half life 2 benchmarks geforcefx/ati???

[Pent]

nVIDIA says that they are coding better drivers to work with half life 2, for a slight image degration, compared to radeon

[1000101]

you can run the game in dx8 mode on the nvidia card and it will really fly.
won't look as cool though.

Theres also the option of the detonator 50's; which thus far look like they kill alot of IQ for not much gains.
not sure how nvidia will solve a hardware fault through drivers though.

[richcz3]

Originally posted by Levitateme
I ran 3dmark a week ago. and on one game got amazing results. but then like the next 3 were terrible resutls...so im curious what nvidia is up to...if anything
____________________________________

What is nVidia up to?
That's a good way of putting it.
1000101 is right, the new Det 50's that are set to be released really cut down on visual quality to up the frame rates. A search on various HW sites show what the new drivers accomplish.

John Carmack chimes in
http://english.bonusweb.cz/interviews/carmackgfx.html

Unfortunately, these FX series of cards have been a bit of a thorn in nVidas side from the get go. Their next gen card is based on the same archtecture so nVidia is going to be hard pressed to engineer a remedy in hardware somehow.
It even looks like the NV38 is going to be as big as the original Dust Buster FX 5800
http://www.hwupgrade.it/articoli/908/index.html

For the moment ATI has it together in Hardware as well as Drivers. I never thought I would write that...ever.


richcz3

[Levitateme]

thanks richz, i like the idea of playing a game and it looking terrible. wow im really dissapointed in nvidia...the reason i got this card is because the specs on it looked so amazing. but i really have not seen it do anythign amazing yet...

[CgFX]

Don't take any of our words on this. Wait and see what happens.

The quote by Carmack discusses a reduction in precision, not necessarily in image quality. There is an enormous difference that escapes most of the people arguing about this. He mentions this has no affect on Doom3's image quality so we will have to wait and see what happens with HL2.

For example, the Radeon 9700/9800 have 1/256th the sub pixel precision of the GeForce FX and that has a direct impact on image quality for every frame rendered. You won't hear Valve talk about that thanks to their "$pecial relation$hip" with ATI.

Rich is also wrong about NVIDIA's next gen card if he is refering to NV40.

[Levitateme]

thanks cgFX, that is really promising, thanks to everyone... i guess next week ill find out for sure...thanks agian.

[1000101]

The quote by Carmack discusses a reduction in precision, not necessarily in image quality. There is an enormous difference that escapes most of the people arguing about this. He mentions this has no affect on Doom3's image quality so we will have to wait and see what happens with HL2.

The reason it never impacts doom is because he never goes beyond using mostly 12 bit integer precision and in a few places 16 bit FP.
Dx9 specs call for a minimum of 24 bit precision; which means any Dx9 games out there will have much more depth. In addition he also says that future dx9 titles will indeed have noticable IQ degredation.
And yes; Nvidia does support FP32; however it does so at about 1/4th (if not slower) the speed that ATI does 24 bit precision. This is a hardware fault and not a driver issue. The only way for the nvidia card to remain competitive is to use mixed mode (which will essentially drop most shaders off to fp16) Reviewers have allready noticed IQ differences in the released benchmark. I'm sure the final game will be no different.

For example, the Radeon 9700/9800 have 1/256th the sub pixel precision of the GeForce FX and that has a direct impact on image quality for every frame rendered. You won't hear Valve talk about that thanks to their "$pecial relation$hip" with ATI.

Never heard that before. Actually the R3XX series is every bit as precise as the FX (more-so when you factor in performance); it just truncates 128 bit values to 96 bits when it goes to store the date. This isn't anything to do with the special relationship with ATI; as more or less a special relationship with microsoft. If you notice; every title utilizing Dx9 performs much better on ATI hardware. This includes but is not limited to: 3dmark 2003; Half-life 2; Tomb Raider: Age of Darkness (or whatever its called) and several dx9 shader benchmarks.


Rich is also wrong about NVIDIA's next gen card if he is refering to NV40.

If you read Rich's post; he refers specifically to the card in question
:

It even looks like the NV38 is going to be as big as the original Dust Buster FX 5800

Though thats nto really a next gen architecture; so much as a respin of NV35.
Your right though; noone knows about NV40 and it could seriously be a slammin part with no equal..
of course
whats that do for all those NV35 owners out there? Oh wait; spend another 500 dollars on another video card!!

