Why so little parallelism?
|
From: BDH on 10 Nov 2006 20:24 > They'd probably wonder why anyone would bother with such a small > improvement. (Special purpose machines for those purposes have > been built many times.) > > http://www.google.com/search?sa=N&tab=gw&q=fft%20processor That is conservative, because I can't figure all the things that go into real world performance in my head. I suspect a factor of 100 is feasible for large problems. You may be underestimating modern GPUs, though. I am aware of FFT chips, but that is actually substantially more specialized, and they generally handle only 1024 points.
From: BDH on 10 Nov 2006 22:49 > That is conservative, because I can't figure all the things that go > into real world performance in my head. I suspect a factor of 100 is > feasible for large problems. You may be underestimating modern GPUs, > though. I am aware of FFT chips, but that is actually substantially > more specialized, and they generally handle only 1024 points. OK, I did a few numbers, and I'm fairly confident that feasible improvement over GPUs without loss of generality is 100-800x. I was overestimating the number of operation types necessary.
From: Brian Inglis on 10 Nov 2006 23:42 On Fri, 10 Nov 2006 22:37:57 +0800 in comp.arch, prep(a)prep.synonet.com wrote: >eugene(a)cse.ucsc.edu (Eugene Miya) writes: > >> The customer for the first RAMAC I thought was the Fort. Certainly >> close if not the first. CDC had a line of pretty drives resold as >> the RP06 and other models. > >No, the RP06 was a Memorex 677(?) with a massbus `DCL' wart on the >side. It was good enought that another company moved heaven and earth >to drive Memorex to the wall. > >CDC spat off their disk biz early by going into a join venture with >NCR called MAgnetic Peripherals. CDC disks meets NCR printers. That >was about the only thing NCR coud do near right... There's seems to be a plethora of cash registers around (here) with their logo on it, and a few ATMs. -- Thanks. Take care, Brian Inglis Calgary, Alberta, Canada Brian.Inglis(a)CSi.com (Brian[dot]Inglis{at}SystematicSW[dot]ab[dot]ca) fake address use address above to reply
From: Jim Hill on 11 Nov 2006 00:05 Bill Todd wrote: > Of course, there are plenty of people who would like to try to make the > discussion about politics [...] A personal favorite: It is a misfortune, inseparable from human affairs, that public measures are rarely investigated with that spirit of moderation which is essential to a just estimate of their real tendency to advance or obstruct the public good; and that this spirit is more apt to be diminished than promoted, by those occasions which require an unusual exercise of it. -- from the New York /Daily Advertiser/, 11 January 1788.
From: Nick Maclaren on 11 Nov 2006 06:27
In article <1163206926.507783.149890(a)f16g2000cwb.googlegroups.com>, "Andy Freeman" <anamax(a)earthlink.net> writes: |> BDH wrote: |> > |> > So here's another question - what would the reaction of capital sources |> > and companies be to some unknown tiny company with a working prototype |> > with, say, 10x speedup on matrix multiplies, FFTs, and sorting vs GPUs? |> |> They'd probably wonder why anyone would bother with such a small |> improvement. (Special purpose machines for those purposes have |> been built many times.) Grrk. That is true for matrix multiplication, but FFTs and sorting are memory access problems. IBM and others would be VERY interested in something that could do that ten times faster for an economical amount of hardware. Most special purpose systems that have done that for those tasks have handled only a few special cases. IBM did, after all, introduce a couple of new instructions into System/370 simply to speed up sorting by 30%. I doubt that they have lost interest in the problem :-) If BDH can deliver 100 to 800 times improvement for those, then he is clearly a sub-deity, or at least an alien from outer space. Matrix multiply speedup is a known, solved problem. Regards, Nick Maclaren. |