This whole thread is chock full of (willfully or not) "misinformed mumbling." Do we need another thread debating this, particularly one so un-/mis-informed?
The difference is, unlike the aforementioned optimizations nVidia's only offer performance increases in benchmarks, not in actual games. Thats one of FM's problems with what nVidia is doing. According to them if it only increases performance in the benchmark and not in real world situations its not valid.
Ya well HL2 isnt anywhere near done and is Tomb Raider even out yet? You cant really judge performance on beta anything. Remember the Athlon prototypes Tom reviewed in the spring of 99? They showed FPU performance worse than a Pentium 1.
Well from what I gather, Nvidia went to John and asked him what he would like on a GPU. They followed it pretty close so that is why john uses the special extensions and FP16. Not because he has to in order to get Doom3 to work on an FX. But because the card has almost everything he wanted for the engine he was writing. Nvidia has even admitted to building the FX to run Doom3. If ATI had the same extensions I am sure he would use them. I gathered this off the Beyond3d forums.
If you are referring to CG, CG can make code compatible with DX9 from my understanding. So it isnt like Glide where only 3dfx cards can run it. You should be able to use CG and have it work on an ATI card.
Well in some of the games we saw a 70% increase in performance with an increased IQ from 54.xx to 52.xx drivers.
The point is that Futuremark has asked the GPU manufacturers not to use application detection in 3DMark03. This is their call, and NVIDIA should not be trying to sidestep their restrictions. NVIDIA is refusing to go along because they believe this is unrealistic (although ATI seems to have no problem with this!) Whether it is a good idea or not is an entirely separate issue from whether or not NVIDIA should follow the rules that have been established.
First NVIDIA claimed that 3DMark03 was a bad benchmark because it didn't use realistic enough code (and then got caught "cheating" at it after disparaging it as inaccurate and meaningless). Now their problem is that FM won't let them hand-optimize for it (even in ways that could be used for real programs) -- although since a few real DX9.0 games have come out, I notice they've stopped trashing its numbers as unrealistic.
ATI put out a pretty amusing statement a few weeks back about the percentage of games that actually get hand-optimized shaders and the like written for them by the driver team. Basically, only the most popular games and benchmarks will get worked on by NVIDIA and ATI. FM wants 3DMark03 to be representative of the raw hardware/baseline driver performance of the cards, not of how much time the driver team spent working on custom shaders for it. If you don't think that's a good idea, or that it's representative of "real" performance from the cards, then just disregard all 3DMark03 results.
Then shouldn't they provide better results in 3DMark03, even without app-specific optimizations? If not, this means all NVIDIA has been doing for the last 6 months is application-specific enhancements rather than making their drivers faster in general.
First NVIDIA claimed that 3DMark03 was a bad benchmark because it didn't use realistic enough code (and then got caught "cheating" at it after disparaging it as inaccurate and meaningless). Now their problem is that FM won't let them hand-optimize for it (even in ways that could be used for real programs) -- although since a few real DX9.0 games have come out, I notice they've stopped trashing its numbers as unrealistic.
ATI put out a pretty amusing statement a few weeks back about the percentage of games that actually get hand-optimized shaders and the like written for them by the driver team. Basically, only the most popular games and benchmarks will get worked on by NVIDIA and ATI. FM wants 3DMark03 to be representative of the raw hardware/baseline driver performance of the cards, not of how much time the driver team spent working on custom shaders for it. If you don't think that's a good idea, or that it's representative of "real" performance from the cards, then just disregard all 3DMark03 results.
Then shouldn't they provide better results in 3DMark03, even without app-specific optimizations? If not, this means all NVIDIA has been doing for the last 6 months is application-specific enhancements rather than making their drivers faster in general.
- Aug 21, 2002
- 18,368
- 11
- 81
Yes... it SHOULD... but FM has a bug in their ass for some reason. I got better performance in every game I own by switching from 45.xx to 52.xx... but 3DMark2003 scores don't reflect that when FM creates a patch to disable nVidia's "application specific optimizations."
I?ll take a shot at this.
