Message boards :
Number crunching :
-use_sleep
Message board moderation
Author | Message |
---|---|
cov_route Send message Joined: 13 Sep 12 Posts: 342 Credit: 10,270,618 RAC: 0 |
I'm talking about AP here. I have never experimented with -use_sleep much because I find with my setup the AP cpu usage is low unless there is blanking and I don't have the problem with low gpu usage that some AMD users see. Should I look into it more? Could it reduce cpu usage when there is blanking? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I think sleep is more an nVidia thing. The only cure for Blanking is AstroPulse version 7 currently in Beta. If you want to experiment, come on over and test Blanking's worst nightmare, Astropulse 7,00 released for Linux 32&64, Win 32&64, Win32+AMD/NVIDIA/Intel GPU |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Thats correct. AMD cards dont need use_sleep. With each crime and every kindness we birth our future. |
Wedge009 Send message Joined: 3 Apr 99 Posts: 451 Credit: 431,396,357 RAC: 553 |
With recent NV drivers (not sure of the exact version - maybe 275 onwards) AstroPulse tasks on NV GPUs use a full core of CPU processing, it has something to do with the way the drivers handle waiting. I can't remember the exact details right now - those more experienced can give more information on this. The use_sleep parameter helps reduce CPU usage significantly in these situations, often to the point where CPU usage is negligible (on a zero-blanked task). The down-side is that - depending on your specific hardware/software configuration - there may be a considerable increase in the run-time of the GPU processing overall. One would need to test NV GPU processing with and without the use_sleep switch to decide whether or not the reduction in CPU usage sufficiently offsets the potential increase in GPU processing time. As far as I know, AMD GPUs generally don't have this issue, so use_sleep isn't really needed or recommended. Soli Deo Gloria |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Nvidia cards can get similar run times like AMD cards using use_sleep. I tested a lot of different NV hosts. You need to try different ffa_block values. With each crime and every kindness we birth our future. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Interesting! Is there any direct download link for Astropulse v7.01 for NVidia available? Aloha, Uli |
cov_route Send message Joined: 13 Sep 12 Posts: 342 Credit: 10,270,618 RAC: 0 |
Thanks for the replies everyone, I learned something. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Interesting! AP v7 is to be used only at SETI Beta. The change in processing would make many tasks invalid in comparison to AP v6 processing. Those who want to participate in the Beta testing (and perhaps help it get released earlier here) should attach there. Joe |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
AP v7 is to be used only at SETI Beta. The change in processing would make many tasks invalid in comparison to AP v6 processing. Thanks, got it over there at seti beta. It is running smooth on both NVidias GT 640 & GT 430 but "-use-sleep" still leads to a drop to ~50% in GPU usage. The behavior is similar to astropulse V6, no real improvement seen on my side. [edit] Forgot: I'm using driver 337.88 [edit2] For reference here the host: http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=4982 Aloha, Uli |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
AP v7 is to be used only at SETI Beta. The change in processing would make many tasks invalid in comparison to AP v6 processing. No use_sleep in use on your units Uli. You also have to increase ffa_fetch and ffa_fetch block using use_sleep. Add the params into ap_cmdline_win_x86_SSE2_OpenCL_NV.txt. -use_sleep -ffa_block 4096 -ffa_block_fetch 2048 -tune 1 64 4 1 With each crime and every kindness we birth our future. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
No use_sleep in use on your units Uli. Hi Mike, i'm now using -use_sleep -ffa_block 2304 -ffa_block_fetch 1152 -tune 1 64 4 1 I can't go any further, cause i'm streaming DVB-C on this PC and if i go higher, the stuttering becomes unbearable. Unfortunately i can't use different parameter sets for the different cards like i can do with Lunatics seti v7 executable in mbcuda.cfg: [bus1slot0] ;;; Optional GPU specifc control (requires Cuda 3.2 or newer app), example ;processpriority = abovenormal pfblockspersm = 1 pfperiodsperlaunch = 50 [bus5slot0] ;;; Optional GPU specifc control (requires Cuda 3.2 or newer app), example ;processpriority = abovenormal pfblockspersm = 4 pfperiodsperlaunch = 200 So i can take load from the primary card, which is streaming the DVB-C data. I would really like to see this feature also implemented in the astropulse executables. Aloha, Uli |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Lets see. I will check later. With each crime and every kindness we birth our future. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Lets see. Great! *thumbsup* :) Aloha, Uli |
Darrell Wilcox Send message Joined: 11 Nov 99 Posts: 303 Credit: 180,954,940 RAC: 118 |
To Mike: I have been searching for information on the parameters on the "-tune" that explain what they actually do. Is there a source I can read? |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Check the AstroPulse_OpenCL_NV_ReadMe.txt from app package. With each crime and every kindness we birth our future. |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Lets see. With such low ffa_block values use_sleep doesn`t help. I have tested r_2399 on a 640 and it worked perfectly with 4096 2048. Maybe the 430 is slowing you down but i`m no nvidia expert. You can try -tune 1 32 8 1 and -tune 1 16 16 1 to see if this helps. With each crime and every kindness we birth our future. |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Looks much better Uli. I think -tune 1 16 16 1 is a good value for your configuration. At least CPU time looks O.K. With each crime and every kindness we birth our future. |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Looks much better Uli. Yes, i tried 2560 and 1280, but there was still stutter. I'm now using -use_sleep -ffa_block 2304 -ffa_block_fetch 1152 -tune 1 16 16 1 With this setup i reach >90% GPU load and the streaming is stutter free. Thanks a lot! :) Aloha, Uli |
Ulrich Metzner Send message Joined: 3 Jul 02 Posts: 1256 Credit: 13,565,513 RAC: 13 |
Interesting side note: The very moment i stop streaming DVB-C data, the GPU load drops to exactly 50% on both(!!) cards. So, the system throughput for seti is higher with streaming than without it! I'm puzzled. :? This is new with version 2399. So for now i'll only use the sleep function when streaming data. Aloha, Uli |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
You can try higher tune values whilst not streaming. I would create another ap_cmdline_win_x86_SSE2_OpenCL_NV.txt file with -tune 1 64 4 1 or 1 128 2 1 and try if this speeds things up. Just rename one to tx1 if not in use the other in tx2 if not in use. So you only need to rename the file and have 2 different configurations. 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.