Message boards :
Number crunching :
Anything relating to AstroPulse tasks
Message board moderation
Previous · 1 . . . 7 · 8 · 9 · 10 · 11 · 12 · 13 . . . 120 · Next
Author | Message |
---|---|
Fawkesguy Send message Joined: 8 Jan 01 Posts: 108 Credit: 188,578,766 RAC: 0 |
Fawkesguy, I rechecked the readme text and we are both wrong - Right, and when I copied Cliff's line and pasted it into a text editor, it showed up as "I" (upper case i), not "L". That's why I mentioned it. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
LoL, reading ReadMe should be as fun as writing it ;) And indeed, that switch set parameters of oclFFT library that does FFT for us on GPU. |
Todderbert Send message Joined: 17 Jun 99 Posts: 221 Credit: 53,153,779 RAC: 0 |
In the last week I have had major issues with two of my machines. Both machines are Intel 4690Ks running two 750tis SC in each unit. Tuesday morning one of my machines greeted me with a black screen of death, no response or communications on the LAN, had to hard restart. This morning my second machine was wailing away at 100% fan usage on the GPUs, turned on the monitor and had a frozen desktop with no response. Had to hard restart that one too. Looking through event logs and found no errors on why the lockups occurred. The coincidence was that both machines were running 2-APs on a single 750ti at the time, also I had added this line suggested by Zalster just before the problems: ap_cmdline_win_x86_SSE2_OpenCL_NV.txt -use_sleep -unroll 28 -oclFFT_plan 256 16 256 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -tune 2 64 4 1 -hp These machines have been rock solid till now. They are only a month or two old so there is no dust build up and the line voltage to each machine is good. So I'm leaning on the idea that 2-aps per 750 is too much, and/or the parameters above might not be the perfect command line setup. Any thoughts? |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
That -unroll 28 looks WAY too high to me for a 750ti. I only run -unroll 18 with my 970s. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
In the last week I have had major issues with two of my machines. Both machines are Intel 4690Ks running two 750tis SC in each unit. Tuesday morning one of my machines greeted me with a black screen of death, no response or communications on the LAN, had to hard restart. This morning my second machine was wailing away at 100% fan usage on the GPUs, turned on the monitor and had a frozen desktop with no response. Had to hard restart that one too. Looking through event logs and found no errors on why the lockups occurred. The coincidence was that both machines were running 2-APs on a single 750ti at the time, also I had added this line suggested by Zalster just before the problems: Looking at the context of the conversation when that config was mentioned it seems you were discussing the GTX 970. Those parameters are likely for that configuration rather than the 750 Ti. Looking at Zalster machines with 750 Ti's they are not using that config. Their results show they are using a drastically different configuration for their 750 Ti's. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
WezH Send message Joined: 19 Aug 99 Posts: 576 Credit: 67,033,957 RAC: 95 |
I have got some issues too with my host with two 750ti's. Today and yesterday when I got home from work same thing, black screen (no 100% fan usage), forced to hard restart. Today I'm sure that I was running two AP's/card, can't be sure for yesterday. My cmdline is like this: -use_sleep -unroll 15 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1 -hp So different hosts and different cmdlines, same problems. Don't have any thoughts yet... |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Yes, that -unroll is too high for a 750Ti. As are the rest of the values in that commandline. Better to use 10 to 12 for the 750s. Might be able to do 15 but I keep mine around 12 This is what I use for my 750s -use_sleep -unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -tune 1 64 8 1 -tune 2 128 8 1 -oclFFT_plan 256 16 256 -hp Keep in mine each class of GPUs are going to have different Values. |
WezH Send message Joined: 19 Aug 99 Posts: 576 Credit: 67,033,957 RAC: 95 |
Yes, that -unroll is too high for a 750Ti. As are the rest of the values in that commandline. I suspect that You are running 1WU/GPU? |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
I suspect that You are running 1WU/GPU? 2 Work units per GPU |
WezH Send message Joined: 19 Aug 99 Posts: 576 Credit: 67,033,957 RAC: 95 |
I suspect that You are running 1WU/GPU? Ok, I'll try Your cmdline, seems to be faster than mine, didn't catch any over 4,000 sec runtime validated tasks from Your Ti's. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
I thought unroll should be set to 2 times your Complete Units. My 750Ti has 5. I was use unroll 16 because someone suggested it, it worked but would lock up if I tried to view a video with AP tasks running. So I backed it off to unroll 10 and it seems fine now. |
Todderbert Send message Joined: 17 Jun 99 Posts: 221 Credit: 53,153,779 RAC: 0 |
