Speed comparisons... amd64 vs em64t?
From "Re: [suse-amd64] Giving up on 64-bit Linux for now" Thread...
Currently, dual opteron 246 (2 Ghz) with 8 Gbyte memory is the fastest performer running our application, compared with Apple dual G5 (2Ghz) and dual Xeon (2.8 Ghz) by a wide margin.
Speaking of which... Does anyone else have any decent comparisons of more real-world apps (not just synthetic benchmarks) between differing 64bit archs, especially opteron vs Intel? I tossed together a quick set of tests using 'openssl speed rsa' between IA64 and opteron (older 2.4 kernel systems) then amd64 vs em64t on some SLES9 installs (http://www.miguelito.org/openssl) and it looks like amd has a nice lead right now. Would love to see more application comparisons though. -- Mike Marion-Unix SysAdmin/Staff Engineer-http://www.qualcomm.com [It's Halloween Kitty gives out raisins] Kids: "Ewwwww..." Kitty: "No, raisins are good for you. Raisins are nature's candy." [As soon as she closes the sliding glass door eggs hit it] Red: "..and eggs are nature's hand grenade. Kitty, don't give them raisins.. it just pisses them off."
Does anyone else have any decent comparisons of more real-world apps (not just synthetic benchmarks) between differing 64bit archs, especially opteron vs Intel?
"Intel" means Xeon here really, the ia64 is so exotic, I haven't seen any offered in the shops, let alone a Linux distro for it. For my thesis I ran the benchmark of the fftw library on various processors. The fftw code dynamically optimises itself at runtime; for scientific number crunching applications I decided that this test was therefore reasonably good. Results had a strong dependence on advanced processors features (i.e. mainly sse/sse2 for Intel, 3dnow for AMD), which was expected. Results also had a strong dependency on data width. The Opteron is unbeatable for single floats, but the Xeon came off a little better for double floats. The difference between single and double is significant, therefore being able to control the processing width in an application can become important. (Rumour has it matlab runs everything at least double - not so hot). The altivec unit in a G4 performed comparatively poorly and only does single float anyway, leaving that architecture without special hardware support for double. The results showed that. Volker -- Volker Kuhlmann is possibly list0570 with the domain in header http://volker.dnsalias.net/ Please do not CC list postings to me.
On Tuesday 21 September 2004 20:00, mmarion@qualcomm.com wrote:
Does anyone else have any decent comparisons of more real-world apps (not just synthetic benchmarks) between differing 64bit archs, especially opteron vs Intel? I currently have an 3.0GHz Xeon (w/ em64t), a 3.6Ghz xeon (w/ em64t) and some dual opterons (244,248, maybe 250 too). Those are for reviewing.
First I want to compare MySQL benchmarks, using both 32 and 64 bits binaries and a mysql compiled from source with gcc 3.4.x. The benchmark will consist of a copy of our database and running a script that does some real-life queries on it with different concurrency levels. Secondly i want to compare webserver performance, apache 2.x w/ dynamic PHP scripts and tux for static files, lets see which platform is the best webserver. Alternativly I want to time some comilations (how fast can they compile a vanilla kernel 10 times etc, useless but a nice comparison ;)).
Would love to see more application comparisons though.
If you have any more idea's of real-world benchmarks, please say so, and i'll try to get them tested too and post the results back to the list. Ultimatly I want to put together a series of (public available) real-world benchmarks which I want to run on quite some platforms to get a nice overview of speed improvements. - kees
On Thu, 2004-09-30 at 07:24, Kees Hoekzema wrote:
Would love to see more application comparisons though.
If you have any more idea's of real-world benchmarks, please say so, and i'll try to get them tested too and post the results back to the list.
