Folding Forum

Sn1ken wrote:Why Nvidia and AMD is so different I do not know. But this is what I have learned trough my time of folding.

I don't know either, but my guess is that Nvida is looking harder at capturing the game market. If some wasted CPU cycles produce a slightly higher frame rate, then they gain boasting rights. Maybe that doesn't apply to AMD's approach. Who can say?

Umm....Just picked up the first core_15 WU. So now 1 GPU running core_17 and 1 GPU running core_15. Strange thing is the CPU core feeding core_17 is still at 100% (but as we know not really doing anything), but the CPU core feeding core_15 is at just ~5%. Meanwhile the GPU core-17 is still at ~80% load, while the core_15 GPU is at 99-100% load (and the temp of that GPU has gone up by 15c)

The Ciore15 does not need a CPU to poll the GPU the whole time, but you will see bursts of CPU activity to support the GPU when data is transferred.

Whether it is effective to let one CPU thread support more than one nVidia GPU on Core17 or 18 is still AFAIK an open question. While polling, there should be no difference (against a separate thread per GPU), but there is probably some loss if more than one GPU wants data transfer at the same time. If anyone has tried and has comparative figures, please post.

wilding2004 wrote:...the CPU core feeding core_15 is at just ~5%. Meanwhile the GPU core-17 is still at ~80% load, while the core_15 GPU is at 99-100% load (and the temp of that GPU has gone up by 15c)

Core_15 uses CUDA, newer core_17 and core_18 use OpenCL, so they are quite different. I'd be amazed if there'd be an NVidia release where OCL performance equals or outperforms their proprietary CUDA implementation, be it CPU utilization or GPGPU performance.

Yes, quite different.

Core_15 could only run implicit modeling. Much faster science, but with much less detail, roughly speaking.
Core_17 and newer can run both implicit and explicit modeling, like the CPU cores have done for a good while longer.