The 52.16 drivers do contain some genuine upgrades in performance because of the unified compiler technology -- and I think 3Dmark3 reflects this to a certain degree. But Nvidia is also hand-coding replacing shaders in 3Dmark2003 and other games to improve performance. This hand-coding is not a realistic strategy for increasing performance beyond a few games because it is so labor intensive.
All Futuremark did was reorder the shaders slightly, ATI has a ??unified compiler??technology too and it was able to cope with the reordering of instructions so ? ATI?s scores remained the same. NV?s scores dropped because the ?compiler? wasn?t just coping with a simple reordering of instructions they were replacing shaders and a simple reordering of them broke this optimization.
This hand-coded shader replacement can also present a problem if a game is updated or new levels for the game are released -- the hand-coded shaders may not work anymore and you loose the performance gains of the hand-coded shaders.
.
The 52.16 drivers do contain some genuine upgrades in performance because of the unified compiler technology -- and I think 3Dmark3 reflects this to a certain degree. But Nvidia is also hand-coding replacing shaders in 3Dmark2003 and other games to improve performance. This hand-coding is not a realistic strategy for increasing performance beyond a few games because it is so labor intensive.
All Futuremark did was reorder the shaders slightly, ATI has a ??unified compiler??technology too and it was able to cope with the reordering of instructions so ? ATI?s scores remained the same. NV?s scores dropped because the ?compiler? wasn?t just coping with a simple reordering of instructions they were replacing shaders and a simple reordering of them broke this optimization.
This hand-coded shader replacement can also present a problem if a game is updated or new levels for the game are released -- the hand-coded shaders may not work anymore and you loose the performance gains of the hand-coded shaders.
.
- Aug 21, 2002
- 18,368
- 11
- 81
In the same respect, Homeworld 2 players will be disappointed when their 9800 Pro doesn't perform the way it does in 3DMark2003.
Solution to the problem... Industry Standard Benchmark out... game benchmarks in.
If FM wants to stay in business, they should get a crapload of demos of the popular games, and measure performance with those instead of running "games" that nobody will ever play.
Solution to the problem... Industry Standard Benchmark out... game benchmarks in.
If FM wants to stay in business, they should get a crapload of demos of the popular games, and measure performance with those instead of running "games" that nobody will ever play.
Yea! Shame on Intel! SSE2 may perform FPU calculations 3x as fast but thats only because it sacrifices a tiny bit of floating point precision.
Ugh. Don't you see what Nvidia is doing? Any benchmark where scores change on a daily basis without a change in image quality is not a legitimate benchmark. HardOCP pointed that out a few days ago. Nvidia is doing it's best to cast doubt on FM's program. It's worked hasn't it? The next big cards that come out won't be able to woe anyone with just a big 3dmark score, because Nvidia has proven that FM is irrelevant.
I don't care how my video card does in some silly synthetic benchmark. If Nvidia wants to use a shader recompiler thats up to them. The FX series is hampered by a lack of physical registers, so code that uses an unnessesary amount of registers is inefficiant on the Nvidia cards.
Ugh. Don't you see what Nvidia is doing? Any benchmark where scores change on a daily basis without a change in image quality is not a legitimate benchmark. HardOCP pointed that out a few days ago. Nvidia is doing it's best to cast doubt on FM's program. It's worked hasn't it? The next big cards that come out won't be able to woe anyone with just a big 3dmark score, because Nvidia has proven that FM is irrelevant.
I don't care how my video card does in some silly synthetic benchmark. If Nvidia wants to use a shader recompiler thats up to them. The FX series is hampered by a lack of physical registers, so code that uses an unnessesary amount of registers is inefficiant on the Nvidia cards.
Well, maybe NV is using shader replacement (in Homeworld) and their performance is too high.: )
3Dmark3 is only a general DX9 guideline. But Futuremark just didn?t pile together a bunch of nonsensical code with no direction or goal in mind and say here?s a benchmark. The engineers at Futuremark, I gather, asked the Q: ? what type of code and things are DX9 software going to be required to run? They also sat down and consulted with many groups in the software/gaming industry about this exact question too. Based on this, they tried (however successfully) to set up a general performance guideline/benchmark.
