jim@baroque.Stanford.EDU (James Helman) (03/31/91)
In article <1998@kuling.UUCP> irf@kuling writes: SPEC mark int fp HP9000/730,750 G/CRX 72.2 51.0 91.0 HP9000/720 G/CRX 55.5 39.0 70.2 In article <2005@kuling.UUCP> irf@kuling writes: gcc espr. li eqntott spice doduc nasa7 matrix fpppp tomcatv HP9000/730 46.5 55.2 50.3 52.6 60.9 64.0 73.7 273.3 107.0 67.4 HP9000/720 35.2 42.5 36.1 40.6 46.9 48.6 58.0 210.0 81.4 52.9 Somethings off. These 720 spec ratios yield 38.5 rather than 39.0 ISpecs. In article <32580006@hpcuhe.cup.hp.com> linley@hpcuhe.cup.hp.com writes : Although Snakes is not superscalar, PA-RISC instructions such as ADD AND BRANCH, MOVE AND BRANCH and COMPARE AND BRANCH allow a similar amount of parallelism as America for integer-only applications; in fact, the ratio of Integer SPECmarks to MHz for Snakes (65/66) actually exceeds America's (35/42). Putting clock frequencies and irf@kuling's recent postings into spec.sc (from perelandra.cms.udel.edu) bears this out in general, but MIPS is pretty close. Where did this 65/66 ratio come from? From irf's numbers, it appears that Snakes has a slightly lower ratio of ISpecs to MHz than the RS/6000. Why is SPARC so much lower than MIPS in terms of ISpecs/MHz? Int Float Specs/ ISpecs FSpecs/ Machine MHz Notes Specs Specs Specs MHz MHz MHz IBM_RS/6000-320 20.00 7 22.25 15.78 27.99 1.11 0.79 1.40 IBM_RS/6000-320 20.00 10 24.60 16.30 32.40 1.23 0.82 1.62 IBM_RS/6000-520 20.00 7 22.25 15.80 27.95 1.11 0.79 1.40 IBM_RS/6000-530 25.00 7 28.93 20.27 36.67 1.16 0.81 1.47 IBM_RS/6000-540 30.00 2,7 34.70 23.98 44.40 1.16 0.80 1.48 IBM_RS/6000-550??42.00 10 54.30 34.50 73.50 1.29 0.82 1.75 IBM_RS/6000-730 25.00 7 28.97 20.05 37.04 1.16 0.80 1.48 IBM_RS/6000-930 25.00 7 28.86 20.15 36.67 1.15 0.81 1.47 HP_9000/720 50.00 11 55.19 38.48 70.19 1.10 0.77 1.40 HP_9000/730 66.00 11 72.19 51.05 90.95 1.09 0.77 1.38 DEC_DS2100 12.50 1 7.46 8.69 6.73 0.60 0.70 0.54 DEC_DS2100 12.50 4,7 8.29 8.71 8.02 0.66 0.70 0.64 DEC_DS3100 16.67 4,7 11.27 11.78 10.94 0.68 0.71 0.66 DEC_DS3100 16.67 1 10.14 11.45 9.36 0.61 0.69 0.56 DEC_DS5000/200 25.00 4,7 18.55 19.10 18.20 0.74 0.76 0.73 DEC_DS5400 20.00 4,7 11.80 12.64 11.27 0.59 0.63 0.56 DEC_DS5810 25.00 4 11.31 12.84 10.40 0.45 0.51 0.42 MIPS_6280_Beta ??60.00 2,7 42.22 42.05 42.34 0.70 0.70 0.71 MIPS_M/120-5 16.67 1,7 11.20 12.94 10.17 0.67 0.78 0.61 MIPS_M/2000 25.00 1 16.46 19.55 14.67 0.66 0.78 0.59 MIPS_M/2000 25.00 7 17.58 19.69 16.30 0.70 0.79 0.65 MIPS_RC2030 16.67 1 9.26 11.18 8.17 0.56 0.67 0.49 MIPS_RC3240 25.00 7 16.06 17.60 15.10 0.64 0.70 0.60 MIPS_RC3260 25.00 7 17.26 19.27 16.03 0.69 0.77 0.64 MIPS_Rx2030 16.67 7 10.02 11.34 9.22 0.60 0.68 0.55 MIPS_Rx3230 ??25.00 7 17.78 19.29 16.84 0.71 0.77 0.67 SGI_4D/210S 25.00 7 13.97 16.79 12.36 0.56 0.67 0.49 SGI_4D/25S ??20.00 7 12.20 13.98 11.14 0.61 0.70 0.56 SGI_4D/320S 33.00 5 19.45 22.56 17.61 0.59 0.68 0.53 Solbourne_5/801 33.00 7 16.27 17.69 15.39 0.49 0.54 0.47 Sun_4/260 16.67 3 5.71 8.71 4.31 0.34 0.52 0.26 Sun_SS_SLC_est 20.00 7 7.70 9.52 6.68 0.39 0.48 0.33 Sun_SS1 20.00 1 8.26 9.58 7.49 0.41 0.48 0.37 Sun_SS1 20.00 3,7 8.43 9.56 7.75 0.42 0.48 0.39 Sun_SS1+ 25.00 