davis@unc.UUCP (09/30/87)
In article <5270@jade.BERKELEY.EDU>, mwm@eris.BERKELEY.EDU (MikeMeyer) writes: > ...I suspect > going to multi-chip cpus lets you get the same speed with less heat. The pad drivers for high speed integrated circuits use a significant amount of the power supplied to the chip. Although a multi-chip cpu would have more cooling area, it also would have a lot more power hungry pad drivers. I suspect Cray's conservative approach to implementation hardware is more responsible for the selection of multi-chip cpu's than cooling considerations. Mark (davis@cs.unc.edu)
lamaster@pioneer.arpa (Hugh LaMaster) (09/30/87)
In article <1449@unc.cs.unc.edu> davis@unc.cs.unc.edu (Mark Davis) writes: >hungry pad drivers. I suspect Cray's conservative approach to >implementation hardware is more responsible for the selection of >multi-chip cpu's than cooling considerations. Actually, cooling considerations have ALWAYS been one of the primary tradeoffs in supercomputer design. Seymour Cray is not necessarily the world's best "architect", but he sure has been the best at packaging and plumbing. If you don't believe it, ask some CDC engineers how much fun they had in trying to package the original STAR without him. Or why his machines are always smaller, and have faster clocks, than other similar complexity (gate count, etc.) machines from other designers. Current Cray technology is a 250MHz processor. Try clocking your 68020 or MIPS machine at that speed and see how long it takes to melt. Hugh LaMaster, m/s 233-9, UUCP {topaz,lll-crg,ucbvax}! NASA Ames Research Center ames!pioneer!lamaster Moffett Field, CA 94035 ARPA lamaster@ames-pioneer.arpa Phone: (415)694-6117 ARPA lamaster@pioneer.arc.nasa.gov (Disclaimer: "All opinions solely the author's responsibility")
mark@mips.UUCP (Mark G. Johnson) (10/01/87)
In article <2934@ames.arpa>, lamaster@pioneer.arpa (Hugh LaMaster) writes > Current Cray technology is a 250MHz processor. Try clocking > your 68020 or MIPS machine at that speed and see how long it > takes to melt. In the MIPS CPU case, about 3/4 of its 2W power dissipation (at 60 ns cycle) is due to switching (power = C*V*V*f). However, assume the worst, that ALL power consumption is due to switching transients. Then you just scale by (60ns / 4ns) and find that its power at 250 MHz would be 30 Watts. Since it's now a Cray-class device, it gets to use a Cray-sized 50 ton refrigeration unit, so 30 Watts should be no problem :-). Of course, we'd drop to a power supply of 3 volts, in which case the C*V*V*f power consumption would fall by (3/5) squared, giving 10.8 Watts. This power level is routinely acccomodated today, in air-cooled conventional packaging (for example in the larger ECL gate arrays). -- -Mark Johnson *** DISCLAIMER: The opinions above are personal. *** UUCP: {decvax,ucbvax,ihnp4}!decwrl!mips!mark TEL: 408-720-1700 x208 US mail: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
fouts@orville.nas.nasa.gov (Marty Fouts) (10/02/87)
In article <732@obiwan.UUCP> mark@mips.UUCP (Mark G. Johnson) writes: >In article <2934@ames.arpa>, lamaster@pioneer.arpa (Hugh LaMaster) writes > > Current Cray technology is a 250MHz processor. Try clocking > > your 68020 or MIPS machine at that speed and see how long it > > takes to melt. > >In the MIPS CPU case, about 3/4 of its 2W power dissipation (at 60 ns >cycle) is due to switching (power = C*V*V*f). However, assume the worst, >that ALL power consumption is due to switching transients. Then you >just scale by (60ns / 4ns) and find that its power at 250 MHz would >be 30 Watts. Since it's now a Cray-class device, it gets to use a >Cray-sized 50 ton refrigeration unit, so 30 Watts should be no problem :-). > Shouldn't you be taking the increase in volume of circuitry into account? After all, MY Cray 2 has 2 gigabytes of memory in the same cabinetry as 4 CPUs. The total effect that the added part count has on power dissipation through can be estimated by multiplying 30 watts by 120000 (the number of parts in a Cray 2 CPU) and dividing by whatever the number of components in a MIPS CPU case are. Since I don't know this number I'll just divide by 100 and come up with 36 KWatts, which is slightly higher than your estimate.