DOOM 3 Benchmarks :
The GeForceFX 5900 Ultra, 5600 Ultra, Radeon 9800 Pro 256MB, 9600 Pro, and 9500 Pro step up to the plate to show you what they have in what will likely be the deal breaker when it comes to buying a new graphics card.
Let me start by saying, "Thank you!" to id Software and NVIDIA for making this article possible. It is truly an honor to be included in an opportunity of this magnitude. John Carmack and Tim Willits of id Software went the extra mile this last Friday to help us make sure we could show this to you today. Also, Jim Black of NVIDIA extended himself above and beyond the call of duty in order to make this all possible.
The Story
What we are showing you here today are DOOM 3 benchmarks. We have been asked not to comment on game content seen during this testing. So don't ask, as we are not telling.
You will notice that we have thanked NVIDIA for making this possible, which will of course bring the conspiracy theorists out of the closet in droves. Yes, NVIDIA was the driving force behind making these benchmarks a public reality.
With that said, be advised that id is also very aware of what is being shown here. The "test2" demo we used was recorded for us Friday in the id offices here in Texas by Tim Willits. An interesting caveat to the story is that the PC that was to be delivered to us here at [H] on Friday for testing was lost by the airlines and never recovered anywhere close to where it was needed. Jim Black, in NVIDIA developer relations, was keen enough to carry a duplicate hard drive on his person while attempting to deliver the system to us.
To make an excruciatingly long story...not so long, we built a machine here to test DOOM 3 on. However, due to strenuous hardware security measures taken by id and NVIDIA to protect the DOOM 3 data from leaks, we were not able to use the scripted tools that were to make our benchmarking much speedier. To outline the facts we: A. Benchmarked on hardware put together and owned by HardOCP. B. Benchmarked using a demo recorded by id at the id offices the day of the benchmarking previously unseen by NVIDIA. c. No scripting language coded by NVIDA was used.
Every DOOM 3 benchmark you see here was run by myself, by hand. The entire benchmarking and confirmation process took us (including Jim Black who stayed for security reasons) right at 16 hours from start to finish and that did include a 5am to 10am nap. I have watched the demo play through somewhere in the neighborhood of 200 times on various hardware.
John Carmack has described the demo we used as a random code drop from a recent build of DOOM 3 and he believes it to be representative of the final game in a performance sense.
Test Setup
We were allowed to test on whatever video cards we wished to with whatever drivers we wished to. As mentioned in the tagline, we used a GeForceFX 5900 Ultra and GeForceFX 5600 Ultra both supplied by NVIDIA. The 5600 Ultra you see here is the final specification of the NV31 GPU and has not yet been seen by the public. It will be announced soon.
The ATI cards used included a sample 9800 Pro 256MB and a 9600 Pro sample supplied by ATI. We also included a 9500 Pro card from TYAN.
NVIDIA DetonatorFX v44.03, ATI Catalyst v3.4 & v3.2.
How NOT to Benchmark
Seeing as we have been testing the most recent ATI and NVIDIA graphics cards, we of course had the latest versions of the drivers on hand, and that is what we start out using. Directly comparing the 5900 Ultra with the 9800 Pro 256MB, this is what we saw.
If you were saying to yourself that something is not right here, we would have to agree with you. It seems that ATI's latest Catalyst driver is broken when it comes to running DOOM 3.
We pulled back to the driver versions for all cards based on Catalyst v3.2. Now this leaves us with a driver that we are told only has the ability to access 128MB of RAM on the 9800-256 card. While this does not give us a 100% apples to apples comparison between the 5900 Ultra and the 9800 Pro-256, it is the best we could do while showing the 9800 in the best possible light. From the scaling shown, it is our thought that the 128MB of "missing" memory did not impact the scores, but it is something to be taken into account.
![[H]ard|OCP](logo0000.gif){kind=logo}
![[H]ard Menu](mainmenu.gif){kind=small}
Main Page