[elvis]

early indications of the 50 series of drivers due for nvidia cards are implying that various filtering techniques at certain levels is being removed, which gives a slight speed boost for a slight loss in image quality.

the annoying thing for most people is that at a user level the option is available to not use things like trilinear/anistropic filtering etc, but nvidia have felt the need to remove this from the user's control.

at any rate, doom3 is not DX9, and shouldn't really be used to compare nvidia to ati on a DX9 level. doom3 also has separate nvidia and ati paths, which blurs the lines a bit. at this stage half-life 2 will have DX8 and DX9 code paths. sites previewing the game have remarked that the geforceFX series works well under the DX8 path, but poorly in comparison to the radeon 9600 and up in the DX9 paths.

however spending megabucks on a DX9 card like the geforceFX series and having to play games in DX8 mode is rather frustrating. as CgFX said: only time will tell what the real story is. it's not looking good for nvidia at this stage for their geforceFX product, but nvidia being who they are will release a new product every 6-12 months anyway.

the biggest/fastest/best video card is always coming soon. :)

[CgFX]

Originally posted by 1000101
Never heard that before. Actually the R3XX series is every bit as precise as the FX (more-so when you factor in performance); it just truncates 128 bit values to 96 bits when it goes to store the date.
Just to be clear, I was discussing sub pixel precision and you changed that to talk about the rendering precision. The R3xx has 4 bits of sub pixel precision. The NV3x has 12 bits. This means that the R3xx is 256 times less precise when trying to determine if a pixel is on a surface is not. The result is a dramatic increase in rendering errors that show up as holes, cracks, dots, etc.

As for pipeline precision, NV3x is 128 bit FP, front to back and also supports a half-precision 64 bit FP mode. Obviously even the half-precision mode is orders of magnitude more precise than the 32 bit integer of the GeForce4.

As I understand the R3xx pipeline it has a max of 96 bit FP (which is 1/4294967296th the precision of NV3x) but there are various stages in the shader engine where that actually gets factored down to 24 bit FP (1/1099511627776th the precision of NV3x's half-precision mode). Based on that I don't understand how you can suggest that it is as precise or more than NV3x.

I should also note that I don't think the 9800 Pro is the right card for HL2 either. The benchmark is not suggesting that it is locked at 60 fps on 9800 pro (which is where all games should be played). It is suggesting that it averages 60 fps at 1024x768. Not where people will want to be wrt to resolution and performance. Especially when it slows to a crawl with a ton of things going on.

I think it is good but somewhat strage that everyone is expecting a brand new game with an important next generation graphics engine to be played with every last graphics featured cranked and at high resolutions on current cards. This has never been the case in the past. I am glad that it is not possible, otherwise it would mean that the software wasn't being written to push the envelope at all.

[richcz3]

Aside from HL2's release, there are so few DX9 titles coming out this or next year.
-
Much like the hoopla surrounding shader capabilities in hardware. The first GeForce 4 cards at the time were pitched as the best cards you would want to play Doom 3 with. So early adopters payed $400 for features that weren't in wide use yet.
The gaffe was when nVidia yanked the HW shaders out in the MX line to make them more affordable. Hapless consumers bought them in truckloads. But it doesn't matter now anyway as by the time Doom 3 comes out there will be a need to upgrade the hardware again.
http://www.hothardware.com/hh_files/S&V/gf4_mx_roundup.shtml
http://www.digit-life.com/articles/gf4mx/

Point is, game development takes 2 to 3 years minimum and accelerators come out ever 6-8 months and even longer now. HL2 is an exception to that rule. Unlike the Geforce4 MX line. People who own DX9 cards by nVidia have to deal with the bad DX9 press now instead of later.

By the time DX9 games are in force on the market, rest assured that nVidia will have made their cards to spec.


richcz3

[CgFX]

Originally posted by richcz3
The gaffe was when nVidia yanked the HW shaders out in the MX line to make them more affordable. Hapless consumers bought them in truckloads.
That shows a lack of understanding on how things work and what nVidia actually did. GF4 MX is based on NV17 which was a strong evolution of NV15 (GF2). There was never programmability in there to yank.

And why pick on NVIDIA for this? During the same period ATI was desperately trying to sell its DX7-class cards and even went as far as renaming the same card as something new (even more weasely in my book). What exactly is the difference between the Radeon and the Radeon 7200?

NV got such flak for the GF4 MX naming. Where are all those people when it comes to the Radeon 9000 and 9200's??? I thought that first number was supposed to the the DX class?

[richcz3]

CgFX, that's good for hardware buffs to be able to differentiate, but the general consumer sees "GeForce4 MX", not "GeForce 2 SE". Thats sleight of hand marketing isn't it. I thought I was being nice for saying yanking out.

Here’s what John Carmack (id Software) had to say about the name confusion thing:

NVIDIA has really made a mess of the naming conventions here. I always thought it was bad enough that GF2 was just a speed bumped GF1, while GF3 had significant architectural improvements over GF2. I expected GF4 to be the speed bumped GF3, but calling the NV17 GF4-MX really sucks.
http://www.techspot.com/reviews/hardware/evga_gf4mx440/index.shtml

Oh but what about the current 9600 ATI nonsense. ATI is no angel. The 9600 is a hardware downgrade to the 9500 series.
I refuse to buy a 9600.

