this is just another example.
I feel that the current state of the NV35 is sub-standard at this point in time. If I'm right about the driver settings and the 9700 NP can beat it clock for clock, you can almost guarantee with certainty the the 9800 will also.
"Quality", yeah rightXmas said:
Another attempt by nvidia is mislead the consumer and reviewers. The only "quality" mode nvidia has is the "Application" setting. Making "Quality" the default does a serious disservice to consumers: Fake trilinear and other shortcuts are the default!Plus, nvidia has yet to release a WHQL certified driver for the GeForce FX series cards, yet you can buy the GeForce FX 5800 in stores! So much for nvidia's vaunted driver quality.
And of course nvidia insists that websites use a "benchmark driver" (42.76) for GeForce FX reviews...nvidia has gone on and on about "cinematic rendering" yet they release drivers with inferior image quality by default and the products are extremely slow when doing anything remotely "cinematic".
HardOCP already did this: http://www.hardocp.com/article.html?art=NDQ0LDg=. Granted, this wasn't the 5800, but the results are still relevant. Note the huge gains with AA and AF.
For image quality analysis check out http://www.hardocp.com/article.html?art=NDQ0LDEw.
From HardOCP's conclusion:
Deflection
Newcomer
Some of you guys have short memories.
16 bit z vs. 32 bit z
16 bit vs 24 bit vs 32 bit fp
AA differences
now
ati's aniso not as good aniso on angles vs whatever Nvidia's doing on balanced mode
Both co's are known to cut corners on quality where it makes sense for performance. IMHO if you need to flip between screenshots 20 times to notice a difference the extra performance you gain from this corner cutting is well worth it.
16 bit z vs. 32 bit z
16 bit vs 24 bit vs 32 bit fp
AA differences
now
ati's aniso not as good aniso on angles vs whatever Nvidia's doing on balanced mode
Both co's are known to cut corners on quality where it makes sense for performance. IMHO if you need to flip between screenshots 20 times to notice a difference the extra performance you gain from this corner cutting is well worth it.
