Date: Sat, 18 May 1996 15:56:07 +0200 From: se@zpr.uni-koeln.de (Stefan Esser) To: Michael Smith <msmith@atrad.adelaide.edu.au> Cc: hackers@freebsd.org Subject: Re: EDO & Memory latency Message-ID: <199605181356.AA16136@Sisyphos> In-Reply-To: Michael Smith <msmith@atrad.adelaide.edu.au> "Re: EDO & Memory latency" (May 17, 10:29)
next in thread | previous in thread | raw e-mail | index | archive | help
On May 17, 10:29, Michael Smith wrote: } Subject: Re: EDO & Memory latency } Warner Losh stands accused of saying: } > I take it then the amd chip is pin compatbile with the 486 I have and } > that there will be *NO* problems in pulling one out and putting the } > other in? } } There _should_ be no problems; if your board has jumper settings described } for an AMD 486DX4 or similar it should work fine. No! The AMD 486DX4 used a pinout slightly different from the i486DX4. The 5x86 fixes this, and has to be jumpered identical to an iDX4, except the clock multiplier pin selects between 2x and 3x on the intel chip, and between 4x and 3x for the AMD. ==> Treat it like an i486DX4, just be sure to set the clock multiplier jumper to the 2x position. There is a simple DOS utility that measures cache and memory latency and throughput, and it does also check the internal clock frequency of the CPU. It is called 'ctcm' and should be available from lots of FTP sites. } > What kind of performance increase should I expect? Say on a make } > world and also on CPU bound things. Make world times, ordered by real time. Since the world might have looked a little different in each case :), don't take these numbers to be too meaningful ... In one case, the compile might have been a -stable system, in the other a (bigger) -current. (But the DX2 and 5x86 values were obtained on the same system with -current, just before and after the upgrade.) P5/133, 32MB?, AHC?: 13449.10 real 8789.22 user 2073.65 sys P5/150, 32MB, IDE: 14049.84 real 8798.35 user 1448.19 sys AMD5x86, 16MB, NCR: 15240.39 real 11011.81 user 2551.85 sys iDX4/100, 32MB, BT747: 17411.11 real 12021.55 user 2848.28 sys DX2/66, 16MB, NCR: 22902.31 real 17493.97 user 3495.63 sys } Well, some numbers out of the ol' Dhrystone-2 test gave my DX2/66 about 30K, } the P5-83 about 70K and the P100 here at work about 100K. This was using } the same binary on unloaded systems. Switching the cache settings results in dhrystone 2 results that are up to 50% different, with no actual improvement in real applications. } 'world' times are harder to compare because I went to an NCR PCI SCSI } controller and a faster motherboard, sorry. Well, but they are much more meaningful, since dhrystone can be off by a factor of 1.5! Regards, STefan -- Stefan Esser, Zentrum fuer Paralleles Rechnen Tel: +49 221 4706021 Universitaet zu Koeln, Weyertal 80, 50931 Koeln FAX: +49 221 4705160 ============================================================================== http://www.zpr.uni-koeln.de/~se <se@ZPR.Uni-Koeln.DE>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605181356.AA16136>