Re: Core 17 eats 1 cpu core with Nvidia GPU?
Posted: Mon Sep 09, 2013 6:25 am
I run GPU folding on a straight 3.2 GHz P4 using a GTX 460 and it produces the same PPD as my higher-end machines regardless of the core used. However, I do need to dedicate the full CPU to it i,e, no CPU folding.
Part of the problem may be your CPU folding with the GPU folding. By default the CPU will use both cores but Core_17 requires a full CPU core with Nvidia video cards which is not true for previous GPU cores. Over-subscribing the CPU is a disaster for PPD.
The next part is that the GT 620 is a lower-end GPU and because of that it will not be as productive on Core_17 as previous cores. Core_17 decreased the base points and added QRB (Quick return bonus) points to compensate. However, because your GT 620 is low-end (The GTX 460 is the equivalence point) it will not gain more QRB points than lost base points producing a net loss in PPD.
your best solution currently is not to use client-type=advanced so as not get Core_17. Unfortunately, that may not be a solution long-term because eventually Core_17 will become general-release. PG knows about this issue and is working on a long-term solution.
Part of the problem may be your CPU folding with the GPU folding. By default the CPU will use both cores but Core_17 requires a full CPU core with Nvidia video cards which is not true for previous GPU cores. Over-subscribing the CPU is a disaster for PPD.
The next part is that the GT 620 is a lower-end GPU and because of that it will not be as productive on Core_17 as previous cores. Core_17 decreased the base points and added QRB (Quick return bonus) points to compensate. However, because your GT 620 is low-end (The GTX 460 is the equivalence point) it will not gain more QRB points than lost base points producing a net loss in PPD.
your best solution currently is not to use client-type=advanced so as not get Core_17. Unfortunately, that may not be a solution long-term because eventually Core_17 will become general-release. PG knows about this issue and is working on a long-term solution.