Valve Half Life 2 Presentation and Benchmarks

The R420, now please quit trolling.

 

That question doesn't seem to be trolling, at least not yet.

An argument dictating that the NV40 is the only possible "wise" answer, contrary to the trend through R3xx/NV3x, might be trolling. So could an argument dictating that the R420 is the only possible "wise" answer, except less unreasonable in the current situation.

Please save the until that happens, though?

Why I don't think there is only one "wise" answer to the above question about "taking the crown"...


Advancement (I think of shader featureset):
Up in the air...I think there are valid reasons to expect the NV40 to be more likely, but I tend to think ATI is intending to implement "R3xx based" with a surprising amount of improvements.

Quality (I think of AA and AF):
I think ATI has a head start, but they also seem to have less obvious paths of advancement because of how significant their head start is. I think NV40 is a shoo-in for "most improved", though, with the final lead somewhat in doubt.

Performance:
It seems to me that the R420 is quite assured, with known info, to lead with regard to shaders. Fixed function is completely up in the air, AFAICS.

Plenty of uncertainty to ask such a question without having the label "troll" thrown around, I think?

 

The already well-known ATI R420 graphics processor probably boasting with a lot of new innovations, such as Shaders 3.0 and DirectX 9.1 support, will be announced at Comdex. ATI’s R420 VPU is expected to have 12x1 architecture, blazing core-speeds and skyrocketing memory.

According to the information from the same web-site that watches all the rumours all the time from all around the web, the NV40 will have 8x2 architecture, include 175 million of transistors and will be equipped with 1500MHz memory. The target clocking of the chip is 600MHz.

It is difficult to judge who will take the crown.

 

Sorry folks, but we have absolutely no clue how R420 will match up against NV40, and anyone who says otherwise is wrong (or breaking an NDA or two).

We don't even know whether the tremendous improvements in shader performance with the new 52.10 drivers (compared to 45.23, running ShaderMark 2.0) are the result of legitimate, general improvements in the scheduler or new cheats. Perhaps we should figure that out before we declare how future drivers will perform on hardware we know essentially nothing about??

 

Or they are pointing out how silly absolutes can be.

I'd agree that we have no basis to justify proclaiming absolutes (and thus with the sentiment of correcting such comments that seem to have been made). But what "clues" we do have...their not being suitable for such proclamations doesn't cause them to stop existing.

 

Are you working for ATi's PR ?

We know NOTHING of the R420.
And sorry, I haven't heard of DX 9.1 either.

 

Pfffft, SPECULATE Parhelia....take a guess and have some fun trying to back it up.

 

This info came from http://www.xbitlabs.com

 

i guess whoever made the comment about ATi using SM 3.0 and DX9.1 is kicking themselves now............

 

adding to my prior post above they still use PS 2.0 and 9.0b

 

Last years best?

Does anyone wonder how the new 6800 Nvidia boards will do? Why are there no beanchmarks with the newest boards going head to head?

 

Re: Last years best?

Well, according to Gabe, they all either blue-screen or burst into flame after the loading screen, and so his assessment is therefore rather vague, as he says: "Obviously, then, there's not much I can say about them because I don't know. I can say with authority that our fire-extinguisher budget has increased by 40% ever since we started trying to get them to run the game--and that after 500% more programming time spent on the attempt than we had to send to get ATi hardware to run the game."

[Nah, just kidding!...]

 

Last Years best..

I would be interested to see the difference when they compare the lastest from each, seem like lots of shader stuff going on; Nvidia is always taking about Sm 3.0. I wonder if it does them good here. Does anyone know if they have plans to release benchmarks on those? Why do you think they chose not to benchmark the newest cards?

 

Re: Last Years best..

because this thread is almost a year old. The benchmarks are over a year old..

Gabe said ati's best is 30% faster than nvidia's best.

THat is the latest news

 

Thanks

Ahh I didnt know. Where did Gabe say this? I would like to read that article too.

 

nm

 

NV40/NV35 and HL2

Hi, i'm new, excuse me if i don't write english very well


NV3X ,in truth ,is a DirectX 8.1 compilant card, more pixel shader 1.x friendly, that has a lot of problems in architecture when run with DirectX9 and complex Pixel Shader 2.0.

