Has anybody ever installed CASINO on system that uses a Graphycal Processing Unit (GPU)?
QMC seems the right calculation for GPUs.
thank you
casino on gpu?
-
Mike Towler
- Posts: 239
- Joined: Thu May 30, 2013 11:03 pm
- Location: Florence
- Contact:
Re: casino on gpu?
Hi Mauro,
I had a go at this last year using OpenACC and the PGI compiler. It was very frustrating as the compiler was full of stupid bugs, and I had to spend some considerable time negotiating with the PGI people to get them fixed. When I was finally able to get a working GPU-enabled CASINO executable it turned out to be about 27 times slower on 1 CPU and 1 GPU than on a CPU alone. The conclusion was that CASINO would have to be rewritten from the ground up to re-order various loops and so on before the GPU-ization would work, and there was still no guarantee that you would get much benefit. As the modular structure of CASINO makes this very difficult to do, well, frankly I couldn't be bothered. In the end, I decided to wait for the technology and the software to improve; some other efforts continue in the background but don't expect anything useful anytime soon..
There is a more extended discussion of this in the talk I gave at 'QMC in the Apuan Alps' last year, which you can find in my slides on the conference web page: http://www.vallico.net/tti/qmcitaa_12
Best wishes,
Mike
I had a go at this last year using OpenACC and the PGI compiler. It was very frustrating as the compiler was full of stupid bugs, and I had to spend some considerable time negotiating with the PGI people to get them fixed. When I was finally able to get a working GPU-enabled CASINO executable it turned out to be about 27 times slower on 1 CPU and 1 GPU than on a CPU alone. The conclusion was that CASINO would have to be rewritten from the ground up to re-order various loops and so on before the GPU-ization would work, and there was still no guarantee that you would get much benefit. As the modular structure of CASINO makes this very difficult to do, well, frankly I couldn't be bothered. In the end, I decided to wait for the technology and the software to improve; some other efforts continue in the background but don't expect anything useful anytime soon..
There is a more extended discussion of this in the talk I gave at 'QMC in the Apuan Alps' last year, which you can find in my slides on the conference web page: http://www.vallico.net/tti/qmcitaa_12
Best wishes,
Mike