For the record I had been buying nVidia sine the TNT line. Even when 3DFX was on the ropes I bought a 5500. For nVidia there was solid measurable perfomance in each HW revision up until the GeForce 4 line.

Do I have a beef with nVidia. Heck no. I have 4 systems based on the nForce2 MBs. And I intend to keep away from VIA as long as nVidia puts out true performance advantages over VIA. I am looking forward to an nForce 3 system. :love:
Should nVidia stall or stumble, I will read and research and make my decisions accordingly. Not loyalty.


richcz3

[1000101]

Just to be clear, I was discussing sub pixel precision and you changed that to talk about the rendering precision. The R3xx has 4 bits of sub pixel precision. The NV3x has 12 bits. This means that the R3xx is 256 times less precise when trying to determine if a pixel is on a surface is not. The result is a dramatic increase in rendering errors that show up as holes, cracks, dots, etc.

I'll read some paperwork on this. Even if its true I've never seen any such artifacts on surfaces in games or in maya. So it seems to be nvidia would be doing more than they have to here.

As for pipeline precision, NV3x is 128 bit FP, front to back and also supports a half-precision 64 bit FP mode. Obviously even the half-precision mode is orders of magnitude more precise than the 32 bit integer of the GeForce4.

Indeedy doo.

As I understand the R3xx pipeline it has a max of 96 bit FP (which is 1/4294967296th the precision of NV3x) but there are various stages in the shader engine where that actually gets factored down to 24 bit FP (1/1099511627776th the precision of NV3x's half-precision mode). Based on that I don't understand how you can suggest that it is as precise or more than NV3x.

Your failing to factor in a couple things. Firstly; though ATI's final output is a 96 bit FP value; everything is calculated internally at 128 bit. Also; ATI's 24 bit mode is 4 times faster(at least) than nvidia's 32 bit mode. In order to keep up nvidia must drop to FP16 and 12 bit integer which is indeed less precise than ATI. Don't just read nvidia's marketting; think of real world tests.

I should also note that I don't think the 9800 Pro is the right card for HL2 either. The benchmark is not suggesting that it is locked at 60 fps on 9800 pro (which is where all games should be played). It is suggesting that it averages 60 fps at 1024x768. Not where people will want to be wrt to resolution and performance. Especially when it slows to a crawl with a ton of things going on.

Okay; lets break this down. The benchmark shows the 9800 pro getting an average of about 60fps. At 1280x1024 it gets an average of about 55fps. I suppose at 800x600 it would probably get close to 65-70fps but not much higher. The problem here that slows the cards down is mostly the shaders that comprise a majority of the effects.
Just out of curiosity; what are you refering to that slows down when there is a ton of things going on? Also, what card doesn't? I don't think you know exactly what your talking about with that statement so I'll have to ask you to clarify what you mean with "Especially when it slows to a crawl with a ton of things going on."


I think it is good but somewhat strage that everyone is expecting a brand new game with an important next generation graphics engine to be played with every last graphics featured cranked and at high resolutions on current cards. This has never been the case in the past. I am glad that it is not possible, otherwise it would mean that the software wasn't being written to push the envelope at all.

I think you totally misunderstand peoples expectations. Firstly; based on the benchmarks shown; we can see an over 1 year old 9700 pro is going to perform quite well in HL2 at 1024x768 with all the goodies turned on (sans AF and AA) and manage an average of 50-60fps. We can also see that with all the goodies turned on that the brand spankin new 5900 ultra is half as fast at almost 100 dollars more than the competition.

There is plenty of other evidence to show that the 5900 ultra many people just bought is going to be an attrocious directX 9 performer. Its a fact of life. The hardware is incapable of providing adequate performance while running within directX 9 specifications.


Point is, game development takes 2 to 3 years minimum and accelerators come out ever 6-8 months and even longer now. HL2 is an exception to that rule. Unlike the Geforce4 MX line. People who own DX9 cards by nVidia have to deal with the bad DX9 press now instead of later.

By the time DX9 games are in force on the market, rest assured that nVidia will have made their cards to spec.

Dx9 has been in the hands of developers for at least a year. ATI has had a dx9 board out since the summer of 2002. We're starting to see ALOT of Dx9 titles tricke through right now. By christmas there will probably be 2 or 3 major ones and a half dozen minor ones. We won't see a major architectural update from nvidia till maybe a paper launch in december and a real card in febuary.

the Dx9 boat has totally been missed by nvidia.
Its a shame so many 5200 and 5600 cards have been sold. They'll be what holds back all the pretty visuals in games.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.