Fired up Team Viewer at work here and logged in to fix that error. Thank you Zalstar. Being a newb to this AP setup, is there a chart to use for different GPU configurations? I have one machine with 2-960s w/1-750, and another with 2-970s w/1-960. If mixed cards will cause a problem I can pull that 750 and move the 960 over to that machine. |
WezH Send message Joined: 19 Aug 99 Posts: 576 Credit: 67,033,957 RAC: 95 |
Fired up Team Viewer at work here and logged in to fix that error. Thank you Zalstar. Being a newb to this AP setup, is there a chart to use for different GPU configurations? I have one machine with 2-960s w/1-750, and another with 2-970s w/1-960. If mixed cards will cause a problem I can pull that 750 and move the 960 over to that machine. Unfortunately no, there is no chart for different GPU configs, just these forums and it could be hard to find right answer. I can be wrong but: Mixed cards, if they are same "brand" (NVIDIA/AMD/INTEL) will cause problem, they are using same cmdline files. Different brands, like my hosts NVIDIA/AMD, has different cmdline files, so no problem. Correct me if I'm wrong. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
If you are going to mix then go with the setting for the lowest gpu. I think the read me file in lunatics talks about how to set. I cheated, went to the source. Mike. Back when all these started to come out, we were quizzing him about what settings to use. He will probably be along at some point. He's over in Germany if I remember correctly. But my memory tends to be faulty lol Keep that 960 with the 970. Think the 750 is too low for them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Being a newb to this AP setup, is there a chart to use for different GPU configurations? Most advanced chart inside app's ReadMe currently. Additional attempts to systematize gave nothing so far. There are many hints over many threads here but in scattered form. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Just re-read recent ReadMe. Per-device config supported. |
WezH Send message Joined: 19 Aug 99 Posts: 576 Credit: 67,033,957 RAC: 95 |
I stand corrected. |
Todderbert Send message Joined: 17 Jun 99 Posts: 221 Credit: 53,153,779 RAC: 0 |
If you are going to mix then go with the setting for the lowest gpu. I think the read me file in lunatics talks about how to set. I cheated, went to the source. Mike. Back when all these started to come out, we were quizzing him about what settings to use. He will probably be along at some point. He's over in Germany if I remember correctly. But my memory tends to be faulty lol I was thinking of running the three 960s all together, and just running a dual 970 for now. I found a mini 970 but its an ASUS with a backplate, looks like it would be pretty snug against the reference 970. I'll have to read that txt file about the parameter settings and figure it out. A dual 980 build will be happening soon. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Then that should be fine. Good luck with the build |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Per-device config supported. Speaking of that device-specific configuration, I should mention that I got a bit of a surprise several weeks ago after I upgraded drivers on my xw9400. The device numbers got shuffled! OLD ORDER OpenCL: NVIDIA GPU 0: GeForce GTX 660 (driver version 337.88, device version OpenCL 1.1 CUDA, 1536MB, 1498MB available, 2047 GFLOPS peak) OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 337.88, device version OpenCL 1.1 CUDA, 2048MB, 2010MB available, 2082 GFLOPS peak) OpenCL: NVIDIA GPU 2: GeForce GTX 750 Ti (driver version 337.88, device version OpenCL 1.1 CUDA, 2048MB, 2003MB available, 2409 GFLOPS peak) OpenCL: NVIDIA GPU 3: GeForce GTX 660 (driver version 337.88, device version OpenCL 1.1 CUDA, 2048MB, 2013MB available, 2132 GFLOPS peak) NEW ORDER OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 350.12, device version OpenCL 1.2 CUDA, 2048MB, 2009MB available, 1388 GFLOPS peak) OpenCL: NVIDIA GPU 1: GeForce GTX 660 (driver version 350.12, device version OpenCL 1.2 CUDA, 2048MB, 2012MB available, 2132 GFLOPS peak) OpenCL: NVIDIA GPU 2: GeForce GTX 750 Ti (driver version 350.12, device version OpenCL 1.2 CUDA, 2048MB, 2003MB available, 1606 GFLOPS peak) OpenCL: NVIDIA GPU 3: GeForce GTX 660 (driver version 350.12, device version OpenCL 1.2 CUDA, 1536MB, 1497MB available, 2047 GFLOPS peak) Only Device 2 stayed the same. I didn't notice the change until I went to check the results of the device-specific tuning and found that 3 of the 4 GPUs had been running with the wrong parameters. Nothing harmed, because I don't have that big a spread between the 4 devices, but the resulting run times were far enough off my expectations that it finally caught my eye. Is there any way to have a different method of identifying the individual GPUs in the device-specific configuration other than Device 0, Device 1, etc., to avoid this sort of shuffling when upgrading drivers? |
©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.