gnu@sun.uucp (John Gilmore) (12/04/84)
Quote without comment from Dtack Grounded, issue #37, Nov/Dec 84, pg 23: 68020 WATCH We have been guilty of overthusiasm in estimating the performance of the 68020. Motorola has now published a fairly detailed performance analysis. When they speak of a 4-1 speed improvement, they are talking about a 16.67MHz 68020 vs an 8MHz 68000, each with no wait state, and code optimized for the 68020. A no-wait-state 16.67MHz 68020 needs, according to Motorola's literature, 90 nsec SYSTEM access time on the RAMs - and that includes the delays of address multiplexers, decoders, buffers, and data buffers. Translation: 68020 systems will have wait states unless you have unlimited DOD-type dollars. What you and we are presumably interested in is the speed improvement using 150nsec DRAM and comparing the 16.67MHz 68020 with the 12.5MHz 68000 - and that looks to be about 2-1 even. The initial sample 68020s run at 12.5MHz so we are looking at less than 2-1. (These figures assume the use of a 32-bit data bus for the 68020.) MATH CHIP STUFF The 68881 has finally seen first silicon; now we will see if Motorola can break the existing record time of 2.5 years between first silicon and a working (production) complex F.P. math chip. We would like to be optimistic, but anybody who 'bets the company' on early arrival of production 68881s would be a fool, considering the historical record. (We DO think Motorola can beat that 2.5 years, but by how much?) --------- PERMISSION IS HEREBY granted to anyone whomever to make unlimited copies of any part or whole of this newsletter provided that a copy of THIS page, with its accompanying subscription information, is included. [Sorry, FNE -- I didn't retype the whole page, just the sub info --gnu] SUBSCRIPTIONS: Beginning with issue #19, subscriptions are $15 for 10 issues in the US and Canada (US funds) or $25 for 10 issues elsewhere. Make the check payable to DTACK GROUNDED. The address is: DTACK GROUNDED 1415 E. McFadden, Ste. F SANTA ANA CA 92705
davet@oakhill.UUCP (Dave Trissel) (12/10/84)
We have run several benchmarks now on an MC68020 plugged into an MC68010 socket. This means the MC68020 was running at the very same clock speed as the MC68010 (12.5 Mhz) and with the very same memory speeds (I think it was a one wait state memory for the '010.) The results have been about the same on all benchmarks run so far -- almost exactly double the speed (1/2 the time) and one-half the total number of bus cycles for the '020. Remember also that the '020 was running on only a 16-bit bus. Note that the benchmarks are the standard Pascal and C set with only the EDN quicksort and Berkley Puzzle being of significant length. This implies that the MC68020 on-chip cache probably made up the bulk of the improvement. It would be interesting to run UNIX System V (or any other OS for that matter) and see what results come up. (We are'nt planning that but several of our customers already have.) Concerning the MC68881, I believe samples are scheduled to go out this week. Due to the dedication and heroic efforts of our '881 people it appears that first silicon wafers are yeilding 95 to 98 percent bug-free parts. Sounds impossible, I admit, but soon you should have verification right here on the net as customers check out the parts for themselves. My understanding is that many early parts are running at 16 Megahertz. (I wish somebody on the '881 project would post here as they have a right to brag about thier accomplishment.) I am posting the MC68881 times in another message here for general use and information. I would also request anyone having any information on the Net (including Richard Mateosian of National) for equivalent transcendental times for the National NS32081 to please post them to me. We have not been able to determine any transcendental times for that chip since it does not have any on-board support for them. Motorola Semiconductor Dave Trissel Austin, Texas {ctvax|gatech|seismo|ihnp4}|ut-sally|oakhill|davet