But is also true that Valve is the aggravting for NV3X (and also NV40) because it has taken advatage from the dipendence between ALU and TMU in the architecture of NV3X/NV40.

NV40 is a brand new chip regarding at NV3X and has a shader architecture efficient and similar to Ati cards, but unfortunately
has the same dipendence between ALU and TMU of NV3X.

R3XX and R420 instead have indipendent ALU and TMU unit ,and isn't a case if Valve has implemented in HL2's graphic engine the "Radiosity Normal Mapping" technology:

The "Radiosity Normal Mapping" is implemeted with 1920 different pixel shaders. Longest shader has 42 ALU operations and until 7 texture fetches that requires a Floating Point precision of 32bit (FP32)
This is an operation in which the indipendence between ALU and TMU of R3X0 and R420 chips take a lot of advantage at the same clock speed.

This explains because in very intensive FP32 map like "canals", Geforce fx cards in DX9 mode are unplayable.

In this map 6800Ultra has also the same performance of 9800XT (and inferior performance regarding to X800XT) because:

1)Half Life 2 uses a very complex water and illumination algorithm(reflections, refractions etc..) , in wich are necessaries textures fetch and mathematic operations at the same time.
But NV40, because of has dipendent ALU and TMU, can't execute mathmatics operations and textures operations both at the same cycle.

2)NV40/NV3X have the "mini-alu(s)" that supports, in the calculations, the principal operating ALU, and Half-Life 2 is programmed in the way to make not working these mini-ALU(s).

3) Analyzing NV40, R3X0 and R420 pipelines we can see that, imagining the second ALU not working, NV40 remains with a power of only 16 operations per clock(texture fetch or mathematic operation) while R3X0/R420 , with their indipendent ALU and TMU, have a theoretical power of 16 and 32 operation per clock.

4)In this case ("canals map") NV40 and R3X0 they have the same power of calculation per clock and, moreover, R3X0 has a bit superior frequencies and also work at FP24 (against FP32 of geforce cards)

and the shaders are written in this indigestible mode for NV40 and so the 6800Ultra has an advatage of only 8 operations per cycle regarding at 9800XT.

Conclusions:

Half Life 2 shaders are written in the way to maximize the problem of dipendence between ALU and TMU of Geforce card,
also Valve had used a lot of FP32(that is a problem for NV3X ad fortunatley not for NV40) in the game, also where it is superflous, in the way to get worse NV3X performance.

The water shader(very complex in "canals" map) is very Ati-friendly (a lot of mathematic instructions and a bit of texture fetch).


Valve and Id have analyzed Ati and Nvidia architecture and have programmed these in the way to do spite at the rival.

Carmack had programmed Doom3 utilising the lookup table (Ati advise to not use this system) in spite to Ati cards privileging so Nvidia chips and in the way to doubling pixel fill rate performance of Geforces (this, with the Ultrashadow , is the cause of better performance of Nvidia cards in doom3),
while Valve follow the recommendations of Ati , a lot of mathematics calculations and ratio 1:1 between texture fetch and pixel shader (ratio indigestible for NV chips).

Hi

 

Interesting to come back to view this discussion after playing on HL2 for a while now. I waited until I could pick up a copy cheap as I am running a Geforce FX 5900, and didnt expect great things with all the bad benchmarks about.

But then I came across a posting on Guru3D (i think it was) where someone had found a way around the inherent problems supposedly evident with the FX series of cards and HL2.

If you replace the settings for the Geforce FX 5900, with those of the ATi Radeon 9800 Pro within the config text file, and then force Direct X9, then it is possible to get the FX to run in DX9 (using the ATi code path) without graphical glitches, and to run quite playable/ fast at decent frame rate.

Having got the game to do just that - run in DX9 with all water effects rendering properly, and at enjoyable framerate - I dont see why Valve selects DX8 for these cards with no water effects? It runs fine!!

Anyone would think from this that Valve either gave up too early on the FX series, or were paid by ATi not to code its path properly. :?