Message boards :
Number crunching :
4x AMD Radeon R9 Fury X
Message board moderation
Author | Message |
---|---|
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Maybe you remember my old thread here. 4x HD7990 didn't worked. I installed four R9 Fury X VGA cards. The last Windows 8.1 Pro x64 DVD didn't boot. After two months I got a new DVD... - and yes, this disk didn't boot also. (The old Windows 8.1 x64 DVD from an other PC worked (the OS wanted to install).) So Microsoft sell just now DVDs which don't boot? So I burned an .ISO DVD - and finally I could install Windows. Motherboard drivers installed. All updates for Windows. Newest AMD VGA card driver (v15.7.1) installed. Installed BOINC. Opti Lunatics (v0.43b) apps installed (just AP and MB for ATI GPU). 1 MB WU/GPU... - and it 'worked*'. Then I tested 2 MB WUs/GPU... with '-cpu_lock -instances_per_device 2 -hp' and it 'worked*'. In GPU-Z nearly no 'GPU Load'. I opened Task-Manager and I saw all 8 MB GPU WUs were fixed at CPU-thread#0. [* not really] I deletd all in 'cmdline.txt' file. Started BOINC again and all 8 MB WUs were fixed still at CPU-thread#0. CPU-thread#0 was full loaded - the other 23 CPU-threads were idle (2x Xeon (each 6 Core/12 threads = 24 threads in whole)). Why all GPU apps are fixed at CPU-thread#0? Thanks. |
woohoo Send message Joined: 30 Oct 13 Posts: 972 Credit: 165,671,404 RAC: 5 |
i recommend that you turn off hyper threading |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Did you try <cpu_usage>0.3</cpu_usage> That would be 2 cores shutdown for feeding. |
qbit Send message Joined: 19 Sep 04 Posts: 630 Credit: 6,868,528 RAC: 0 |
Nice machine Dirk! Looking forward to see how it performs. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Hi, Your results show affinity mask 1. That limits the number of CPUs to 1. 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 |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I disabled HT in BIOS. Windows see now 12 CPU-threads. But the same like I wrote above. All 8 GPU apps (2 MB WUs/GPU) are fixed at CPU-thread#0. I need to manually choose the affiliation in Task-Manager: 'use all CPU-threads'. Then nearly all CPU-threads do something, ~50% CPU is working (2 MB WUs/GPU= 8). The CPUs do nothing, just GPU app support. I'm very disappointed and angry - I collected the money, build this build, I see the finish line - and then it don't work. I have no idea why all GPU apps are fixed at CPU-thread#0. The motherboard have BIOS v5206. ASUS have v5701 online for upgrade. Could be the BIOS the problem? Or maybe the motherboard drivers? |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
petri33 wrote: Hi, Hi, what you mean? The cmdline.txt file is empty. You mean this? - - - - - - - - - - (In time with HT off, just 12 CPU-threads) (...) <stderr_txt> Running on device number: 3 Priority of worker thread raised successfully Priority of process adjusted successfully, below normal priority class used OpenCL platform detected: Advanced Micro Devices, Inc. BOINC assigns device 3 7 slot of 64 used for this instance Info: BOINC provided OpenCL device ID used Info: CPU affinity mask used: 1 Build features: SETI7 (...) - - - - - - - - - - I can't change this, or? So it's an app problem? |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
have you tried what happens with less GPU tasks? and with only 1 task/device? not that you want to run that way, but I'd want to know if it was something that happens always. I'd also look what happens with only 2 (3) cards in the system. IOW if it's systemic or threshold behaviour. A person who won't read has no advantage over one who can't read. (Mark Twain) |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
The max until now was 2 MB WUs/VGA card. I tried 1 MB WU/VGA card, all 4 GPU apps are still fixed at CPU-thread#0. In Task-Manager it's named: 'CPU 0 (Knoten*: 0)' 'Knoten*: 1' is also there, I guess this is the second CPU (socket). [* german] (BTW. CrossFireX is disabled via AMD tool.) |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Info: CPU affinity mask used: 1 This means that there is a bug (or a new decision to reduce lag?) in the app. If "cmdline.txt file is empty" and you do not have <cmdline> in app_info.xml but the app acts as if you use -cpu_lock_fixed_cpu 0 - it should be a bug I don't remember if BOINC have any ability to set affinity of apps - check app_info.xml - try older app (but not very old) (I was about to suggest using -no_cpu_lock switch but I do not find it in new builds - in older builds (2012) existed -no_cpu_lock switch (it is visible in the exe - e.g. AP6_win_x86_SSE2_OpenCL_ATI_r555.exe)) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
If BIOS has ability to completely disable secondary socket CPU disable it then post results. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
