Message boards :
Number crunching :
Are Benchmarks Accurate?
Message board moderation
Author | Message |
---|---|
Mark Stevens Send message Joined: 15 Feb 01 Posts: 28 Credit: 177,705 RAC: 0 |
Having just started on boinc I was interested to see the benchmarks on my PCs. I was puzzled though by the integer MIPs though which didn't seem to bear much relationship to the cpu speed. For example, I have 2 Dell Precisions 450. One with twin 3.06GHz Xeons and the other with twin 2.66GHz Xeons. Both are running Windows XP Pro SP1. Their respective benchmarks were: 2 x 3.06Ghz fp MIPs: 1964 int MIPs: 2379 2 x 2.66GHz fp MIPs: 1719 int MIPs: 3517 As you can see the slower PC has integer MIPs which are 50% higher which doesn't make much sense?! Both were doing anything else at the time as far as I was aware. Anyhow, I ran them again to make sure. Nothing else was running for sure this time. The floating point MIPs were about the same but the integer ones varied again. Not as much as before but still different. As follows: 2 x 3.06Ghz fp MIPs: 1983 int MIPs: 2632 2 x 2.66GHz fp MIPs: 1727 int MIPs: 2904 (expected 2292 based on above) Does anyone know what's causing this difference? Thanks Mark |
Heffed Send message Joined: 19 Mar 02 Posts: 1856 Credit: 40,736 RAC: 0 |
Benchmarks are still a bit iffy. But they are much better than they have been. ;-) |
Planet Cox Send message Joined: 20 Aug 02 Posts: 4 Credit: 31,712 RAC: 0 |
Well I don't know but I have been looking up some other people's PC numbers and I accept that there may be differences in mother board specs and all but it seems odd to me that some running near enough the same processor have very different ratings. Prompted me to download some benchmark software. This PC i'm on has a gig of DDR 400 ram and all but it (boinc) still says its much slower then a PC with a processor speed of only 100Mhz more. The other PC running boinc here crawls along and that has ratings 1/4 of this and seems to take 4 times longer so maybe they are sort of correct. I might try running it in Linux on the same PC and see if it makes any difference. |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
AMD64 3000+ at 2100 mhz winxp 410 mhz ram i tune up it a little bit ;-) its for me fast more speed not need 2004-07-03 22:06:53 - 2633 double precision MIPS (Whetstone) per CPU 2004-07-03 22:06:53 - 5168 integer MIPS (Dhrystone) per CPU [/url] [/url] Bei der Eroberung des Weltraums sind zwei Probleme zu lösen: die Schwerkraft und der Papierkrieg. Mit der Schwerkraft wären wir fertig geworden.Wernher von Braun |
Thierry Van Driessche Send message Joined: 20 Aug 02 Posts: 3083 Credit: 150,096 RAC: 0 |
The results of benchmark are depending from different factors, a.o. 1. the OS, 2. the running processes, 3. the available RAM. Here is only one example of result of a benchmark, using SiSoftware Sandra Professional version 2004.2.9.104 on a PC running at 2.88GHz: With all processes running Dhrystone ALU : 8587MIPS Whetstone FPU : 3517MFLOPS Whetstone iSSE2 : 6207MFLOPS Unnecessary processes killed and after freeing RAM Dhrystone ALU : 8818MIPS Whetstone FPU : 3597MFLOPS Whetstone iSSE2 : 6363MFLOPS |
Mark Stevens Send message Joined: 15 Feb 01 Posts: 28 Credit: 177,705 RAC: 0 |
Yes, I forgot I had Sandra (v1). The results from that are more consistant and in the correct proportion. Running the boinc benchmark again the fp MIPs are always about the same but the integer ones vary wildly (from 2379-3697). Seems the boinc benchmark is a little off for some reason?! They mention on the description of scoring that bionc uses the benchmark results. Does this mean my scores (I've yet to get any) will be off? |
D. Gustibus Send message Joined: 3 Jan 04 Posts: 9 Credit: 242,311 RAC: 0 |
Speaking of benchmarks, can someone shed some light on the benchmark situation with HT processors? I thought I read a post a few days ago (Heffed?) that mentioned that the benchmarks were faulty on the HT processors and recommended running a manual benchmark with HT disabled. But I also understand that BOINC runs new benchmarks from time to time, and these of course will be done while the system is in normal operation with HT enabled. On my particular system, there is a difference in the integer BOINC benchmark of almost 100% with HT enabled/disabled. What gives here? :) |
