tkh’s blog

Doom 3

Yesterday, I finally got Doom 3. After the installation, I launched Doom 3, but performance was terrible: I got less than 1 fps even at 640×480. I searched on the web about this problem, but nobody has the same problem. My machine has two Athlon MP 2000+, GeForce4 Ti 4600, and 1GB of memory, and it surely meets the minimum requirements. I checked all device drivers are the latest versions, and I was about to conclude that there was hardware problem, but still not convinced because programs using OpenGL or DirectX run just fine.

After several hours struggling with settings, I accidentally found nvemulate.exe in my home directory. As you can guess from the file name, it emulates NV30 (GeForce FX series) on an older card. I had turned on NV30 emulation, and forgot about it. After turning it off, Doom 3 runs very smoothly. I did benchmark using “timedemo demo1.” The results are 19.1 fps at 1024×768 and 24 fps at 800×600 with the high quality graphics setting. It’s not fast, but playable.

I hope id software would release the Linux binary soon so I wouldn’t have to reboot into Windows only to play Doom 3.

GPU languages

In the paper sesssion today, two languages which combine CPU and GPU programming were presented. Brook for GPU is pretty much like C and introduces new contructs which were adopted from Brook, a stream model programming language. In the Q&A session, somebody pointed out that a functional language would be better for reseahchers, and the answer was obvious “it’s a language for general audience, so functional languages are not suitable.” like justification. I thought functional language would be better, but after seeing the second language, it turned out to be wrong. Michael McCool presented Sh, and it’s an C++ API. It seems to be unncecessarily complex and lacks intuition. I need to use both languages, but Brook for GPU seems to be easier to program.

L.A.

I’ve been in L.A since last Saturday. We drove from Albuqurque and it took about 13 hours. The total distance is about 850 miles. The driving wasn’t so bad because I never got sleepy. I don’t know why, but maybe I had enough sleep before leaving home, or I just got used to long driving. I’m attending SIGGRAPH. There are lots of things to write, but don’t have time…

Multipipe machine

We’ve got a loaner machine from SGI. It’s a four Itanium2 processors and two graphics pipe machine running Linux. Since each graphics pipe has two video output, we can use the machine to drive four screen powerwall in the lab. As it’s an early access machine, I don’t mention performance, but with OpenGL Multipipe for Linux, we can run OpenGL apps with four screens. Here are the photos of the machine.