Message boards :
Number crunching :
Added GTX670 Not getting expected performance Boost
Message board moderation
Author | Message |
---|---|
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I replaced my C1060 Tesla card with a GTX670. Its been a couple of weeks now and no performance increase. At the same time I moved the os from win xp x64 to win 7 pro x64. Computer is an i7 960/Asus P6X58D-E mb with 24 GB ram. I'm running Boinc 6.10.60 and Lunatics_Win64_v0.40. Will my current Boinc version suffice or do I need to move up? Also, what is this about a change in the environment variable? Boinc at start up rates the 670 at 457 gflops peak. That seems awfully low. The GTS250 in the same machine is rated higher than that. the 250 is for video only and not running boinc. The GTX670 is running driver 310.90. Delta-V |
Vipin Palazhi Send message Joined: 29 Feb 08 Posts: 286 Credit: 167,386,578 RAC: 0 |
The current issues with the server and unavailability of enough WUs to fill the cache will certainly impair your system from delivering its maximum potential. You should see better RAC once the WUs start flowing freely. I am using 6.10.60 as well and it shouldnt affect performance as long as you have the right optimized apps. You may want to upgrade your Lunatics MB file to x41zc and set up your GPU to crunch astropulse. The following links should help. x41zc Astropulse You can download the latest files from Arkayn's website I will let the more experience crunchers explain the gflops of the GPU. |
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
Boinc at start up rates the 670 at 457 gflops peak. That seems awfully low. Agree. Here's the BOINC startup messages for my factory overclocked 670: 2/6/2013 4:33:31 AM | | CUDA: NVIDIA GPU 0: GeForce GTX 670 (driver version 310.90, CUDA version 5.0, compute capability 3.0, 2048MB, 1992MB available, 2810 GFLOPS peak) 2/6/2013 4:33:31 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 670 (driver version 310.90, device version OpenCL 1.1 CUDA, 2048MB, 1992MB available, 2810 GFLOPS peak) I am using the same driver, so I'm not sure what causes the difference. Unavailability of work and running other projects is depressing my average by 8-10k with a hexcore cpu and a BETA version of BOINC. Edit: I understand the environmental variable for Kepler cards is only needed when running the stock applicationsl. Another Fred Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14644 Credit: 200,643,578 RAC: 874 |
Lee is running BOINC v6.10.60, released 25 March 2011. The GTX 670 wasn't released until 10 May 2012. All this means is that his version of BOINC doesn't know how to do the correct maths to estimate the card speed - it assumes the wrong number of shaders per multiprocessor. So, the information line is wrong, but that doesn't affect the actual running speed of the GPU. Use a separate, external (and up-to-date) tool like MSI Afterburner if you want to be reassured that the GTX 670 is running according to specification. |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
Use a separate, external (and up-to-date) tool like MSI Afterburner if you want to be reassured that the GTX 670 is running according to specification.[/quote] I should have mentioned it is an EVGA GTX670 and I'm running the EVGA Precision X 3.0.4 tool. I set the GPU clock offset to + 23MHZ and the Memory clock offset to + 46MHZ. I set the power Target at 110% but I'm not sure that helps. I haven't tried setting the GPU & Memory any higher than current yet. I am also running 2 work units at a time and it seems to average 10 to 14 minutes eack for the longer ones. NVIDIA GPU 0: GeForce GTX 670 (driver version 31090, CUDA version 5000, compute capability 3.0, 2048MB, 457 GFLOPS peak) NVIDIA GPU 1 (not used): GeForce GTS 250 (driver version 31090, CUDA version 5000, compute capability 1.1, 512MB, 470 GFLOPS peak) There is no opencl line. The current cuda version is Lunatics_x41g_win32_cuda32. Delta-V |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Use a separate, external (and up-to-date) tool like MSI Afterburner if you want to be reassured that the GTX 670 is running according to specification. The x41g_Cuda32 app predates the GTX670 by a long time and is not optimised for it, use the Current x41zc_Cuda5 app that has been Optimised for Keplers, http://jgopt.org/index.html Claggy |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
I noticed that you haven't mentioned how much power you have; how big is your psu and is it enough? I have a 1200 watt psu running an i7/950 with 1 x GTX660SC/2Gb @ 1 x GTX460SE/1GB. I have implemented the evironmental variable because I was having error problems with my Opencl tasks. Currently I'm running Win7/64bit, BOINC 7.0.47, _x41zc cuda 5.0, nVidia 310.90. If you continue to run Opencl make sure that you reserve 1 core via your app_info.xml. BTW, both devices are EVGA and am using PrecisionX 3.0.4. I keep PrecisionX minimized and have it start at startup, automatic fan control and force fan control for each interval checked. If you don't keep it minimized, you won't know what the card temps are and the card may over heat. I don't buy computers, I build them!! |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I noticed that you haven't mentioned how much power you have; how big is your psu and is it enough? I have a 1200 watt psu running an i7/950 with 1 x GTX660SC/2Gb @ 1 x GTX460SE/1GB. I have implemented the evironmental variable because I was having error problems with my Opencl tasks. Currently I'm running Win7/64bit, BOINC 7.0.47, _x41zc cuda 5.0, nVidia 310.90. If you continue to run Opencl make sure that you reserve 1 core via your app_info.xml. The power is 1200 watts. I use EVGA precision on the 670 and a custom fan curve and also keep it minimized. The MSI version didn't work very well with this card but I use the MSI with my fermi cards. I track the temps with GPUZ. The Kepler with 1 work unit at a time is hanging out at about 60C. If time allows, I'll move up to the x41cz tomorrow. I may move to a 7 series Boinc first as well. Delta-V |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
The power is 1200 watts. I use EVGA precision on the 670 and a custom fan curve and also keep it minimized. The MSI version didn't work very well with this card but I use the MSI with my fermi cards. I track the temps with GPUZ. The Kepler with 1 work unit at a time is hanging out at about 60C. If time allows, I'll move up to the x41cz tomorrow. I may move to a 7 series Boinc first as well. PrecisionX works very well with all of my cards. My other machine runs 1 x EVGA GTS250 & 1 x GTX275/GTS250 -- which is a very unique card in that it has two different physical GPUs on the same card. Both machine use the PrecisionX with the same settings with the temps on auto. My office is in a room of my house that has no a/c and no breeze in the summer and the temps on the GTX660SC only got above 70 once when it hit the 90s for a couple of days. I haven't checked my temps on either of machines since mid August. Just checked the temps on the i7/950 and the 660SC is doing about 62c and the 460SE is doing about 52c. I don't buy computers, I build them!! |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
My other machine runs 1 x EVGA GTS250 & 1 x GTX275/GTS250 -- which is a very unique card in that it has two different physical GPUs on the same card. Both machine use the PrecisionX with the same settings with the temps on auto. My office is in a room of my house that has no a/c and no breeze in the summer and the temps on the GTX660SC only got above 70 once when it hit the 90s for a couple of days. I haven't checked my temps on either of machines since mid August. Just checked the temps on the i7/950 and the 660SC is doing about 62c and the 460SE is doing about 52c.[/quote] Never heard of a GTX275/GTS250. My i7 machine is in the house and makes little or no impact on the room temp. The other 3 local machines are in the shop. Its about 22'x26'. It is well insulated and has a heat pump but I've rarely turned it on this winter as the computers put out enough heat to keep it at 60 or above. One pc has a GTX660 Ti that runs 60 to 65C. The other 2 have GTX470s and they run about 70 to 75C. I got tied up working on a Dell laptop today and didn't get to move up to x41zc. Maybe tomorrow or Tuesday. Delta-V |
Cliff Harding Send message Joined: 18 Aug 99 Posts: 1432 Credit: 110,967,840 RAC: 67 |
My other machine runs 1 x EVGA GTS250 & 1 x GTX275/GTS250 -- which is a very unique card in that it has two different physical GPUs on the same card. Both machine use the PrecisionX with the same settings with the temps on auto. My office is in a room of my house that has no a/c and no breeze in the summer and the temps on the GTX660SC only got above 70 once when it hit the 90s for a couple of days. I haven't checked my temps on either of machines since mid August. Just checked the temps on the i7/950 and the 660SC is doing about 62c and the 460SE is doing about 52c. Never heard of a GTX275/GTS250. My i7 machine is in the house and makes little or no impact on the room temp. The other 3 local machines are in the shop. Its about 22'x26'. It is well insulated and has a heat pump but I've rarely turned it on this winter as the computers put out enough heat to keep it at 60 or above. One pc has a GTX660 Ti that runs 60 to 65C. The other 2 have GTX470s and they run about 70 to 75C. I got tied up working on a Dell laptop today and didn't get to move up to x41zc. Maybe tomorrow or Tuesday. It is slowly becoming useless![/quote] Not many people have heard of it Part Number: 012-P3-1178-TR Part Desc: EVGA GeForce GTX 275 CO-OP PhysX Edition; I got it in March, 2010. Didn't know much about cards then and if I could afford it, I would replace it yesterday. 02/07/2013 17:02:06 | | CUDA: NVIDIA GPU 0: GeForce GTX 275 (driver version 310.70, CUDA version 5.0, compute capability 1.3, 896MB, 825MB available, 933 GFLOPS peak) 02/07/2013 17:02:06 | | CUDA: NVIDIA GPU 2 (ignored by config): GeForce GTS 250 (driver version 310.70, CUDA version 5.0, compute capability 1.1, 384MB, 314MB available, 705 GFLOPS peak) 02/07/2013 17:02:06 | | OpenCL: NVIDIA GPU 0: GeForce GTX 275 (driver version 310.70, device version OpenCL 1.0 CUDA, 896MB, 825MB available, 933 GFLOPS peak) 02/07/2013 17:02:06 | | OpenCL: NVIDIA GPU 2 (not used): GeForce GTS 250 (driver version 310.70, device version OpenCL 1.0 CUDA, 384MB, 314MB available, 705 GFLOPS peak) I don't buy computers, I build them!! |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I just upgraded to Boinc 7.0.28 from 6.10.60 and all seems to be working BUT what has happened to the old Messages Tab and where can I get to the info it used to display? Delta-V |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
And where do I go to set local preferences? Delta-V |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I found the "Messages" relabeled under the Advanced Tab as "Event Log" but still can't find any path to local preferences. Delta-V |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
I found the "Messages" relabeled under the Advanced Tab as "Event Log" but still can't find any path to local preferences. Tools -> Computing Preferences... ? |
Lee Gresham Send message Joined: 12 Aug 03 Posts: 159 Credit: 130,116,228 RAC: 0 |
I found the "Messages" relabeled under the Advanced Tab as "Event Log" but still can't find any path to local preferences. Sorry I missed that one. It used to be under the advanced tab. It getting just like the local groceries. always re-arranging the store! Thanks Delta-V |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I prefer to use the older BOINC Manager with the never BOINC. There are a few options not in the GUI in the older manager, but not anything I need to bother with since I generally don't use local settings. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.