mjg@ecsvax.UUCP (Michael Gingell) (10/31/85)
How fast is the Amiga ?. In the cause of Science I tried the Dhrystone benchmark published recently in net.sources. Running in CLI mode only and compiling with the current release of the Lattice compiler (no Enums) I found it executed 50000 Drystone loops in 110 secs. Thats approximately 450 Dhrystones/sec. But what happens if you are running other processes simultaneously ?. I tried running it with one of the graphics demos and the speed dropped to half. So I thought if I run 3 graphics demos and the benchmark the speed should halve again - right ? - Wrong. It stayed about the same. What is happening ?. For comparison here are some results from selected machines in the table given in the last Drystone posting : From: rer@vaximile.UUCP (R.RICHARDSON) * * MACHINE MICROPROCESSOR OPERATING COMPILER DHRYSTONES/SEC. * TYPE SYSTEM NO REG REGS * -------------------------- ------------ ----------- --------------- * IBM PC/XT 8088-4.77Mhz PCDOS 2.1 Lattice 2.15 403 - @ * PDP-11/34 - UNIX V7M cc 387 438 * ATT PC6300 8086-8Mhz MSDOS 2.11 b16cc 2.0 632 684 * IBM PC/AT 80286-6Mhz PCDOS 3.0 CI-C86 2.1 666 684 * Macintosh 68000-7.8Mhz 2M Mac Rom Mac C 32 bit int 694 704 * Macintosh 68000-7.7Mhz - MegaMax C 2.0 661 709 * Macintosh 68000-7.8Mhz 2M Mac Rom Mac C 16 bit int 877 909 S * IBM PC/AT 80286-6Mhz PCDOS 3.0 MS 3.0(small) 1063 1086 * IBM PC/AT 80286-6Mhz PCDOS 3.1 Lattice 2.15 1250 - @ * VAX 11/785 - UNIX 5.2 cc 2083 2083 * VAX 8600 - VMS VAX-11 C 2.0 7142 7142 * Amdahl 580 - UTS 5.0 Rel 1.2 cc Ver. 1.5 23076 23076 ----------------------------------------------------------------------------- Mike Gingell ....decvax!mcnc!ecsvax!mjg
hamilton@uiucuxc.CSO.UIUC.EDU (11/03/85)
hmmm i tried the same dhrystone benchmark on my amiga and got:
~368 no register variables
~371 register variables
~300 while running dots, lines & boxes at same time
~100 while running mandrille at same time
i think i was using the 3.02 Lattice compiler, operating from the CLI.
i'm hoping either 1) the Lattice compiler is very crude, or 2) somehow
i transcribed the benchmark incorrectly...
wayne hamilton
U of Il and US Army Corps of Engineers CERL
UUCP: {ihnp4,pur-ee,convex}!uiucdcs!uiucuxc!hamilton
ARPA: hamilton@uiucuxc.cso.uiuc.edu
CSNET: hamilton%uiucuxc@uiuc.csnet
USMail: Box 476, Urbana, IL 61801
Phone: (217)333-8703nigel@minster.UUCP (nigel) (11/04/85)
In article <663@ecsvax.UUCP> mjg@ecsvax.UUCP (Michael Gingell) writes: > > > >How fast is the Amiga ?. In the cause of Science I tried the Dhrystone >benchmark published recently in net.sources. Running in CLI mode only >and compiling with the current release of the Lattice compiler (no >Enums) I found it executed 50000 Drystone loops in 110 secs. Thats >approximately 450 Dhrystones/sec. > Ok - so we ran it on an Atari 520ST ... 1092 Dhrystones/sec. This is, of course, independent of graphic mode, but if you wave the mouse about it slows down a bit. Nigel Roles P.S. The benchmark is wrong. malloc() should be declared as char *.