I don't remember if BOINC have any ability to set affinity of apps BOINC doesn't have the ability to set CPU affinity - people keep asking for it, but for reasons which seem of dubious validity, whenever they crop up. The developers feel - and I think they're right - that as far as BOINC's concerned, it should be the operating system which schedules CPUs. If anyone has a particular need to run a particular app on a particular CPU, the app developer is at liberty to accommodate their needs - as Raistmer has done, though I did think it was an optional switch. I must re-read the ReadMes, again. None of the above impacts on BOINC's ability to schedule GPUs independently. That makes far more sense, and indeed it's essential to allow BOINC to act as ringmaster if applications of different types from different projects aren't to end up contending for the same silicon. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Is <use_all_gpus>1</use_all_gpus> in cc_config.xml set ? _\|/_ U r s |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
I must re-read the ReadMes, again. From the ReadMe file for the MultiBeam OpenCL application, r2929, which is the one Dirk is using. -cpu_lock :Results in CPUs number limitation for particular app instance. Also attempt to bind different instances to different CPU cores will be made. Can be used to increase performance under some specific conditions. Can decrease performance in other cases though. Experimentation required. -no_cpu_lock : To disable affinity management (opposite to -cpu_lock option). For ATi version CPUlock affinity management enabled by default. So, affinity management is enabled by default (-cpu_lock mode), but should "bind different instances to different CPU cores". That doesn't seem to be happening here, though I believe there is some doubt over the reporting by the new Windows 8.1 Task Manager. It's possible that Dirk may need to set -total_GPU_instances_num N : To use together with -cpu_lock on multi-vendor GPU hosts. Set N to total number of simultaneously running GPU OpenCL SETI apps for host (total among all used GPU of all vendors). App needs to know this number to properly select logical CPU for execution in affinity-management (- cpu_lock) mode. Should not exceed 64. to accommodate his 8 eight concurrent instances. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Of the readme.txt file: -no_cpu_lock : To disable affinity management (opposite to -cpu_lock option). For ATi version CPUlock affinity management enabled by default. [In the meantime it was mentioned already.] So I used '-no_cpu_lock' and then all CPU-threads were allowed/used. BOINC showed inter alia: Task postponed: Suspicious spike results, host needs reboot or maintenance ...or... Task postponed: Triplet data corruption, retry from checkpoint. So L2-Cache miss? GPU app got support from CPU#0 and then from CPU#1? Then I tried '-cpu_lock -total_GPU_instances_num 12 -hp'. The result like above with -no_cpu_lock', all CPU-threads allowed/used. First I tried '-cpu_lock -total_GPU_instances_num 3 -hp', the result was all GPU apps were fixed at CPU-thread#3. -cpu_lock -instances_per_device 3 -hp All GPU apps fixed at CPU-thread#0. Either I use '-no_cpu_lock' and get the above mentioned errors, or all GPU apps fixed at one CPU-thread, which will be overloaded (one whole thread and all others idle) and reduce very much GPU crunching. What could I do? I can't disable one CPU (socket) in BIOS. 1 WU/GPU (in whole 4 GPU apps): 12 CPU-threads = 30 % CPU 24 CPU-threads = 15 % CPU (HT on) [HT on or off, it's the same CPU support, or?] 2 WUs/GPU (in whole 8 GPU apps): 12 CPU-threads = 45 % CPU 3 WUs/GPU (in whole 12 GPU apps): 12 CPU-threads = 50 % CPU Until now there is no cc_config.xml file. BOINC use all 4 VGA cards, it shows: '0,04C + 0,33 AMD/ATI GPUs (d0)' d0, d1, d2 and d3 (each 3 times) |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Until now there is no cc_config.xml file. Without that setting in cc_config.xml BOINC will only use the first GPU. _\|/_ U r s |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Make sure your config supported by AMD OpenCL runtime. Test few AMD OpenCL samples from their SDK. If some fail then interaction with AMD support required. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
Until now there is no cc_config.xml file. I don't think that's true. It will normally use the "best" GPU, and GPUs close enough to the 'best' to be a loose match. In this case, with four identical GPUs, all should be enabled by default. That would be clear from the device detection report in BOINC's Event Log at startup, and it also seems to be true from Dirk's report of 8 tasks running concurrently. I'll look through the device allocation reports of the tasks reported so far. OpenCL Platform Name: AMD Accelerated Parallel Processing 4433115394 Running on device number: 0 4433115415 Running on device number: 1 4433115674 Running on device number: 2 4433115667 Running on device number: 3 No obvious problems with the GPUs, which takes us back to the CPU question. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Also, any task result I looked into has many restarts. Try to be patient a little and not fiddle with settings. Allow few tasks to complete on their own, w/o restarts and re-sheduling between GPUs. Then provide links to their results on web page. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Raistmer wrote: Make sure your config supported by AMD OpenCL runtime. I'm new related AMD things... I just installed the AMD Catalyst Software Suite (v15.7.1) [of 29.07.2015]... There is also a 'v15.9.1 Beta' [of 30.09.2015] available. Could you (or someone other) give me little bit more infos and URLs? Thanks. |
©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.