Message boards :
Number crunching :
CPU leaderboards
Message board moderation
Author | Message |
---|---|
Gabriel Siegel Send message Joined: 2 Apr 00 Posts: 10 Credit: 1,430,196 RAC: 0 |
Hi! I have been looking at the CPU list at https://setiathome.berkeley.edu/cpu_list.php and have some questions. I am quite surprised to see the 3.1 Ghz Intel(R) Core(TM) i5-5675R at the very top of the list, when the list is ordered by per-core performance. It even lies above the 3.30 Ghz Intel(R) Core(TM) i7-5775R CPU which makes no sense at all to me. I wonder if this can this be due to overclocking? Perhaps many users having the i5 5675R have overclocked their CPUs leading to the improved performance? And how does BOINC generally deal with overclocking in CPU benchmarks? My own 4.2 Ghz Intel Core i7 7700K model is rated significantly lower in per core performance compared with the aforementioned i5s. I wonder if this has to do with the fact that BOINC considers a hyper-threaded CPU such as the i7 to have 8 cores, thereby reducing the actual per-core performance. So perhaps the "per core" performance is really double that of what BOINC calculates for hyper-threaded CPUs? Let me know your inputs all you clever people out there :-) |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13736 Credit: 208,696,464 RAC: 304 |
I'm not too sure just how they come up with some of the numbers (eg 3.93 average cores per computer?). Looking at my computers, compared to what is in the list, and the GFLOPs/core appears to be based on the Measured Floating Point speed which I would expect is the same benchmarking code for all systems, which wouldn't account for the performance of systems with AVX support, nor higher levels of SSE or SSSE support. So unfortunately the list doesn't give a true or accurate reflection of actual performance. If it were based on the Average Processing Rate it would probably be a much more realistic listing of actual CPU performance. EDIT- the GPU list is broken at the moment, but it too doesn't really reflect the reality of actual GPU performance. Shaggies graphs give a good indication of relative performance, and aren't affected by the running of multiple WUs per GPU as much as Average Processing Rate is. Unfortunately the incredibly variable nature of credit allocation for any given WU still makes it difficult to get a truly accurate idea of performance, but his graphs are still excellent for comparing between different cards work output and the amount of work done for power used. Something that compared run times of the same type of WU (GBT v Arecibo) and the same angle range and was able to differentiate between units running 1,2,3 or more WUs per GPU would be ideal, but is suspect rather difficult to implement. Grant Darwin NT |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I'm not too sure just how they come up with some of the numbers (eg 3.93 average cores per computer?). Probably something as simple as: Total number of reported CPUs/Total number of computers There are a number of users that restrict the number of cores they use for BOINC via the cc_config.xml option <ncpus>. So you will see their CPU information reported something like GenuineIntel SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Darrell Wilcox Send message Joined: 11 Nov 99 Posts: 303 Credit: 180,954,940 RAC: 118 |
@ Gabriel Siegel: Not only what you have noticed, but there are many instances of the same CPU being listed in two places. I counted 126 of these. For example, line 6 (Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz [x86 Family 6 Model 60 Stepping 3] ) and line 21 (Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz [Family 6 Model 60 Stepping 3] ) are the same CPU, but listed with slightly different specs (notice the "x86" in front of "Family"). I haven't found who/how to address such things since there is no contact info on the webpage. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
@ Gabriel Siegel: I believe the Mac BOINC client reports the CPU with the extra x86 information that the other clients do not. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
@ Gabriel Siegel: . . I think that is the difference between users running a 32 bit OS versus a 64 bit OS. They identify differently, the lines with x86 in them are running 32 bit. Yet strangely they seem to outperform the same machines running on 64bit. All so counter-intuitive. . . Also I think the strange values for number of cores is because some users run with hyperthreading off. So when you have 2,400 i7-6700K machines listed and 242 of them (not actual figures, just using them for example) are not running with HT and therefore reporting only 4 cores not 8, then the average will be an odd value like 7.52666. Sorry Hal but I don't think it is to do with the manager configuration to control the number of cores crunching. The benchmarks are run independently of that. Stephen :) |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
@ Gabriel Siegel: You are correct. In that changing "Use at most n% of the CPUs" will not change the number of reported processors by BOINC. However if you read my post I indicated using the <ncpus> value in the cc_config.xml. Which when modified will cause BOINC to report the number of processors you entered. So if you set a value such as <ncpus>100</ncpus> on your i5-6400 system then it would appear on the website and in the stats as: GenuineIntel Intel(R) Core(TM) i5-6400 CPU @ 2.70GHz [Family 6 Model 94 Stepping 3] (100 processors) I feel this image may explain what I was referring to regarding the "[x86 Family" processor information. Additionally "Darwin" is how BOINC reports OS X. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
©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.