ASUS V9950 GeForceFX 5900 Ultra 256MB

[Rys]

It's not often that a reviewer attempts to eat a graphics card. When he or she does, you should stop and stare. I did just that with ASUS' current range topper. Here's a snip.

With 256MB of memory, meaning chips on the back side of the PCB, ASUS have created the worlds first AGP8X copper sandwich. Not very tasty (I did try to eat it, the dentist appointment is next week) in the traditional sense, but in the world of consumer graphics and all that entails, very tasty indeed.

Check it out in full, here.

Rys

[Judas]

There are few things I'd like to discuss about the review, thanks.

Page 13:

So with excellent performance (yes, even in upcoming 'proper' DX9 class titles like HL2 and DooM3), ASUS have done a good job wrapping it up with as much copper as they could, ... .

HL2 is a complete disaster for Nvidia and Doom3 code is still in production and won't be out until early 2004 at least, which makes it impossible to ensure all codepaths are working properly. Carmack mentioned Nvidia hardware won't perform very well without lower quality Nvidia specific path is being used. Is there something I missed?

Page 13:

It might have slower shader performance than ATI's R350 in some circumstances, but unless *****ing about your card on forums is your favourite pastime, those that like to play their games instead of talk about them will enjoy an FX 5900

I can't see how this judgement might be connected to Asus V9950's review. The reviewer should use the conclusion page to summarize his thoughts about the card, not the readers. The fact that Hexus.net's own forum software censors words from the review should tell something.

Page 13: Hmm, hard to recommend when you're forced to take a £400 hit in the wallet. But, 256MB of memory does make a difference on NV35 boards when pushed, ... .

How? There's no argument to back this up in benchmarks. I can't see any benchmarks done with 128MB card working at same frequencies so that this conclusion may be drawn.

Page 10:

Gun Metal 2 is a new benchmark at HEXUS, testing PixelShader 1.1 and VertexShader 2.0 from the DirectX 9 specification

DirectX 9 specifications are PS2.0 and VS2.0 not PS1.1 and VS2.0 Gun Metal uses only portions of DX9, if the reviewer felt the need to mention the VS2.0 was from DX9, the fact that PS1.1 is from DX8 should have been mentioned not to misguide uninformed readers.

Page 9:

Given an accepted baseline of driver quality from ATI just now, the ASUS draws UT2003 just fine.

IQ tests without screenshots are like performance tests without numbers.

Page 9:

What's nice to see, from an "is this a good driver?" perspective, is almost the same line gradient from each card, indicating similar methods of IQ being applied on both sets of hardware.

I can't see why the reviewer doesn't mention the bilinear filtering "bug" in Detonators, and what has the performance drop gradient has to do with image quality comparison anyway? This is something done with high quality images, and they aren't provided in review. The reviewer also failed to notice performance quality differences in absence of trilinear filtering which makes his observations about image quality very questionable.

It's certainly not earth shattering, but a win nonetheless. It's nice to see some parity in performance at the end of testing, in a real world DX9 game engine, to restore some waning faith in NVIDIA.

Is that so? Atrocious image quality of Detonator 50 drivers used in review are proven by THG with screenshots in Aquamark3 benchmark. 45.23 drivers already introduced image quality lowering cheats. What did the reviewer do to ensure no image quality degrading cheats and/or no shader replacement instructions were executed to claim waning faith in Nvidia is being restored ?

[Rys]

Originally posted by Judas
There are few things I'd like to discuss about the review, thanks.

Thanks for taking the time to do so

Originally posted by Judas
HL2 is a complete disaster for Nvidia and Doom3 code is still in production and won't be out until early 2004 at least, which makes it impossible to ensure all codepaths are working properly. Carmack mentioned Nvidia hardware won't perform very well without lower quality Nvidia specific path is being used. Is there something I missed?

Complete disaster? Can't really see that happening and I'm assured it will run just fine on NVIDIA hardware. It's too early to make a conclusion, nobody can run the game yet, but I base the statement on confidence that NVIDIA can sort out the problems. Doom3 comments are even less relevant, that's even further away.

Originally posted by Judas
I can't see how this judgement might be connected to Asus V9950's review. The reviewer should use the conclusion page to summarize his thoughts about the card, not the readers. The fact that Hexus.net's own forum software censors words from the review should tell something.

The forum software is over zealous, the word isn't inappropriate to use, since bitching is exactly what goes on I wasn't commenting on HEXUS readers either, rather forum users with a vendetta. The statement makes that clear.

Originally posted by Judas
How? There's no argument to back this up in benchmarks. I can't see any benchmarks done with 128MB card working at same frequencies so that this conclusion may be drawn.

We've covered 128MB vs 256MB FX5900 performance at HEXUS many times before, the statement is correct. Go look at other reviews for that. Even versus the Radeon on the graphs, you can see the advantage of 256MB on an FX5900, it generally has more performance at the 1600x1200x32 4AA 8AF point, where the memory is being used the most.

Originally posted by Judas
DirectX 9 specifications are PS2.0 and VS2.0 not PS1.1 and VS2.0 Gun Metal uses only portions of DX9, if the reviewer felt the need to mention the VS2.0 was from DX9, the fact that PS1.1 is from DX8 should have been mentioned not to misguide uninformed readers.

Good point, I'll correct the article

Originally posted by Judas
IQ tests without screenshots are like performance tests without numbers.

Not really. The reader can rely on the reviewer to give them accurate information without empirical evidence. We usally don't have the time to compare IQ explicitly, but I'll make the effort to do so in the future. To my eyes, 45.23 is a nice driver set in terms of IQ.

Originally posted by Judas
I can't see why the reviewer doesn't mention the bilinear filtering "bug" in Detonators, and what has the performance drop gradient has to do with image quality comparison anyway? This is something done with high quality images, and they aren't provided in review. The reviewer also failed to notice performance quality differences in absence of trilinear filtering which makes his observations about image quality very questionable.

See above, we generally have no time to explicitly cover IQ. The bilinear bug isn't driver wide, it's applied on some apps and tests. And the fact I didn't notice it suggests that it had little effect on IQ on the tests I ran, otherwise I'd definitely have said so. I mention the driver drawing incorrectly in Comanche 4, why wouldn't I do so in other tests I notice being drawn badly?

Originally posted by Judas
Is that so? Atrocious image quality of Detonator 50 drivers used in review are proven by THG with screenshots in Aquamark3 benchmark. 45.23 drivers already introduced image quality lowering cheats. What did the reviewer do to ensure no image quality degrading cheats and/or no shader replacement instructions were executed to claim waning faith in Nvidia is being restored ?

45.23 on the whole is a good driver for IQ. No driver is perfect, I hope you understand that. ATI's drivers have IQ bugs, and unless everyone tests screenshots against the refrast all the time, it's not pertinent to tar an entire driver with a "bad IQ" brush, since that's clearly not the case.

Thanks for your comments, at least you took the time to read everything, many don't The very state of drivers, industry wide, makes performance evaluation of cards a risky business these days. I stand by all my conclusions and statements however, despite that.

Cheers,

Rys