bzs@world.std.com (Barry Shein) (12/28/90)
Is anyone aware of any correlative analysis (statistical methods) comparing Dhrystone to SPEC. In particular, to what extent one is or isn't a predictor of the other? -- -Barry Shein Software Tool & Die | {xylogics,uunet}!world!bzs | bzs@world.std.com Purveyors to the Trade | Voice: 617-739-0202 | Login: 617-739-WRLD
mikes@iuvax.cs.indiana.edu (Michael Squires) (12/28/90)
In article <1990Dec28.043824.28267@world.std.com> bzs@world.std.com (Barry Shein) writes: >Is anyone aware of any correlative analysis (statistical methods) >comparing Dhrystone to SPEC. In particular, to what extent one is or >isn't a predictor of the other? This would be fairly simple to do. Where would I find the data? At this point are there non-proprietary benchmarks that are useful (in the sense that a single user could obtain and run the benchmark without having to pay more than access costs for the sources)? I personally find the Dhrystone to be useful since I use the same code on all the machines that I test, and there is an excellent relationship between the results and my subjective evaluation of the hardware/software. For example, the Dhrystone showed me that I was getting sub-par performance from my 486/25 and the manufacturer has accepted the motherboard for rework. I didn't need the AIM suite or the SPEC suite for this. Things that I've seen/have: Dhrystone 1.1 and 2; iobench2 (for testing disk I/O); the old Byte UNIX benchmarks (I understand the disk I/O part is faulty, gets fooled by caches); the Whetstone and Linpack floating point benchmarks; the NCR System Characterization benchmark; and the Monash University benchmark. The Dhrystone, Whetstone, Byte, NCR, and Monash sources were posted to various newsgroups. -- Mike Squires (mikes@iuvax.cs.indiana.edu) 812 855 3974 (w) 812 333 6564 (h) mikes@iuvax.cs.indiana.edu 546 N Park Ridge Rd., Bloomington, IN 47408 Under construction: mikes@sir-alan.cica.indiana.edu
rnovak@mips.COM (Robert E. Novak) (12/29/90)
I'm willing to take Barry up on his offer if someone will fill in the Dhrystone 1.1 data for the machines that I list below. I have the SPEC data on all of the following machines loaded into a spreadsheet. In article <1990Dec28.043824.28267@world.std.com> bzs@world.std.com (Barry Shein) writes: >Is anyone aware of any correlative analysis (statistical methods) >comparing Dhrystone to SPEC. In particular, to what extent one is or >isn't a predictor of the other? > >-- > -Barry Shein > >Software Tool & Die | {xylogics,uunet}!world!bzs | bzs@world.std.com >Purveyors to the Trade | Voice: 617-739-0202 | Login: 617-739-WRLD System MHz Chip Cache Spec Iss Alacron AL860 25 i860 2 Alacron AL860 Accelerator 40 i860 12 3 Alacron AL860 40 i860 12 4 AT&T StarServer E 33 i486 128 3 CDC 4340 25 R3000 128 2 CDC 4360 25 R3000 128 2 CDC 4380 25 R3000 128 2 CDC 4680(Beta) 60 R6000 592 2 CYBER 920B-450 20 R3000 96 2 CYBER 910B-621 25 R3000 128 2 CDC 4340 25 R3000 128 3 CDC 4360 25 R3000 128 3 CDC 4380 25 R3000 128 3 CDC 4680(Beta) 60 R6000 592 3 CYBER 920B-450 20 R3000 96 3 CYBER 910B-621 25 R3000 128 3 CDC 4680 60 R6000 592 4 CDC 4330 