lowenz
21-03-2005, 17:10
E' uscita oggi la versione finale di Sciencemark 2
-Per sistemi a 32 bit:
http://www.sciencemark.org/download/ScienceMark2%2032-bit%2021%20MAR%2005.zip
-Per sistemi a 64 bit:
http://www.sciencemark.org/download/ScienceMark2%2064-bit%2021%20MAR%2005.zip
Changelog
Changelog between 07FEB05 -> 21MAR05
-64-bit support for Windows x64 edition
--------------------------------------------
Important notes for 64-bit support.
: 64-bit is a beta version at the moment.
: Performance may change in the future depending on how we tune for fp performance. You should see significant speedups on both EM64T and AMD64 processors in 64-bit mode. Part of this is due to compiler improvements. : Intel EM64T systems are supported, however, ScienceMark2 won't be able to detect them properly in MemBench.
This will be remedied in a future version.
: CipherBench is not supported in 64-bit mode. Running AES returns an incorrect result, that is impossibly too fast.
This will be remedied in a future version.
: Both SGEMM and DGEMM benchmarks are restricted to compiler only under 64-bit edition.
This will be remedied in a future version.
: There are several new memory cores that are present in the 64-bit build of Membench that are not present in 32-bit edition.
This is due to the new ISA.
: STREAM is tweaked a bit
--------------------------------------------
-Fixed a rare crash that would occur immediately upon startup.
-Sciencemark2 can now be run from the command line!
Command line options:
-automate : Run through each benchmark without user interaction. Upon completion, ScienceMark2 will automatically close.
-cipherbench : Run through the cipher benchmark
-dgemm : Run through the dgemm benchmark using the compiler core
-dgemm:compiler : Run through the dgemm benchmark using the compiler core
-dgemm:sse2s : Run through the dgemm benchmark using the scalar sse2 core
-dgemm:sse2v : Run through the dgemm benchmark using the vector sse2 core
-dgemm:x87 : Run through the dgemm benchmark using the x87 core
-help : Display all command line options
-membench : Run through the memory benchmark
-moldyn : Run through the molecular dynamics benchmark
-moldyn:file.in : Run through a custom simulation using file.in
-primordia : Run through the primordia benchmark
-primordia:file.in: Run through the primordia benchmark using the custom file provided.
-runallbench : Run through the entire benchmark. -automate is implied.
-sgemm : Run through the sgemm benchmark using the compiler core
-sgemm:compiler : Run through the sgemm benchmark using the compiler core
-sgemm:sse : Run through the sgemm benchmark using the sse core
-sgemm:3dnow : Run through the sgemm benchmark using the 3dNow! core
Note that ScienceMark2 in 64-bit edition does not support sgemm:sse, sgemm:3dnow, dgemm:sse2s, dgemm:sse2v, or dgemm:x87.
To run a fully automated benchmark of everything, try:
sciencemark2 -automate -runallbench
- Changed the behavior of the "Run All Benchmarks" function. Sciencemark will no longer run through the entire collection of primordia and molecular dynamics files, but will run through primordia simulation of silver, and molecular dynamics simulation of helium and argon. (these are the default benchmarks that everyone is used to)
- Sciencemark2 calculates a composite score if all benchmarks are run from in the Run all benchmarks mode.
- Sciencemark2 can compare your score with other scores you have generated, or from other people.
For instance, a score of 1000 in an area means your system is comparable to a dual Opteron 246 (2.0 Ghz) with 4 GB registered PC2700 RAM, on a non-NUMA motherboard.
Score is weighted thusly:
(((Memory Score + Stream Score) / 2) + Cryptography Score + Molecular Dynamics Score + Primordia Score + BLAS Score) / 5
Memory Score is a composite between memory latency at 16MB, 512 byte stride and peak memory bandwidth at 16MB.
BLAS Score is a composite between SGEMM peak performance and DGEMM peak performance.
- Added hooks for Intel MKL 7.2 and AMD's ACML 2.5 libraries. We've tested them, and they raise performance quite a bit in BLAS.
Until licensing issues are straightened out, we can't distribute the binaries for these.
Let us know at www.sciencemark.org of any issues with the latest version. There have been quite a few changes in the benchmark, and most of the bugs have been stomped out.
