Milkyway at home still requires double precision, right?<br /><br />
<br /><br />
If so, maybe that should be noted down since people with the newer GPUs are going to have an awful time , even ones with GTX TITAN X Pascal or VEGA FE.<br /><br />
<br /><br />
IIRC the best GPUs for it:<br /><br />
Tesla P100<br /><br />
Firepro W8100<br /><br />
GTX TITAN Black / GTX TITAN (original) / GTX TITAN Z<br /><br />
Quadro K6000<br /><br />
Tesla K20 , K40, K20X, K80<br /><br />
R9 280X , R9 280 (aka HD7970, HD7950)<br /><br />
HD 6970 , HD 6950<br /><br />
R9 390X, R9 390 (aka R9 290X, R9 290) ... at more power usage than the older Terascale GPUs<br /><br />
HD 5870 , HD 5850
Thanks for the reminder. I'll get it added to the OP.
I just grabbed the stats dump for all of the hosts on [email protected] There are three hosts that are running a Titan V, but so far no results have been posted. I am interested to see what it can do on MilkyWay. It's fp64 performance should blow every other "consumer" GPU out of the water. Peak fp64 on the Titan V is roughly 4x better than the Titan Black.
Anandtech's review in [email protected] had roughly 4x the result of a GTX TITAN (Kepler) in FP64. The same review has 6x original TITAN result for pixel shader compute and OpenCL compute (financial analysis).
It's a shame generally the newer Geforce , Quadros , Radeon Pros (formerly Firepro), and Radeon cards are actually not great for this project.
From reading the thread over on RakeSearch, it sounds like the the avx2nopext should be a bit faster, but no one has posted any times for the app on Zen. I will be loading it up on my 1700 tonight, which is running at 3825 MHz.
*Edit* I grabbed 100 tasks to test with the avx2nopext application. I will grab another 100 to test with the avx app. My 1700 is currently running in Windows.