Message boards :
Number crunching :
Any way to control which cores BOINC uses?
Message board moderation
Previous · 1 · 2 · 3 · Next
Author | Message |
---|---|
A-D Send message Joined: 28 Feb 04 Posts: 19 Credit: 13,373,059 RAC: 11 |
Ah, there we go. I didn't realize what "Affinity" was. So now I have Cores 1-3 at max and Core 0 idling. I think something may be screwy with the Core 0 sensor, since it now shows Cores 0, 2, and 3 about the same, and Core 1 a few degrees cooler. So Core 0 is the same temp idling that the others are at max. I'll re-seat the heatsink next week anyway, but surely I have a sensor anomaly. Edit: Cores 0, 2, and 3 at about 55. Core 1 at about 51. Edit 2: Core 0 would be around 60 if it were at max. |
zpm Send message Joined: 25 Apr 08 Posts: 284 Credit: 1,659,024 RAC: 0 |
thats actually pretty normal!!!! it's no harm, i classify anything above 70 c on cpu's as HOT!. mine q6600 runs at about 50-60 c maxed out. what are you using to monitor cpu temps? you not really going to see a true idle temp if other cores are going full steam. and at 2.83 ghz, it'll be a little warmer just b/c of the frequency. you need to do a few test. first series is a 5 minute monitoring of cpu temps at full 100/100 on boinc, second is a 5-10 minute idle rest peroid..... also, if your case doesn't have enough fans to move air, hot air will at a few F degress to the temp but not much, make sure you got a fan to draw air out of the case. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
From my experiences, CPU-affinity give you ~ 1 % or something speed up with BOINC/SETI@home.. It help to eliminate cache miss. At my QX6700 it ran well. It help if you have two (or more) separated CPUs. Or CPUs which use 2 (or 4) separated L2 caches. This are mobos with two (or more) sockets or mobos with one socket and AMD (4 caches)/Intel (2 caches) Quad-Core CPUs. Crunch3r's BOINC V6.1.0 installed over stock Berkeley BOINC V5.10.45 . So no co-processor (CUDA) support. [http://calbe.dw70.de/boinc.html] If you are ~ 10 °C less than the max. CPU temp, then you are well for 24/7 crunching. Q9550 max. 71.4°C [http://processorfinder.intel.com/details.aspx?sSpec=SLB8V]- E0 [http://processorfinder.intel.com/details.aspx?sSpec=SLAWQ]- C1 I read about, you should be ~ 20 °C less than the max. CPU-Core temp for happy CPU life. [Delta to Tjunction max temp] |
A-D Send message Joined: 28 Feb 04 Posts: 19 Credit: 13,373,059 RAC: 11 |
My concern wasn't so much the absolute magnitude of my temps, even that of Core 0. It was the 7 degree disparity I was wondering about. I was interested in setting BOINC to use three cores regardless, in part so I might back off on my fan RPMs a bit, and I just figured that if there were a way to force BOINC always to use Cores 1-3 and not 0, that would be the thing to do. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
My QX6700 have sometimes ~ 8 °C different CPU-Core temps. It's 'normal' that CPU-Core #0 is little hotter than the other 3 CPU-Cores. Current I have at my QX6700 OCed @ 3.14 GHz : 33-38-33-38 [Delta to Tjunction max temp] The temps can also vary depend of the AR of the WUs. [http://www.alcpu.com/CoreTemp] |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Ah, there we go. I didn't realize what "Affinity" was. It could just be your motherboard is giving you weird data. I was using an Asus Striker II NSE board with my E8400 and I was seeing up to 8C different between the cores. When I replaced that with a Gigabyte EP45T-UD3P I now notice the cores are withing 1-2ÂÂC of each other on CPUID's Hardware Monitor. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Now here's the rub......... AFAIK, no. I think, the BOINC devs have no interest for to enable CPU-affinity in the stock BOINC Versions. Since my QX6700 (~ 3 years) I didn't got a positive sign. I got everytime the answer that I need to make a test for to prove that it give a speed up.. But, hurt it - for to 'enable' [make it possible] it in the stock BOINC client and the user can enable/disable it with an cc_config.xml entry? |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Now here's the rub......... I think Crunch3r (where is he, BTW?) did extensive testing and thought that it was worth the few % to include it in his client version back then. And my RAC on the 920 rig has been dropping, so I think I am gonna disable the ncpu test...... "Freedom is just Chaos, with better lighting." Alan Dean Foster |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Just for S&G, I set it to 10......... 10 MB and 2 Cuda.......LOL.. If there is a spare CPU cycle in there anywhere, let Intel find it. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Current QX6700 CPU-Core temps: 32-37-34-39 [Delta to Tjunction max temp] (every sec. jumping higher/less) AFAIK, Crunch3r is only at the Lunatic's forum.. Mark, maybe you would like to PM/EMail the BOINC devs, maybe you have better good luck.. :-) AFAIK, the ncpu entry in cc_config.xml will only 'masquerade/lead to believe' BOINC that your equipment have more CPUs. With BOINC >= V6.6.x you don't need it. BOINC V6.10.x I don't know.. I guess also no.. ;-) |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
|
Lint trap Send message Joined: 30 May 03 Posts: 871 Credit: 28,092,319 RAC: 0 |
re: Affinity If you are running XP Pro, goto THG and look for Task Assignment Manager. It supposedly will let you set affinity where you want and for the program you want. Should work as long as TAM is running. Written back when dual cores ruled, but a screenshot showed 8 core slots (0-7). I found it when looking up 'affinity'. Martin |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