Keck_Komputers Send message Joined: 4 Jul 99 Posts: 1575 Credit: 4,152,111 RAC: 1 |
The benchmarking system does still leave a bit to be desired. It has been majorly improved however. At one time it was so bad that HT processors normally claimed double the credit they should and as much as 10 times what they should. @D. Gustibus Running benchmarks with HT off is not a good idea, unless that is how you normally run your computer. I think the post you saw was one where we were trying to evaluate how well the new benchmarks were working. John Keck BOINCing since 2002/12/08 |
Thierry Van Driessche Send message Joined: 20 Aug 02 Posts: 3083 Credit: 150,096 RAC: 0 |
These are some of my results of benchmarking. HT enabled --- - 2004-06-25 12:28:35 - Number of CPUs: 2 --- - 2004-06-25 12:28:35 - 1585 double precision MIPS (Whetstone) per CPU --- - 2004-06-25 12:28:35 - 1876 integer MIPS (Dhrystone) per CPU --- - 2004-06-26 10:20:39 - 1568 double precision MIPS (Whetstone) per CPU --- - 2004-06-26 10:20:39 - 1786 integer MIPS (Dhrystone) per CPU HT disabled, all processes running, no RAM optimization --- - 2004-06-26 10:32:36 - Number of CPUs: 1 --- - 2004-06-26 10:32:36 - 1798 double precision MIPS (Whetstone) per CPU --- - 2004-06-26 10:32:36 - 3868 integer MIPS (Dhrystone) per CPU HT disabled, unnecessary processes killed, after RAM optimization --- - 2004-06-26 10:37:59 - Number of CPUs: 1 --- - 2004-06-26 10:37:59 - 1851 double precision MIPS (Whetstone) per CPU --- - 2004-06-26 10:37:59 - 4001 integer MIPS (Dhrystone) per CPU All above benchmarks have been run manually. Running CPU benchmark AUTOMATICALLY --- - 2004-07-01 10:39:21 - Number of CPUs: 2 --- - 2004-07-01 10:39:21 - 1560 double precision MIPS (Whetstone) per CPU --- - 2004-07-01 10:39:21 - 1823 integer MIPS (Dhrystone) per CPU Conclusions: 1. between the two dates, 6/25 and 7/1, there are not that big differences, 2. between the 2 consecutive benchmarks of the 6/25, the differences are quite acceptable, 3. benchmark results are quite depending on processes that are running and on optimization of RAM, 4. Whetstone numbers does not change that much using HT or not but Dhrystone is dramatically affected. Greetings from Belgium. |
Mark Stevens Send message Joined: 15 Feb 01 Posts: 28 Credit: 177,705 RAC: 0 |
Assuming the benchmarks are off target. In relation to the scoring, what's worse: having a benchmark which is too low or too high? |
Thierry Van Driessche Send message Joined: 20 Aug 02 Posts: 3083 Credit: 150,096 RAC: 0 |
> Assuming the benchmarks are off target. In relation to the scoring, what's > worse: having a benchmark which is too low or too high? In relation to the claimed credit, the best benchmark is the one with the highest score. |
SURVEYOR Send message Joined: 19 Oct 02 Posts: 375 Credit: 608,422 RAC: 0 |
Check this Benchmark Results ============================== Intel Pentium 4HT, 3400 MHz (17 x 200) 1024 MB (DDR SDRAM) Microsoft Windows XP Home Edition Service Pack 1 (05.01.2600.00) Boinc 3.19 Setiathome version 3.08 ...Number of CPUs: 2 ...1867 double precision MIPS (Whetstone) per CPU ...2250 integer MIPS (Dhrystone) per CPU ----------------------------- FreshDiagnose Version 6.60 by fRESH dEVICES http://www.freshdevices.com/ ..CPU Bench Result ..WhetStone FPU 4,859 MWIPS ..DhryStone ALU 7,394 MDIPS ..Speed 3400 MHz ----------------------------- SiSoftware Sandra Version 2004.SP1 ...Benchmark Results ...Dhrystone ALU : 10096MIPS ...Whetstone FPU : 4244MFLOPS ...Whetstone iSSE2 : 7494MFLOPS ============================ Intel Pentium 4A, 2533 MHz (19 x 133) 512 MB (DDR SDRAM) Microsoft Windows XP Home Edition Service Pack 1 (05.01.2600.00) Boinc 3.19 Setiathome version 3.08 ...Number of CPUs: 1 ...1625 double precision MIPS (Whetstone) per CPU ...3634 integer MIPS (Dhrystone) per CPU ------------------------------ FreshDiagnose Version 6.60 by fRESH dEVICES http://www.freshdevices.com/ ..CPU Bench Result ..WhetStone FPU 2,699 MWIPS ..DhryStone ALU 5,778 MDIPS ..Speed 2524 MHz ------------------------------ SiSoftware Sandra Version 2004.SP1 ...Benchmark Results ...Dhrystone ALU : 6478MIPS ...Whetstone FPU : 1860MFLOPS ...Whetstone iSSE2 : 3394MFLOPS =========================== BOINC Alpha Tester BOINC Beta Tester |
©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.