7 10.01 11.22 9.28 0.40 0.45 0.37 Sun_IPC 25.00 10 11.80 12.40 11.40 0.47 0.50 0.46 Sun_SS330 25.00 1 10.89 12.25 10.08 0.44 0.49 0.40 Sun_SS330 25.00 3,7 11.82 12.22 11.56 0.47 0.49 0.46 Sun_SS490 33.00 3,7 17.60 18.62 16.96 0.53 0.56 0.51 Sun SS/2 40.00 8 21.01 20.24 21.53 0.53 0.51 0.54 (Caveats: spec.sc really needs to have redundancies removed to leave only the most recent numbers. The clock numbers came from various sources and could be wrong.) Notes 1. SPEC Newsletter, Volume 1, Issue 1, Fall 1989 2. 39283@mips.mips.COM (mash@mips.COM (John Mashey)) 3. SPEC Newsletter, Volume 2, Issue 1, Winter 1990 4. 'Digital's RISC Family Graphics and CPU Performance Summary' 5. 'Supercomputing Review', September, 1990 6. tom@hcx2.ssd.csd.harris.com (Tom Horsley) 7. Calculated from 'Your Mileage May Vary', John R. Mashey 8. Steven.Schlick@SAM.CS.CMU.EDU 9. Michael Z. Slater, 'Microprocessor Report' 10. Bo Thide, Uppsala U., comp.arch, <1998@kuling.UUCP> 27-Mar-91 11. Bo Thide, Uppsala U., comp.arch <2004@kuling.UUCP> 30-Mar-91 Jim Helman Department of Applied Physics Durand 012 Stanford University FAX: (415) 725-3377 (jim@KAOS.stanford.edu) Work: (415) 723-9127
linley@hpcuhe.cup.hp.com (Linley Gwennap) (04/02/91)
An error has been pointed out in my previous detailed posting on the Snakes processor. I would like to correct it. I apologize for the confusion. > Although Snakes is not superscalar, PA-RISC instructions such as ADD AND > BRANCH, MOVE AND BRANCH and COMPARE AND BRANCH allow a similar amount of > parallelism as America for integer-only applications; in fact, the ratio of > Integer SPECmarks to MHz for Snakes (65/66) actually exceeds America's > (35/42). In fact, the Series 730 produces about 50 Integer SPECmarks, not 65. Thus, the ISPEC/MHz ratio is about 75%, as opposed to about 83% on IBM's America. Still, by this measure, the Series 730 exceeds all non-superscalar systems that I am familiar with. --Linley Gwennap Hewlett-Packard
linley@hpcuhe.cup.hp.com (Linley Gwennap) (04/02/91)
DISCLAIMER: I have nothing to do with HP's marketing of the Series 700.
I personally hate the use of MIPS as a performance measurement. Particularly
in 12" high digits in the Wall Street Journal.
But I wouldn't hesitate to use $/SPECint as a performance measure, although
I would assume that most workstation users are doing at least some floating
point, or else they would be using PCs :-). Anyway, here is an amendment
to Bo's table. I hope I have the SPECint numbers right....
====================================================================
The HP 720: How It Stacks Up
COMPANY/PRODUCT PRICE SPEC SPEC Price Per Price Per
int marks SPECint SPECmark
Hewlett-Packard/ $12,000 39 55.5 $308 $216
HP 9000 Model 720
IBM/ $9,725 16.3 24.6 $597 $395
RISC System/6000
Model 320
Digital Equipment/ $12,500 20.5 19.9 $610 $628
DECstation 5000
Model 200 MX
Sun Microsystems/ $15,000 20.2 21 $743 $714
SPARCstation 2
====================================================================
John's right; the DECstation is about as good as the RS/6000 for
integer-only workloads. The Series 700 is only twice as good.
--Linley Gwennap
Hewlett-Packard