Date: Thu, 19 Mar 1998 10:00:03 -0800 (PST) From: Hugh LaMaster <lamaster@george.arc.nasa.gov> To: freebsd-current@FreeBSD.ORG Cc: lamaster@george.arc.nasa.gov Subject: Re: Stream_d benchmark... Wow, there really are differences in Message-ID: <199803191800.KAA01635@george.arc.nasa.gov> In-Reply-To: <199803191010.CAA21492@rah.star-gate.com> from "Amancio Hasty" at Mar 19, 98 02:10:43 am
next in thread | previous in thread | raw e-mail | index | archive | help
For the discussion of the chipsets, I refer to my previous post. This is just to put the numbers in one place so that they can be compared to previous numbers. This discussion should probably be elsewhere - in - hardware perhaps? > This is an Asus motherboard dying to double or more my memory system. > > My PPro200 is about a 1.5 years old and I hope that the new 100Mhz bus > based systems fair better than my system. > ------------------------------------------------------------- > Function Rate (MB/s) RMS time Min time Max time > Copy: 113.7778 0.1557 0.1406 0.1719 > Scale: 107.7895 0.1565 0.1484 0.1719 > Add: 118.1538 0.2158 0.2031 0.2344 > Triad: 118.1538 0.2213 0.2031 0.2344 These numbers are quite good. Hard to tell how much of the differences seen are due to board design, aggressive BIOS settings, memory technology, and chipset. And, perhaps, compilers, although the compiler can't do anything about the poor PPro200/Natoma write bandwidth. These numbers seem high to me based on what I have read previously for generic EDO. Anybody using BEDO out there? > > Soeren Schmidt (sos@FreeBSD.org) wrote: > > > In reply to Jaye Mathisen who wrote: > > > > > > Hmm, Then I should be proud of my noname system (p6/200/128MB 72pEDO): > > > > > > Function Rate (MB/s) RMS time Min time Max time > > > Copy: 117.0286 0.2758 0.2734 0.2812 > > : > > > Triad: 125.3878 0.3917 0.3828 0.4219 Higher yet for generic EDO. > > > > All boxes are P6-200's, 256MB RAM (all RAM is 60ns FP as far as I know). > > > > > > > > Box 1 is a SuperMicro P6DNE: > > > > Function Rate (MB/s) RMS time Min time Max time > > > > Copy: 60.7395 0.2704 0.2634 0.2832 > > > > Triad: 71.1647 0.3494 0.3372 0.3565 > > > > Box 2 is a Digital Prioris HX6000 > > > > Copy: 73.3551 0.2197 0.2181 0.2249 > > > > Triad: 77.4268 0.3108 0.3100 0.3122 > > > > Box 3 is a Digital Prioris ZX6000 > > > > Function Rate (MB/s) RMS time Min time Max time > > > > Copy: 84.8807 0.2018 0.1885 0.2834 > > > > Scale: 97.5461 0.1661 0.1640 0.1720 > > > > Add: 111.6549 0.2179 0.2149 0.2247 > > > > Triad: 100.9468 0.2659 0.2377 0.4237 > > > > Box 3 uses 256bit interleaved memory, rather than whatever the > > > > "standard" is. The web site for stream is http://www.cs.virginia.edu/stream and down in ../standard/Bandwidth.html we see the following for x86 boards tested. Note that some people have complained of the difficulty approaching Intel's "Alder" numbers, for the Orion chipset. That board presumably had a very aggressive memory design, and used Orion with full memory interleaving. Various magazines have reported on what bandwidth the consumer actually gets in a typical system with typical software, and the picture has usually been unpleasant. So -- Interesting that some of the numbers above