linux driver 331.20
Posted: Thu Jan 30, 2014 9:47 pm
I had read the descriptions of low ppd for drivers past 327 for Fermi's in windows, but had seen nothing about the same problem in linux (xubuntu 12.04 in my case). Figured it didn't matter much until a driver past 327 showed up in the repositorys. A couple of days ago driver 331.20 showed up and I installed it on one machine experimentally.
I had been running driver and settings 304.116 or driver and setting 319 with no observable ppd difference. Also did not know if Nvdia driver numbers were the same in linux and windows. Still don't, but linux nv driver 331.20 does drop ppd . The machine I tried it on is a Q6600 @ 3.08 running a 3 cpu slot and a 650 ti @ 980 (stock) slot. The normal performance of the 650 Ti with a 8900 wu when undisturbed is ~23,500 to 26,500 ppd on a 965P with a v1 pcie graphics slot.
On the same machine and OS, undisturbed driver 331.20 dropped the performance to ~15,000 ppd.
I ran it about 8hrs and double-checked the TPF with the SMP2 calculator. OK....no sweat back to driver 319. That doesn't work because 319 auto-updates itself to 331.20 (in the versions tab in synaptic). My fix was to go all the way back to driver 304.116 and before rebooting highlight the driver in synaptic, toggle between description and version to update the versions tab...and then using the "Package" drop down submenu ....force version 304.116 and lock version 304.116 so the system doesn't try to update to a poorly performing (for FAH) driver in the future.
Three notes: None of this means anything if you are not folding on a fermi in linux.
There are doubtless more elegant ways to get this done, but it should work for all 50 people running a fermi in linux.
A NF4 939 Lanparty will run two 650 Ti's in linux, but the ppd is capped at 50,000 for the pair and the A64 at 2.3ghz or 2.6ghz is maxed out as in 100% on both processors and very hot. (informational)
I had been running driver and settings 304.116 or driver and setting 319 with no observable ppd difference. Also did not know if Nvdia driver numbers were the same in linux and windows. Still don't, but linux nv driver 331.20 does drop ppd . The machine I tried it on is a Q6600 @ 3.08 running a 3 cpu slot and a 650 ti @ 980 (stock) slot. The normal performance of the 650 Ti with a 8900 wu when undisturbed is ~23,500 to 26,500 ppd on a 965P with a v1 pcie graphics slot.
On the same machine and OS, undisturbed driver 331.20 dropped the performance to ~15,000 ppd.
I ran it about 8hrs and double-checked the TPF with the SMP2 calculator. OK....no sweat back to driver 319. That doesn't work because 319 auto-updates itself to 331.20 (in the versions tab in synaptic). My fix was to go all the way back to driver 304.116 and before rebooting highlight the driver in synaptic, toggle between description and version to update the versions tab...and then using the "Package" drop down submenu ....force version 304.116 and lock version 304.116 so the system doesn't try to update to a poorly performing (for FAH) driver in the future.
Three notes: None of this means anything if you are not folding on a fermi in linux.
There are doubtless more elegant ways to get this done, but it should work for all 50 people running a fermi in linux.
A NF4 939 Lanparty will run two 650 Ti's in linux, but the ppd is capped at 50,000 for the pair and the A64 at 2.3ghz or 2.6ghz is maxed out as in 100% on both processors and very hot. (informational)