Tweet
I've performed the RAMspeed and Stream (as a benchmark) on the exact same system (same CPU and RAM) on two different days, resulting in sets T1 and T2 of performance values:
Notice that the Stream results are quite consistent between T1 and T2, but the RAMspeed results are like night and day. The ratio of T1 to T2 results for RAMspeed ranges between .228 to .245 (The T1 RAMspeed performance values are not even 1/4 as good as T2 RAMspeed performance values.)
The RAMspeed performance values probably should be almost as good as the Stream results, and they are in T2, but aren't in T1.
This happens only on Intel Core 2 Duo. The lower values are easily repeatable and higher (expected) values I've gotten only a few times on the Core 2 Duo. I've tested on two different Core 2 Duo processors with similar results, so its probably not the case that my Core 2 Duo is deficient in some way (both CPUs would have to be deficient in the same way).
I have not seen any issue like this with other CPUs.
I noticed that the RAMspeed test always compiles with a WARNING that the build script could not identify the ARCH, but even after fixing this, the anomalous RAMspeed results persist.
The "fix" (it doesn't seem to fix much though) is adding the line
ARCH=i386
at the beginning of the ramspeed/build_pts.sh script. Otherwise the ARCH variable is left undefined and RAMspeed is compiled with some "catch all" compiler options. Actually, the only difference is the RAMspeed test is complied with optimization option O2 versus just O.
Anyone else experience such issues with RAMspeed? Either with Core 2 Duo or any other CPU?
Any suggestions for diagnosing this issue and resolving it?
Ken
Code:
Test Options T1 T2 ======== ========= ==== ==== RAMspeed Copy FP: 584 2381 RAMspeed Scale FP: 563 2367 RAMspeed Add FP: 647 2831 RAMspeed Triad FP: 647 2837 RAMspeed Ave. FP: 610 2601 Stream Copy 2526 2523 Stream Scale 2488 2483 Stream Add 2960 2962 Stream Triad 3003 3007
The RAMspeed performance values probably should be almost as good as the Stream results, and they are in T2, but aren't in T1.
This happens only on Intel Core 2 Duo. The lower values are easily repeatable and higher (expected) values I've gotten only a few times on the Core 2 Duo. I've tested on two different Core 2 Duo processors with similar results, so its probably not the case that my Core 2 Duo is deficient in some way (both CPUs would have to be deficient in the same way).
I have not seen any issue like this with other CPUs.
I noticed that the RAMspeed test always compiles with a WARNING that the build script could not identify the ARCH, but even after fixing this, the anomalous RAMspeed results persist.
The "fix" (it doesn't seem to fix much though) is adding the line
ARCH=i386
at the beginning of the ramspeed/build_pts.sh script. Otherwise the ARCH variable is left undefined and RAMspeed is compiled with some "catch all" compiler options. Actually, the only difference is the RAMspeed test is complied with optimization option O2 versus just O.
Anyone else experience such issues with RAMspeed? Either with Core 2 Duo or any other CPU?
Any suggestions for diagnosing this issue and resolving it?
Ken
Comment