Agreed. That's pretty much the thing holding the architecture back....
There were two huge mistakes Nvidia initially made with the GeForce FX line:
1) 128 Bit memory interface, corrected as of NV35
2) About half as many PS2.0 registers available as the R3xx architecture
I'd bet dollars to pesos that NV40 is chock full of PS2.0 registers and pixel pipes. I'd be heartily suprised if Nvidia hasn't learned a few things here.
Exatly. It's like racing a F1 with a turbo (for does who don't follow F1: turbo is disalowed). Then FIA comes and say there are some rules to run in F1 and Nvidia says: we have proven our cars are faster than competitors'. Case in wich FIA says: Ok, you're disqualified, go and put a turbo into a WRC car and make it the fastest if you want (turbo is alowed in WRC).
Is just that.
I find it very sad that people actually like nVidia cheating, lying to, and betraying cusomters. Go ahead and rejoice about nVidia once again cheating to decieve the general public. Too bad they dont put forth the energy into making good products that they do acting like babies.
They are only trying to fool the complete dumbassess of the world.
are you a dumbass?
so why are you crying?
/cheer @ Nv
/rude @ FM
are you a dumbass?
so why are you crying?
/cheer @ Nv
/rude @ FM
Some stunningly brilliant discourse here...
*Any* company lying to make their products look better than they really are is a Bad Thing. It's called deceptive advertising, and in most industries this gets you in trouble with the FTC. Since NVIDIA doesn't specifically tout its 3DMark scores as a selling point of its cards (there are plenty of enthusiast sites to do that for them), it can get away with stuff like this.
Futuremark is asking *everyone* (not just NVIDIA) to not use application detection in 3DMark03. NVIDIA is refusing to go along. This is a Bad Thing, as the whole point of a benchmark is that it should be a level playing field -- and in this case, FM has decided they want to measure the baseline driver performance, not how it looks after they've replaced all the shaders with hand-tuned versions. If you don't like what Futuremark is doing, fine! Go talk about them and their unrealistic benchmarks and how they're (supposedly) out to get NVIDIA. Don't hold up NVIDIA as a shining beacon of virtue here, because what they're doing is, at best, verging on deceptive. I would have more respect for them if they stopped trying to do these optimizations -- you can't tell me the time wouldn't be better spent fixing driver bugs and improving performance in real games -- and just went back to their old stance of "3DMark03 isn't representative of real DX9-class games."
*Any* company lying to make their products look better than they really are is a Bad Thing. It's called deceptive advertising, and in most industries this gets you in trouble with the FTC. Since NVIDIA doesn't specifically tout its 3DMark scores as a selling point of its cards (there are plenty of enthusiast sites to do that for them), it can get away with stuff like this.
Futuremark is asking *everyone* (not just NVIDIA) to not use application detection in 3DMark03. NVIDIA is refusing to go along. This is a Bad Thing, as the whole point of a benchmark is that it should be a level playing field -- and in this case, FM has decided they want to measure the baseline driver performance, not how it looks after they've replaced all the shaders with hand-tuned versions. If you don't like what Futuremark is doing, fine! Go talk about them and their unrealistic benchmarks and how they're (supposedly) out to get NVIDIA. Don't hold up NVIDIA as a shining beacon of virtue here, because what they're doing is, at best, verging on deceptive. I would have more respect for them if they stopped trying to do these optimizations -- you can't tell me the time wouldn't be better spent fixing driver bugs and improving performance in real games -- and just went back to their old stance of "3DMark03 isn't representative of real DX9-class games."
vshah
Lifer
- Sep 20, 2003
- 19,003
- 24
- 81
Nvidia created CG because when they started working on their dx9 parts, there was no set standard for dx9, i.e. microsoft HLSL did not exist, and they needed a programming interface/language for their gpus. ATI also had their own compiler/language etc. when the dx9 spec was finalized by microsoft, it ended up closer to ATI's implementation.
also, Futuremark is retarded.
long live my Geforce 2 Ultra
-Vivan
also, Futuremark is retarded.
long live my Geforce 2 Ultra
-Vivan
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter