Message boards :
Number crunching :
Looking for Advice (GTX 780)
Message board moderation
Author | Message |
---|---|
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
Good Day, I got another system up. Its got two gtx 780 (I know older cards) and I7-3930k 6 cores. Only got 4gb of ram. Whats the max or best number of max concurrent work units to run on each GPU. Should adding more ram, make things faster? Also got the optimized apps installed. <app_config> <app> <name>astropulse_v7</name> <max_concurrent>2</max_concurrent> <gpu_versions> <gpu_usage>.5</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <max_concurrent>9</max_concurrent> <gpu_versions> <gpu_usage>.5</gpu_usage> <cpu_usage>.04</cpu_usage> </gpu_versions> </app> </app_config> Thank you |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Good Day, The 780s are good cards, but they run really hot so make sure you have your fans turned up to keep them cool (I've burned out 3 of them from overheating too much) I don't think you really need the max concurrent (unless you are doing other projects at the same time) You could do up to 3 APs at time on the 780 and the same amount of MBs (but they are going to generate alot of heat and consume electricity) <app_config> <app> <name>astropulse_v7</name> <gpu_versions> <gpu_usage>.33</gpu_usage> <cpu_usage>0.33</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>.33</gpu_usage> <cpu_usage>.04</cpu_usage> </gpu_versions> </app> </app_config> With this setting it should only use 2 cores of the 6, you want to leave 2 more cores free (1 for the OS and the other to help feed the GPUs) so 4 total used of the 6 cores This would be my suggestion. Lets see what other people have to say.. Zalster |
rob smith Send message Joined: 7 Mar 03 Posts: 22222 Credit: 416,307,556 RAC: 380 |
The maximum is between two and three per GPU and is dependant on the exact hardware in use. Be aware you may hit a limit due to your 4GB of RAM - realistically you need about 1GB per CPU core, despite what M$ says :-( (Out of interest, having two computers, one with dual GTX780 and the other with dual GTX980 both have 16GB RAM and are "fed" by AMD FX8 series CPUs I find the GTX780 to have very similar performance to the newer GTX980, however the former does draw a fair bit more power so needs better cooling - the '780 powered beast is down just now while I sort out CPU cooling problems.) Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Hi, 1) I would turn HT on (12 cores). 2) I'd run 6 CPU tasks (Check that you computer does not run out of RAM). Leave 6 HT cores to 'feed' the GPUs. 3) I'd run 3 MB tasks (usage 0.33) and two AP tasks (usage 0.5) I have had a similar (except for 8Mb RAM) configuration for years. p.s. Now I am running a special software for MB. It runs one at a time. Shorties take about 72 seconds. My host To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
Good Day, I took your advice, and made the changes. Right now seti says its running 0.04 cpu + 0.33gpu and six instances running. Does this sound right? Also my GPU temps are at 78c/76c is this good or bad? I also found some ram sticks and moved upto 16gb in quad channel. Not sure how much for a difference this will make. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Yes, 6 total is correct. For the Temps that is about what mine were doing. What program are you using to monitor the Temps? I used to use 76C as my max so fans would be at 100% at that temp. Keep an eye on the Temps. When MB and AP run together, the AP will complete quicker while the MB will take longer just FYI. I run mixes of AP and MB, some people do either 1 or the other but that choice is up to you. Not a big problem as usually there are more MB than APs Zalster Edit. Petri makes a good point, Hyperthreading on those i7 is good, you could do some more work on the extra "cores" but you will want to make sure that you don't use up all the cores. In that case, you would want a <project_max_concurrent>12</project_max_concurrent> added to your config file. This would allow for 6 GPU work units and 6 core work units while still leaving 4 virtual (2 actual) free to support GPU and OS Edit... I would add it between the </app> <project_max_concurrent>12</project_max_concurrent> </app_config> |
Darth Beaver Send message Joined: 20 Aug 99 Posts: 6728 Credit: 21,443,075 RAC: 3 |
Gees guys your not really fair are you hehehehe They haven't told you that you can add -use_sleep in the ap_cmdline_win_x86_SSE2_OpenCL_NV.txt or ap_cmdline_win_x64_SSE2_OpenCL_NV.txt depending on witch one you have and then you will reduce the CPU usage and only need to have 1 core free per GPU instead of 1 core free per AP unit done on the GPU .I would tell you that 1 core is all you need to leave free but your asking it to do 3 units on each card so 1 core mite not be enough you may get away with only 1 core free if you only ask it to do 2 AP per GPU MB's it's not necessary to worry about leaving a core free Also <gpu_usage>.5</gpu_usage> <cpu_usage>1.0</cpu_usage> Drop these two line totally as you have the same type of commands in your appinf.xml . Witch is how I get mine to do more units on the GPU not via app_config In the appinf.xml you change all the lines in it that say <count>?<count> 1 = 1unit 0.5 = 2 units 0.33 = 3 units 0.25 = 4 units 0.20 = 5 units but I would only do 3 max AP's , MB 3 but you can try more doing MB's just watch the temps as it will be close to 100% usage Also you can tweek CUDA open the mbcuda file read it carefully and use in the last 3 lines High 16 200 good luck |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Darth Raccoon makes a good point. You will need some command lines for that 780 to get it running at it's best. (if you don't already have one) I don't remember what mine was for the 780. Mike would know what the best settings would be. For the MB, you don't have to tweak the mbcuda.cfg but if you want you can change the last 3 lines like Darth states, just make sure to remove ;;; from before the last 3 lines to let them be utilized. If you have any more questions, just let us know |
elec999 Send message Joined: 24 Nov 02 Posts: 375 Credit: 416,969,548 RAC: 141 |
Good Day, I use GPUZ, and also like to have open hardware monitor http://openhardwaremonitor.org/ Because I can keep its dashboard open on the desktop and easily monitor the temps |
ace_quaker Send message Joined: 12 Oct 01 Posts: 17 Credit: 33,678,474 RAC: 1 |
Darth Raccoon makes a good point. What's the best command line settings for the 780? |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
-use_sleep -unroll 22 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 4 1 -tune 2 64 4 1 -hp This is what I use for my 980s. It should be close to what we were using for the 780s Mike would be the one for sure to know of the setting. Do you know where to install this? Edit.. In the setiathome.berkeley.edu folder find ap_cmdline_win_x86_SSE2_OpenCl_NV uses right click and notepad to open. Highlight the complete line at the beginning of post and copy and paste. Make sure not to add extra spaces. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
-use_sleep -unroll 22 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 4 1 -tune 2 64 4 1 -hp Looks good to me. With each crime and every kindness we birth our future. |
©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.