Valve Half Life 2 Presentation and Benchmarks

Not only that, but ATi immediately spoke up about it, admitted they'd made some genuine optimizations, but stated in the interest of maintaining appearances if nothing else, they would remove the optimizations relative to 3dmk from their drivers which they did. That's a sharp contrast to nVidia, which has only said: "Damn the torpedos, full cheat ahead!"....

 

Thx dave

 

HL2 article

Excellent article, clear writeup.

The linking to the Excel file (showing detailed FPS data) is very comprehensive and professional.

Very nice article Dave, I thought your commentary on what was said/not said was about the most objective and informative of anything I've read on this to date.

 

Re: HL2 article

Mike G

 

i cant verify the authenticity of this...but if true, i believe it spells "DOOM" for NV35 (no pun intended) ... http://english.bonusweb.cz/interviews/carmackgfx.html

 

"Degrees matter." so we are to ignore little cheats but go hell for leather with bigger ones ? a cheat is and should be critisised for such. your example of a wife beater is ridiculous but working off it i will try and give a better example. man1 hits his wife once and man2 hits here til near death. result is man1 may get a fine or custodial sentace etc and man2 would/should go to jail. the point is both would suffer from their actions yet you want man1 to simple be ignored. thats plain wrong. or how about 2 men rob a bank and 1 takes $1 million while the other takes $10, according to your example the guy who stole $10 should not even be charged.

"When ATI cheated; it only boosted 1%-2% percent." - not quite correct. the cheat boosted speed by 24% in 1 test with an overall of 1%-2% for total tests. so 24% in that test is not enough ? read what i just said above.

"What Valve did was within the specifications of DX9" - incorrect. centroid sampling is NOT a part of DX 9.0. it is planned for DX9.1. problems came about beause neither DX9.0(software) and nvidia DX9.0(hardware) support it. ati hardware does have it but they had to do a hack to access it because of DX9.0(software). this is just 1 example we know of though.

"The bastards from nVidia hadn't even followed the specifications for DX9 as given by MS" - incorrect. read above and you will see valve wanting to do what DX9.0 doesnt support is also the cause. so valve wanting to do what DX9.0 and nvidia hardware doesnt support is nvidias fault ? ye right lets just blame nvidia for everything regardless of the facts.

"admitted they'd made some genuine optimizations" - so detecting and changing operations for whats meant to be a benchmark program(or games that are commonly used for benchmarking) is ok ? bull. anyone who says that should not critisise nvidia for doing the same thing.

well since some people are so one eyed i would be surprised if they didnt have a wooden leg and a parrot on their shoulder so i dont think i will bother reading this anymore. as far as some are concerned valve are to be worshiped(with ati) and have not caused any of these problems by not following current hardware and software limitations. also that ati cheating is ok because they called it legitimate and it wasnt as much as nvidia so it can be ignored as if it didnt happen.

 

Wow, great article. The information in the Excel file on high degrees of AA and AF were especially interesting as I hadn't seen that on any other site.

Now I'm just left wondering what will happen when HDR is fully implemented. Does anyone have any guesses as to the performance hit we might expect with HDR on?

 

If you in Love with Nvidia then you don't see clearly in ATI eyes. (Don't be Offendent please, what I just said)

 

ATI reordered some shader instructions. They produced the same output and did the same calculations. This is a perfectly valid approach to a gaming optimisation, but because it was questionable for a loading based benchmark, ATI removed the optimisation in order to be seen to be acting fairly

Nvidia replaced shaders with low precision versions that ouput low quality images. They placed static clip planes that failed when the camera went off rail. They disabled buffer clears. These are not valid game optimisations AT ALL. They can have no use at all during gameplay, and actively degrade the image quality.
Nvidia did this for no other reason than to artificially boost their score on the benchmark (something they said they would never do), refused to admit it was deliberate, claimed it was "Futuremark out to make us look bad", that it was a "driver bug", and continue to cheat and lie to this day.

So Nvidia blatently cheats to mislead customers, lies about it, and continues the same behaviour of dropping IQ for speed. ATI put in a true optimisation with no loss of quality or calculation, admits what it did, but removes the optimisation to ensure fair play anyway.

How can you say the two companies are the same? If you think they are in the same league of cheating, I'm afraid you really lack an understanding of what happened, despite me explaining it in small words.

 

- For ps_2_0 compliance, the minimum level of internal precision for

temporary registers (r#) is s16e7

ATI R3x0/Rv350 NV30/31/34 NV35
rXX partial precision s16e7 s10e5 s10e5


This is just one of the examples for NV non-compliance with the standard. I don't see here nVidia cards compliant with the PS 2.0 standard. For lackeys of nVidia who tried to put some rules of their own and cheated everyone (fortunately not for a very long time) - stop being ignorant.

 

for the prevoius post little unclear typed, so i will repost it:
ATI R3x0/Rv350 - s16e7
NV (all from 30 to 35) - s10e5

 

If it's partial precision, it's DX9 compliant, i don't see your point? :?

 

Oh, there is quite a similar example - Splinter Cell and its shadows. It's just the other way round. (And there was no time to implement the shadow buffer stuff in the PS on ATi cards, according to a post in these forums of the developer.)

 

Evildeus : do you make a difference between partial precision and minimal internal precision? Cause by minimum i (and hope all sane ppl) uinderstand the LEAST acceptable number. as you can see

Look carefully at the numbers, and i hope that they finally ring a bell in your otherwise obviusly empty head.

 

Heh... Justice would truly be blind, but in the most negative way, if the law did not consider degree such that picking up a quarter off the floor of a bank's lobby and putting it into your pocket was considered the same "crime" as armed robbery.... You are being quite ridiculous in your quest to justify nVidia.

 

No need for such sentence, if you don't feel like explaining further, just don't, you can always hope someone else who is more inclined to explain will do it, but don't behave like this please.

 

You are talking of partial precision, so my remark is correct.

For the rest, well, i know the requirements ...

Ingenu, thx

 

People who don't understand the truth will buy NV3x video cards. Because they heard the name Geforce fast/best/cool but don't know how it is now!!!

 

as for centroid sampling- it IS a part of DX9 already, and there is no such thing as DX9.1. The issue is that PS3.0 is required by DX in order to use centroid sampling, even though it's technically possible to use centroid sampling with less-than-PS3.0. And, yes, PS/VS 3.0 is already in DX9.

oh, and on your wife-beater alaogy man1 did not hit his wife, he just accused her of being lazy (which he then proceded to apologize to her for).