Thanks for posting that Martin..need to look into it! Been using "Process Lasso" to achieve similar effect but it has it's own issues. An integrated solution (like above) would seem best. "The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
I don't really see the need for so much control. There is a "pile" of work units, and a "pile" of processors (cores). If you let it, BOINC will pull work from the work pile, choose processors from the processor pile, and it's done. |
Alinator Send message Joined: 19 Apr 05 Posts: 4178 Credit: 4,647,982 RAC: 0 |
Well, technically BOINC doesn't choose which specific core the task will run on, the OS (or GPU driver/application) does unless affinity is asked for at runtime (or overridden manually). In the case of GPUs, that's one of the major reasons for the biggest complaint by users when running secondary computational tasks on them. Degraded video performance, due to the drivers not being sophisticated enough yet to realize the secondary task is going to burn so much of their time and resources they can't perform their primary task adequately (putting pictures on your screen). Secondly, it depends on ones POV and what the objective trying to be accomplished is. If you're trying to squeeze every last cobblestone out of a dedicated cruncher, then yes this kind of control granularity might come in handy. OTOH, if your running BOINC on a GP workstation or server, it has been well documented that messing around and tying various processes to a specific cores can actually hurt the OSes (and that applies to all of them, not just Windows) ability to manage the overall work flow through the machine. Should Berkeley build that functionality into the stock CC? Probably not, but that doesn't mean it wouldn't be useful in certain special circumstances. Fortunately, there are apparently the means to take that level of control available if one so chooses. Alinator |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13736 Credit: 208,696,464 RAC: 304 |
My concern wasn't so much the absolute magnitude of my temps, even that of Core 0. It was the 7 degree disparity I was wondering about. Try using a different programme, such as Core Temp. Grant Darwin NT |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Well, technically BOINC doesn't choose which specific core the task will run on, the OS (or GPU driver/application) does unless affinity is asked for at runtime (or overridden manually). In the case of GPUs, that's one of the major reasons for the biggest complaint by users when running secondary computational tasks on them. Degraded video performance, due to the drivers not being sophisticated enough yet to realize the secondary task is going to burn so much of their time and resources they can't perform their primary task adequately (putting pictures on your screen). Quite correct, but that wasn't what I was trying to get across: While it may be possible wring out a tiny bit more performance under some limited circumstances, it requires a lot more effort on the operator, and that effort likely isn't worth the trivial improvement. |
Peter M. Ferrie Send message Joined: 28 Mar 03 Posts: 86 Credit: 9,967,062 RAC: 0 |
I'm going to do that next week when I install a new PSU. However, the rig is quite new. I spent a great deal of time lapping the base of my heatsink, and use Tuniq TX-2 thermal grease, so this is irritating. I don't have the courage to lap the CPU heat-spreader. you prob lapped one side a little to much it might not look like it to the naked eye .. my advice is to relap and .. (just my advice : use artic silver 5) p.s. a razor blade is a good tool to see if your lap is pretty flat ( cant be concave or convex ).. they use great precision to make the blade side straight.. also check your screws and make sure there all tight.... |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I'm going to do that next week when I install a new PSU. However, the rig is quite new. I spent a great deal of time lapping the base of my heatsink, and use Tuniq TX-2 thermal grease, so this is irritating. I don't have the courage to lap the CPU heat-spreader. Ops.. maybe I overread this.. ;-) Why you lapped the heatsink? Sometimes the bottoms of the heatsinks are convex because (ONLY) of the Intel CPUs which are concave. Because of this they don't match well with the straight AMD CPUs. If you didn't lapped the heatsink, then they had matched maybe very well. Or it was a heatsink which bottom will be straight if installed. Which heatsink you have? I was 'crazy' and lapped the CPU heat spreader. Ohh what a joy.. ;-( Then a well and straight heatsink, Noctua NH-U12P and everything well. ;-) I was surprised how easy this heatsink can manage 130* W TDP of the QX6700. And only with the stock Noctua fan @ ~ 1,300 RPM. IIRC, Arctic Cooling MX-3. Everybody have his favorite. Only to the CPU. [EDIT: * After OC it's more.. after calculation it's ~ 150 W now..] |
A-D Send message Joined: 28 Feb 04 Posts: 19 Credit: 13,373,059 RAC: 11 |
The heatsink is a Thermalright Ultra 120 Extreme. I'll do the razor blade test when I next have the system apart. Assuming the base looks fine, I'll turn the heatsink 180 degrees and note whether anything changes. |
©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.