They did a nice real world test at http://anandtech.com/linux/showdoc.aspx?i=2213, with desktop focus though and not explicitly em64t/AMD64. But maybe you can use some of their benchmarks as welll... BB
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Kees Hoekzema wrote: | On Tuesday 21 September 2004 20:00, mmarion@qualcomm.com wrote: | |>Does anyone else have any decent comparisons of more real-world apps (not |>just synthetic benchmarks) between differing 64bit archs, especially |>opteron vs Intel? | | I currently have an 3.0GHz Xeon (w/ em64t), a 3.6Ghz xeon (w/ em64t) and some | dual opterons (244,248, maybe 250 too). Those are for reviewing. | | First I want to compare MySQL benchmarks, using both 32 and 64 bits binaries | and a mysql compiled from source with gcc 3.4.x. The benchmark will consist | of a copy of our database and running a script that does some real-life | queries on it with different concurrency levels. Secondly i want to compare | webserver performance, apache 2.x w/ dynamic PHP scripts and tux for static | files, lets see which platform is the best webserver. | Alternativly I want to time some comilations (how fast can they compile a | vanilla kernel 10 times etc, useless but a nice comparison ;)). | | |>Would love to see more application comparisons though. | | | If you have any more idea's of real-world benchmarks, please say so, and i'll | try to get them tested too and post the results back to the list. | | Ultimatly I want to put together a series of (public available) real-world | benchmarks which I want to run on quite some platforms to get a nice overview | of speed improvements. | | - kees | InfoWorld's September 20 issue has a test/comparison of the em64t vs. Opteron. http://www.infoworld.com/article/04/09/17/38FE64shootout2_1.html John -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFBXBSH5JB+bPW3+KwRAsIRAJ97X9y/3vaqEWEiPZ13rifxYL4s3gCeIqdt UbZx82y1ms1d7qJ58ml3fxg= =rgQJ -----END PGP SIGNATURE-----
John Scott wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Kees Hoekzema wrote: | On Tuesday 21 September 2004 20:00, mmarion@qualcomm.com wrote: | |>Does anyone else have any decent comparisons of more real-world apps (not |>just synthetic benchmarks) between differing 64bit archs, especially |>opteron vs Intel? | | I currently have an 3.0GHz Xeon (w/ em64t), a 3.6Ghz xeon (w/ em64t) and some | dual opterons (244,248, maybe 250 too). Those are for reviewing. | | First I want to compare MySQL benchmarks, using both 32 and 64 bits binaries | and a mysql compiled from source with gcc 3.4.x. The benchmark will consist | of a copy of our database and running a script that does some real-life | queries on it with different concurrency levels. Secondly i want to compare | webserver performance, apache 2.x w/ dynamic PHP scripts and tux for static | files, lets see which platform is the best webserver. | Alternativly I want to time some comilations (how fast can they compile a | vanilla kernel 10 times etc, useless but a nice comparison ;)). | | |>Would love to see more application comparisons though. | | | If you have any more idea's of real-world benchmarks, please say so, and i'll | try to get them tested too and post the results back to the list. | | Ultimatly I want to put together a series of (public available) real-world | benchmarks which I want to run on quite some platforms to get a nice overview | of speed improvements. | | - kees |
InfoWorld's September 20 issue has a test/comparison of the em64t vs. Opteron.
http://www.infoworld.com/article/04/09/17/38FE64shootout2_1.html
John
-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.4 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iD8DBQFBXBSH5JB+bPW3+KwRAsIRAJ97X9y/3vaqEWEiPZ13rifxYL4s3gCeIqdt UbZx82y1ms1d7qJ58ml3fxg= =rgQJ -----END PGP SIGNATURE-----
Very interesting article. They didn't get to comparing with SGI, though they mentioned IBM & Apple. I would like to see some of those comparisons as well. We'll give 'em time ....
On Thursday 30 September 2004 16:13, John Scott wrote:
InfoWorld's September 20 issue has a test/comparison of the em64t vs. Opteron.
http://www.infoworld.com/article/04/09/17/38FE64shootout2_1.html
The problem with this test, and anands test too, is that in the mysql benchmark they use mysql's own benchmark. That benchmark does a shitload of inserts and selects, but not the kind you would use in a webpage for instance. I'm going to run them too, and probably get a whole lot different results with my own benchmark based on our website. Last time I ran my own benchmark I tested a 32bits vs a 64bits mysql binary. MySQL's own benchmark gave the 64 bits version quite a lead, but in my testing the 32 bits was roughly 20% faster than the 64bits. I've tested it a couple of times and every time the 32 bits was in the lead up to 20%. Since than I lost my belief in mysql's benchmarks a bit ;).
John
-kees
Alternativly I want to time some comilations (how fast can they compile a vanilla kernel 10 times etc, useless but a nice comparison ;)).
The kernel compilation doesn't have enough computing time because the files are not big enough, it is nearly IO bound on a fast machine. I would recomend something more complicated, e.g. try mozilla source or if you have a lot of time OpenOffice source. C++ tends to compile a lot slower. -Andi
participants (7)
-
Andi Kleen -
Bodo Bauer -
John Scott -
Kees Hoekzema -
mmarion＠qualcomm.com -
Volker Kuhlmann -
William A. Mahaffey III