25 R3000 64 4 CDC 4380 25 R3000 128 4 DEC 6000/410 2 DEC DS2100 12 R2000 128 *1 DEC DS3100 16 R2000 128 *1 DEC DS5400 20 R3000 128 2 DEC DS2100 12 R2000 128 b DEC DS3100 16 R2000 128 b DEC DS2100 12 R2000 128 2 DEC DS3100 16 R2000 128 2 DEC DS5000/200 25 R3000 128 2 DEC VAX 4000/300 28ns VAX 256 3 DEC VAXvecto 6000/410 28ns VAX 1152 3 DG AV 310 20 DM88100 32 2 DG AV 5010 20 DM88100 32 2 DG AV 6200 25 DM88100 32 2 DG AV 410 20 DM88100 32 3 AV410 Workstation 20 DM88100 32 3 HP9000/340 16 DM68030 1 HP9000/370 33 DM68030 64 1 HP9000/834 15 PA 128 1 HP DN10010 18 PRISM 192 2 HP/Apollo 10000 18 PRISM 192 2 HP Apollo 10000 18 PRISM 192 4 IBM 6000/320 20 IBM2032 40 c IBM 6000/520 20 IBM2032 40 c IBM 6000/530 25 IBM2032 72 c IBM 6000/540 30 IBM3064 72 c IBM 6000/730 25 IBM2564 72 c IBM 6000/930 25 IBM2564 72 c IBM 6000/320 20 IBM2032 40 2 IBM 6000/520 20 IBM2032 40 2 IBM 6000/530 25 IBM2564 72 2 IBM 6000/540 30 IBM3064 72 2 IBM 6000/730 25 IBM2564 72 2 IBM 6000/930 25 IBM2564 72 2 Intel Star860 33 i860 12 3 Intel Star860 33 i860 12 4 MIPS RC2030 16 R2000 64 *1 MIPS RC2030 16 R2000 64 3 MIPS M/120 16 R2000 128 1 MIPS RC3240 25 R3000 128 2 MIPS M/2000 25 R3000 128 *1 MIPS RC3260 25 R3000 128 2 MIPS M/2000 25 R3000 128 2 Magnum 3000 color 25 R3000 64 3 Magnum 3000 mono 25 R3000 64 3 MIPS RC3240 25 R3000 128 3 MIPS RC3260 25 R3000 64 3 MIPS RC6280 60 R6000 592 3 MIPS M/2000 25 R3000 128 4 MIPS RC6280 60 R6000 592 4 MIPS RC3230 25 R3000 64 4 Motorola 8608 20 DM88100 32 *1 Moto 8864SP 20 DM88100 64 *1 Moto 8864SP 20 DM88100 128 2 Moto 8864SP 25 DM88100 128 2 Motorola 8612 33 DM88100 32 2 SGI 4D/25S 20 R3000 96 2 SGI 4D/210S 25 R3000 128 2 SGI 4D/320S 33 R3000 768 3 Solbrne 5/801 33 Cypress 128 2 Solbourne 5E/901 40 CY7C601 128 4 Stardent 3010 32 R3000 128 2 Stardent 3010 32 R3000 128 3 Sun SS1 20 LSIL 64 *1 Sun SS1 20 LSIL 64 2 Sun SS330 25 Cypress 128 *1 Sun SS330 25 Cypress 128 2 Sun SS490 33 Cypress 128 2 Sun SPARCstation IPC 25 LSIS1C0010(I64 4 Sun SPARCserver 490 33 CY7C601(IU)128 4 Sun SPARCstation 1+ 25 LSIS1C0010(I64 4 -- Robert E. Novak Mail Stop 5-10, MIPS Computer Systems, Inc. {ames,decwrl,pyramid}!mips!rnovak 950 DeGuigne Drive, Sunnyvale, CA 94086 rnovak@mips.COM (rnovak%mips.COM@ames.arc.nasa.gov) +1 408 524-7183
eachus@linus.mitre.org (Robert I. Eachus) (01/01/91)
A very good starting point. I hope, however, that you get Dhrystone 2.1 numbers--correctly measured--to compare to. A lot of (C) compilers were modified to special case some of the Dhrystone 1.1 compares, especially the string operations. While I do not fault the majority of compiler writers who did this in a way which provides better throughput for everyone (and also for Dhrystone 2.1), it is more appropriate to use Dhrystone 2.1 to catch those compilers which cheat. (It uses all results in subsequent operations and prints the final answers.) For Ada I find that Dhrystone 1.1 and 2.1 compare pretty exactly, since the effort of compiler writers has gone into optimizing procedure call overheads. -- Robert I. Eachus with STANDARD_DISCLAIMER; use STANDARD_DISCLAIMER; function MESSAGE (TEXT: in CLEVER_IDEAS) return BETTER_IDEAS is...