freed@aum.UUCP (Erik Freed) (11/06/85)
> >How fast is the Amiga ?. In the cause of Science I tried the Dhrystone > >benchmark published recently in net.sources. Running in CLI mode only > >and compiling with the current release of the Lattice compiler (no > >Enums) I found it executed 50000 Drystone loops in 110 secs. Thats > >approximately 450 Dhrystones/sec. > > Ok - so we ran it on an Atari 520ST ... 1092 Dhrystones/sec. This is, > of course, independent of graphic mode, but if you wave the mouse about > it slows down a bit. > Could we have a second comfirmation from another ST user this sounds amazing! Also could we find out more details about the 1092 drystone run? -- ------------------------------------------------------------------------------- Erik James Freed Aurora Systems San Francisco, CA {dual,ptsfa}!aum!freed
mjg@ecsvax.UUCP (Michael Gingell) (11/08/85)
> > >How fast is the Amiga ?. In the cause of Science I tried the Dhrystone > > >benchmark published recently in net.sources. Running in CLI mode only > > >and compiling with the current release of the Lattice compiler (no > > >Enums) I found it executed 50000 Drystone loops in 110 secs. Thats > > >approximately 450 Dhrystones/sec. > > > > Ok - so we ran it on an Atari 520ST ... 1092 Dhrystones/sec. This is, > > of course, independent of graphic mode, but if you wave the mouse about > > it slows down a bit. > > I should point out that the original 450 on the amiga was using the regular Lattice Compiler which uses 64 bit floating point arithemetic I believe. If I get time I will try it using the 32 bit routines in the Rom kernel. However these comparisons should be taken with a pinch of salt as it is compiler dependant - for all we know the Atari number could be for 16 bit integers. Of course, waving the mouse round on the Amiga makes no difference, you can run the disk drive, and the sound and have sprites going without the 68000 losing any of it's bus bandwidth. Mike Gingell ...decvax!mcnc!ecsvax!mjg
nigel@minster.UUCP (nigel) (11/11/85)
In article <398@aum.UUCP> freed@aum.UUCP (Erik Freed) writes: ... >Could we have a second comfirmation from another ST user this sounds >amazing! Also could we find out more details about the 1092 drystone run? ... Sorry - I should have given more details, but then again so should the original posting! What is important is whether 16 or 32 bit arithmetic is used, and it is not clear (i.e. I don't know) what the Lattice C compiler uses on the Amiga. The following table benchmarks the ISV development kit (i.e. the DR C compiler), with structure assignment, but no enums. regs noregs 16 bit 1092 1070 32 bit 710 ???? The 32 bit figure was arrived at by changing all integer references to long in the benchmark. Timing was done using the system ticker, and is therefore accurate and repeatable. In addition, the 1092 figure drops to the region of 850 if the cursor is displayed and continuously tracked around the screen during the test. I'm sure other people can confirm these numbers. Similar tests and details of the Amiga test would be very interesting, as even if the Amiga was using 32 bit arithmetic, 450 d/s (or as rumoured around 300 for the Hippo compiler) does not compare well. If, as a recent posting indicated, there would have been no coprocessor contention during the published test, what's going on?
preece@ccvaxa.UUCP (11/13/85)
> I should point out that the original 450 on the amiga was using the > regular Lattice Compiler which uses 64 bit floating point arithemetic I > believe. If I get time I will try it using the 32 bit routines in the > Rom kernel. However these comparisons should be taken with a pinch of > salt as it is compiler dependant - for all we know the Atari number > could be for 16 bit integers. /* Written 11:10 am Nov 8, 1985 by > mjg@ecsvax.UUCP in ccvaxa:net.micro.amiga */ ---------- The dhrystone benchmark does not use floating point, so the floating point in the Lattice compiler shouldn't be relevant. -- scott preece gould/csd - urbana ihnp4!uiucdcs!ccvaxa!preece
dale@amiga.UUCP (Dale Luck) (11/14/85)
I ran the Drystone benchmark using the two different compilers I have available. I can collaborate you lattice-c experience, mine came in at around 450/sec. However running the exact same code through the greenhills compiler that is on the sun for cross development not only produced more compact code but also turned a figure of 850 dhrystones/sec. These both are 32bit integer compilers. Let's here for optimizing compilers!!! Dale Luck
brownc@utah-cs.UUCP (Eric C. Brown) (11/14/85)
In article <719@ecsvax.UUCP> mjg@ecsvax.UUCP (Michael Gingell) writes: >> > >How fast is the Amiga ?. In the cause of Science I tried the Dhrystone >> > >benchmark published recently in net.sources. >I should point out that the original 450 on the amiga was using >the regular Lattice Compiler which uses 64 bit floating point >arithemetic I believe. But there is NO floating point in the Dhrystone Benchmark!! It simply tests Integer arithmetic, Procedure Calls, Array Indexing, etc. Eric C. Brown brownc@utah-cs ...!ihnp4!utah-cs!brownc
tim@ISM780B.UUCP (11/16/85)
/* Written 7:03 pm Nov 13, 1985 by dale@amiga in net.micro.amig */ available. I can collaborate you lattice-c experience, mine came in at around 450/sec. However running the exact same code through the greenhills compiler that is on the sun for cross development not only produced more compact code but also turned a figure of 850 dhrystones/sec. These both are 32bit integer compilers. Let's here for optimizing compilers!!! /* End of text from ISM780B:net.micro.amig */ While the Green Hills compilers are good, I think these results say more about the Lattic C compiler than the Green Hills compiler.