-Per sistemi a 32 bit:
http://www.sciencemark.org/download/ScienceMark2%2032-bit%2021%20MAR%2005.zip
-Per sistemi a 64 bit:
http://www.sciencemark.org/download/ScienceMark2%2064-bit%2021%20MAR%2005.zip
Changelog
Changelog between 07FEB05 -> 21MAR05
-64-bit support for Windows x64 edition
--------------------------------------------
Important notes for 64-bit support.
: 64-bit is a beta version at the moment.
: Performance may change in the future depending on how we tune for fp performance. You should see significant speedups on both EM64T and AMD64 processors in 64-bit mode. Part of this is due to compiler improvements. : Intel EM64T systems are supported, however, ScienceMark2 won't be able to detect them properly in MemBench.
This will be remedied in a future version.
: CipherBench is not supported in 64-bit mode. Running AES returns an incorrect result, that is impossibly too fast.
This will be remedied in a future version.
: Both SGEMM and DGEMM benchmarks are restricted to compiler only under 64-bit edition.
This will be remedied in a future version.
: There are several new memory cores that are present in the 64-bit build of Membench that are not present in 32-bit edition.
This is due to the new ISA.
: STREAM is tweaked a bit
--------------------------------------------
-Fixed a rare crash that would occur immediately upon startup.
-Sciencemark2 can now be run from the command line!
Command line options:
-automate : Run through each benchmark without user interaction. Upon completion, ScienceMark2 will automatically close.
-cipherbench : Run through the cipher benchmark
-dgemm : Run through the dgemm benchmark using the compiler core
-dgemm:compiler : Run through the dgemm benchmark using the compiler core
-dgemm:sse2s : Run through the dgemm benchmark using the scalar sse2 core
-dgemm:sse2v : Run through the dgemm benchmark using the vector sse2 core
-dgemm:x87 : Run through the dgemm benchmark using the x87 core
-help : Display all command line options
-membench : Run through the memory benchmark
-moldyn : Run through the molecular dynamics benchmark
-moldyn:file.in : Run through a custom simulation using file.in
-primordia : Run through the primordia benchmark
-primordia:file.in: Run through the primordia benchmark using the custom file provided.
-runallbench : Run through the entire benchmark. -automate is implied.
-sgemm : Run through the sgemm benchmark using the compiler core
-sgemm:compiler : Run through the sgemm benchmark using the compiler core
-sgemm:sse : Run through the sgemm benchmark using the sse core
-sgemm:3dnow : Run through the sgemm benchmark using the 3dNow! core
Note that ScienceMark2 in 64-bit edition does not support sgemm:sse, sgemm:3dnow, dgemm:sse2s, dgemm:sse2v, or dgemm:x87.
To run a fully automated benchmark of everything, try:
sciencemark2 -automate -runallbench
- Changed the behavior of the "Run All Benchmarks" function. Sciencemark will no longer run through the entire collection of primordia and molecular dynamics files, but will run through primordia simulation of silver, and molecular dynamics simulation of helium and argon. (these are the default benchmarks that everyone is used to)
- Sciencemark2 calculates a composite score if all benchmarks are run from in the Run all benchmarks mode.
- Sciencemark2 can compare your score with other scores you have generated, or from other people.
For instance, a score of 1000 in an area means your system is comparable to a dual Opteron 246 (2.0 Ghz) with 4 GB registered PC2700 RAM, on a non-NUMA motherboard.
Score is weighted thusly:
(((Memory Score + Stream Score) / 2) + Cryptography Score + Molecular Dynamics Score + Primordia Score + BLAS Score) / 5
Memory Score is a composite between memory latency at 16MB, 512 byte stride and peak memory bandwidth at 16MB.
BLAS Score is a composite between SGEMM peak performance and DGEMM peak performance.
- Added hooks for Intel MKL 7.2 and AMD's ACML 2.5 libraries. We've tested them, and they raise performance quite a bit in BLAS.
Until licensing issues are straightened out, we can't distribute the binaries for these.
Let us know at www.sciencemark.org of any issues with the latest version. There have been quite a few changes in the benchmark, and most of the bugs have been stomped out.