Author: Eugene Nalimov
Date: 12:15:41 09/19/03
Go up one level in this thread
When programs are compiled with crap compiler benchmark does not mean much. Let's go to the SPEC web page and look for the official submissions. At least here you'll see best hardware and software configurations. First surprise -- there is no official submission by Apple. Why? If their system is really so good, and they are ready to publish the "official" benchmarks (and not results of the several hand-optimized Photoshop filters), why not officially submit the results? Ok, let's pick Power4+ results submitted by IBM. Power4+ is PPC970 "big brother". IBM presumable used the best compilers they have. There are no 2GHz results, but there are 1.7GHz results, and I suspect Power4+ at 1.7GHz would be not worse than G5 at 2GHz. IBM used Power4+ with 1.5Mb of L2 cache and 128Mb of L3 cache -- for Crafty that means that the entire working set fits into L3. Official IBM results: http://www.spec.org/cpu2000/results/res2003q2/cpu2000-20030505-02136.html Official Dell results: http://www.spec.org/cpu2000/results/res2003q3/cpu2000-20030630-02332.html So Dell is ~7% faster. Not much, but Crafty *likes* 64-bit architecture, and that particular 64-bit architecture was beaten by 32-bit one... Thanks, Eugene On September 19, 2003 at 08:25:00, Steven Edwards wrote: >On September 19, 2003 at 03:11:47, Daniel Clausen wrote: >>On September 18, 2003 at 22:39:34, Steven Edwards wrote: > >>>The gang over at http://www.veritest.com was contracted by Apple to run the >>>SPEC >>>benchmarks on the new PowerMac dual 2.0 GHz PowerPC G5. For comparison >>>purposes, a Dell dual 3.06 GHz Xeon was also used in the testing. The Mac was >>>running Mac OS X 10.2.7 while the Dell ran Red Hat Linux 9.0. >> >>I believe the SPEC numbers once they officially submit them to the SPEC >>organization. > >The Veritest report is worth reading. It seems internally consistent, and it is >also in line with some other third party testing. > >>The "testing" of veritest wasn't the most scientific one, but since there's >>still much room for improvement on the Mac software side (ie a compiler which >>really supports the G5 - I think IBM just released one lately for devs, but I'm >>not sure now) I'm pretty sure it's a very fine machine. > >The new IBM PPC compiler gives about a 20% speed improvement over gcc. However, >it was not used in the Veritest examination. > >>Which top model (Mac or PC) is now 10% faster than the other one is not that >>important, given the fact that Mac were _by far_ slower so far compared to PCs. >>So at least a potential CCT-6 member who uses a new Mac shouldn't have to >>complain about speed. :) > >The leader position changes from time to time. Back in late 1999, the Apple >PowerMac G4 was the only consumer machine to be on the US State Department >export blacklist because the box could do better than 1 GFlop/sec sustained. >Then there were the years where Motorola fell behind by quite a bit. Now it >looks like things are even again, but now with IBM putting its resources into >PPC fab tech, I think that the PPC970 and its relatives will once again assume >the front position. > >>Personally I will wait for the 2nd revision of the board and until Apple >>releases Panther officially. Then I'll see whether I get one of these boxes or >>whether I will go for an dual Opteron or something. > >I agree that it's rarely a good idea to buy the first release of a wholly new >model from Apple. Also, it's not clear that Panther is going to support a 64 >bit user space API; it may only allow for 64 bit addressing of system space, >i.e., more than 4 GB of virtual memory total but less than 4 GB for any one >process.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.