seem to almost reach the Alder numbers using Natoma w/ EDO. I admit I am surprised. Here are a few numbers, with the big systems for reference and entertainment, and the PC's at the bottom. Note that the highest Intel board tested is a Dell PII_300; unfortunately, chipset is not specified. Note that the way this benchmark counts bandwidth (in and out), a copy shows twice the bandwidth that, e.g., the *rate* of bcopy() would show. All results are in MB/s --- 1 MB=10^6 B, *not* 2^20 B ------------------------------------------------------------------ Machine ID ncpus COPY SCALE ADD TRIAD ------------------------------------------------------------------ [Big Iron - now that's memory bandwidth. About 100X the bandwidth per CPU of the PCs. Too bad the CPUs are so expensive.] NEC_SX_4 32 434784.0 432886.0 437358.0 436954.0 NEC_SX_4 1 15983.0 15984.0 15989.0 15898.0 Cray_T932_321024-3E 32 310721.0 302182.0 359841.0 359270.0 Cray_T932_321024-3E 1 10653.0 10221.0 13014.0 13682.0 Cray_C90 1 6965.4 6965.4 9378.7 9500.7 [Interesting workstation-server numbers, but, not all up to date or the latest models.] SGI_Origin_2000_2 128 21857.6 23351.7 24459.5 22913.6 SGI_Origin_2000_1 32 8556.0 8670.0 9733.0 9435.0 SGI_Origin_2000_1 1 296.0 300.0 315.0 317.0 IBM_RS6000-591 1 711.1 695.7 750.0 800.0 DEC_600au_600 1 227.7 223.0 243.5 248.2 Sun_Ultra2-2200 1 228.5 227.5 258.9 189.9 HP_C180 1 262.3 262.3 244.9 242.4 [PC numbers, unfortunately without the chipset and memory technology info which would help sort this out.] Compaq_Proliant_5000 1 123.1 114.3 141.2 126.3 Dell_P166s 1 119.5 102.4 107.5 104.1 Dell_Pentium_133 1 88.0 125.7 132.0 120.0 Dell_486_DX-2-66 1 33.3 16.5 22.0 18.8 Dell_P6_200 1 102.4 102.4 112.9 112.9 Dell_PII_300 1 188.2 173.0 213.3 188.2 Gateway_2000_P6-200 1 107.9 89.5 100.5 101.6 Gateway_2000_P5-133-66 1 91.4 114.3 126.0 114.0 Intel_Alder_Pentium_Pr 1 140.0 140.0 163.9 167.6 Intel_Pentium-133 1 84.4 77.1 85.7 85.9 Intel_Pentium-100 1 85.1 74.4 77.0 75.2 Intel_Pentium-90 1 46.4 69.9 69.9 69.9 Intel_Pentium-60 1 37.2 62.1 61.3 58.5 PC-clone-AMD-486DX-50 1 38.1 26.2 28.6 23.3 PC-clone-AMD-486DX-80 1 83.9 41.9 39.3 39.3 Viglen_Pentium_60 1 47.1 61.5 63.1 60.0 Micron_P6-200 1 98.4 97.4 106.5 105.0 Micron_P5-120 1 79.3 100.4 109.9 107.7 Asus_Pentium_180 1 76.2 110.3 109.1 100.0 Asus_Pentium_200 1 84.2 123.1 123.1 111.6 Triton_II_Pentium_133 1 93.5 113.3 116.6 110.3 Triton_II_Pentium_133 1 75.9 85.3 87.8 85.3 Gigabyte_586HX 1 88.9 118.5 126.3 117.1 Note: These numbers don't tell the entire bandwidth story - the cache hierarchy, latency, read and write bandwidth at each level, not to mention MP performance, cache-coherency, prefetch, multiple outstanding transactions, etc. etc. etc. are enough to write a (large) book about. However, my experience is that many applications are sensitive to bandwidth and it is worth a little effort to get the most out the CPU. -- Hugh LaMaster, M/S 233-21, ASCII Email: hlamaster@mail.arc.nasa.gov NASA Ames Research Center Or: lamaster@george.arc.nasa.gov Moffett Field, CA 94035-1000 No Junkmail: USC 18 section 2701 Phone: 650/604-1056 Disclaimer: Unofficial, personal *opinion*. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199803191800.KAA01635>