Article View: comp.os.linux.hardware
Article #494Re: Dual Celeron's and SMP Performance Problems
From: Totally Lost
Date: Fri, 28 May 1999 00:00
Date: Fri, 28 May 1999 00:00
54 lines
3750 bytes
3750 bytes
In article <7ilr3u$r6s$1@sparcserver.lrz-muenchen.de>, kuhn@lpr.e-technik.tu-muenchen.de.nospam wrote: > In comp.os.linux.hardware Totally Lost <idletime@my-deja.com> wrote: > > : Ok, we'll use your numbers... $1000 system, with an additional $150 > : that is 15% more, not 50% more, for that you get a 10-20% speed up, > : a very small part of the time (doesn't speed up editing, backups, > : installs, web browsing, or most other single command activities that > : we occupy our time with in front of the computer). It does however > : speed up a few tasks that some people do frequently. > > My Calculation: > > Price SpeedUp > 2xCeleron333 PC $1150 60-80% > 2xPentiumII/350 PC $1700 70-90% Ok Before you asserted $150 difference, now you've raised that to to $550 ... The 30% difference in your vs my CPU costs at $150 wasn't worth the diversion, but the reality is that for someone building a system the difference is about $100 on a dual system: http://www.shopper.com/pfind/cpu.pentii.333mmx.html $125.50 http://www.shopper.com/pfind/cpu.cel.333mmx.128k.html $ 72.95 CPU Price difference: Uni $52.55, Dual $105.10 So, a 9% increase in cost for a Dual PII, delivers a 20% speed up by your benchmarks. The same 9% also makes the system stable for working sets that are 5-8 times larger. Stated in reverse, saving the 9% in system cost, reduced the performance by 20% for your benchmarks ... that is NOT a good value from a price/performance perspective ... expecially when the price/performance tradeoff gets much much worse for applications with larger working sets with Celerons. How is it that Dual Celerons are better? Simply that they may be overclocked higher, which for small working set computational applications makes them a bit faster. However, if the working set includes the normal X11, filesystem and/or networking OS features, the working set is then expected to be significantly larger than 128K, and there is a significant risk the dual Celeron system will run MUCH slower under load, with the posibility of being slower than a uniprocessor PII due to L2 cache thrashing and extensive memory contention. How this maps to a particular users application requires some study, but I expect that the core OS services that boost the foot print are used in more applications that the relatively tight inner loops that dominate gcc's execution profile. Hence my concern that saving $100 for a Dual Celeron, is a very poor tradeoff for most people. And almost certainly for database, Web and file servers. Sent via Deja.com http://www.deja.com/ Share what you know. Learn what you don't.
Technical Details
Message-ID:
<7imdpm$noi$1@nnrp1.deja.com>Path:
arc1-ca.pugleaf.net!unknown!not-for-mailReferences:
<7ieko6$41e$1@nnrp1.deja.com> <7ifeih$8i6$1@sparcserver.lrz-muenchen.de> <7ik5io$4s0$1@nnrp1.deja.com> <7ikd09$cfi$1@sparcserver.lrz-muenchen.de> <7ikk51$fgu$1@nnrp1.deja.com> <7ilr3u$r6s$1@sparcserver.lrz-muenchen.de>Show Full Headers
Complete Article Headers
X-Google-Language: ENGLISH,ASCII-7-bit X-Google-Thread: 1159f7,acbefbd0f12f9cdd X-Google-Attributes: gid1159f7,public X-Google-Thread: 102b32,acbefbd0f12f9cdd X-Google-Attributes: gid102b32,public X-Google-Thread: 100562,acbefbd0f12f9cdd X-Google-Attributes: gid100562,public From: Totally Lost <idletime@my-deja.com> Subject: Re: Dual Celeron's and SMP Performance Problems Date: 1999/05/28 Message-ID: <7imdpm$noi$1@nnrp1.deja.com> X-Deja-AN: 483133997 References: <7ieko6$41e$1@nnrp1.deja.com> <7ifeih$8i6$1@sparcserver.lrz-muenchen.de> <7ik5io$4s0$1@nnrp1.deja.com> <7ikd09$cfi$1@sparcserver.lrz-muenchen.de> <7ikk51$fgu$1@nnrp1.deja.com> <7ilr3u$r6s$1@sparcserver.lrz-muenchen.de> X-Http-Proxy: 1.0 x24.deja.com:80 (Squid/1.1.22) for client 205.169.102.129 Organization: Deja.com - Share what you know. Learn what you don't. X-Article-Creation-Date: Fri May 28 15:47:03 1999 GMT Newsgroups: comp.os.linux.hardware,alt.comp.hardware.overclocking,csu.unix.linux X-Http-User-Agent: Mozilla/4.07 [en] (X11; I; Linux 2.0.36 i586)