I posted about this once before and then it went stale for a while. But now it seems to have received a second wind.
The GPU core is/has been rewritten in OpenCL instead of CUDA only. This brings the AMD/ATI cards back into the fold. In some cases, AMD is out performing the NVidia. AMD is currently using a small percentage of a CPU core where as NVidia seems to want one full CPU core on OpenCL. Probably due to NVidia drivers, but nothing has been confirmed just yet.
My AMD 7970 is getting about 96K ppd and my NVidia GTX 570 is getting about 34K ppd. Just to give a measurable value to current performance differences with OpenCL on the two different platforms. -- time nor tide wait for no man...
And, if I'm reading the various fora correctly, the OpenCL implementation in core 17 will also still require the use of a logical core for nVidia too...right? I remember you mentioning that previously.
Is core17 still a couple months away? Would it be worth doing the public beta to get a feel for it?
Hopefully we are less than a month away. And since proteneer is now testing larger WU, I'll take that as a sign of confidence in the new core. That and he put the linux version to beta very quickly.
NVidia does consume 1 core of a CPU and AMD consumes very small percentages ( based on driver version ) of a core.
As to the beta question, we can always use more reporting eyes with a variety of hardware and configurations. Plus the core 17 WU's run faster and cooler than the current core 15 WU's for NVidia and it utilizes the AMD more efficiently than the current core 16 WU's. -- time nor tide wait for no man...
Wow. Impressive results with core 17, especially if the the TPF I'm seeing in this small work unit sample is consistent across the board. Hopefully this results in a better polling for the research on the back end.
The 680 and 660ti are running 4-5C cooler on air @ stock on average. Did your 7970 temps drop as well?
PPD: 680=66k; 660ti=45.5k both of these are twice the previous output. Surprisingly the Intel 3930k, which I knocked down to 10 threads, seems to show only a slight hit in TPF resulting in about a 1.5k penalty. But that's a small price to pay for the GPGPU throughput. -- How you can make the world a Better Place
I wasn't running the 7970 on prior WU's so I don't have a point of reference for the core 16 WU's.
But from what I have heard, the AMD/ATI's on core 16 were not using the GPU to it's full potential. Mine runs 97% GPU load on core 17 where as reports on core 16 range from 35-55%. -- time nor tide wait for no man...