@Karlos

Quote:

Karlos wrote: @DAX Not sure about GPGPU on all hardware architectures, but CUDA on linux kicks #### plain and simple. I've used it on both linux and windows and although the tools available for windows are perhaps nicer (better integration with VS etc), raw performance is hard to beat on linux. I have no idea why it should be the case other than the fact that perhaps linux is just more efficient at dealing with high load tasks in general. I've also found it to be more stable for long running compute tasks and better at distributing load between multiple GPUs. In fact, the Fastra II (http://fastra2.ua.ac.be/), which AFAIK is the fastest "desktop" GPGPU computer around is only made possible via Linux. It has 13 G200 class GPUs which required a custom motherboard bios just to allow them to be mapped in and they've had to modify the kernel to allow all the cards to be accessed. You won't get away with that sort of thing on Windows :)

http://research.microsoft.com/en-us/um/redmond/events/escience2008/matsuoka-escience2008.pdf

1 rack server has 4 gpu cards(8 GPU chips) with Windows 2008 Server.

http://www.excaliberpc.com/593085/asus-p7p55-ws-supercomputer-intel.html
5 PCI-Ex X16 slots ( 10 GPU chips) with Windows 7 support,

Recent Catayst windows 7 drivers supports >32bit address space i.e. issues with DIY ViDock.

Last edited by Hammer on 21-Apr-2011 at 11:27 PM.
Last edited by Hammer on 21-Apr-2011 at 11:05 PM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Karlos

Quote:

Karlos wrote: @DAX Not sure about GPGPU on all hardware architectures, but CUDA on linux kicks #### plain and simple. I've used it on both linux and windows and although the tools available for windows are perhaps nicer (better integration with VS etc), raw performance is hard to beat on linux. I have no idea why it should be the case other than the fact that perhaps linux is just more efficient at dealing with high load tasks in general. I've also found it to be more stable for long running compute tasks and better at distributing load between multiple GPUs. In fact, the Fastra II (http://fastra2.ua.ac.be/), which AFAIK is the fastest "desktop" GPGPU computer around is only made possible via Linux. It has 13 G200 class GPUs which required a custom motherboard bios just to allow them to be mapped in and they've had to modify the kernel to allow all the cards to be accessed. You won't get away with that sort of thing on Windows :)

Relative to RV770, the double float with G200 GPU is pretty slow.

Current multi-GPU PC setups are based on NV Fermi or AMD Cayman.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)