[RC5] Speed

Roby Van Hoye roby.vanhoye at ficsgrp.com
Tue Apr 21 10:33:11 EDT 1998


At 19:39 20/04/98 -0700, you wrote:
>On Mon, 20 Apr 1998, Bojan Baros wrote:
>> On Mon, 20 Apr 1998, Dan Sugalski wrote:
>> 
>> > At 11:35 AM 4/20/1998 -0400, Bojan Baros wrote:
>> > >Some of you will remember that a month or so ago I asked the question
of 
>> > >Flops that D.Net has.  I got my answer today.  Combined MFlops are : 
>> > >2959855 or at least that is what it says on the stats page (thumbs up
for 
>> > >new look).  That I think officially makes us the fastest computer on
the 
>> > >net.
>> > >	Also, is it possible that someone from the admin write a letter to 
>> > >the magazine byte and tell them that they are wrong (the said that 
>> > >fastest computer has broken 1 terraflops).
>> > 
>> > Except, of course, we haven't broken any flops mark. Flops is short for
>> > Floating point Operations. RC5 and DES are both integer-only
calculations,
>> > so there's no floating point involved at all. (Not that I doubt we'd kill
>> > that terraflop number if there were, but our current numbers don't have
>> > anything to do with it)
>> > 
>> 
>> And I was saying, that is a calculated value.  It is calculated from 
>> taking some refference processor (in X86 case it was P100), calculating 
>> how many of those processors would be doing current load of X86es, and 
>> than multiplying it with the number of known FLOPS of P100.  That was 
>> done by D.Net staff (http://www.distributed.net/statistics/stats.html.
>> I know that D.Net processing does NOT involve fp ops to be measured...
>
>You can't scale the numbers at all meaningfully, though. I've got a couple
>of Alpha CPUs involved, and their numbers, relatively speaking, suck. The
>speed of a CPU for RC5 purposes is almost entirely based on the speed
>that a particular CPU can shift data within a register. It has zilch to do
>with floating-point performance.

Even better : have a look at the 68k clients, a 68040 has no bitshifting
and therefore did a very poor job on theDES challenge (don't know if this
implies to RC5 too) but on FPU  operations it, relatively spoken that is,
surely beats your P100.

>> Can you at least check out the info before you start flaming people!!!!!!
>
>Um, if you check, you'll see it's not a flame. You just made an incorrect
>statement based on an incorrect assumption. I'm quite sure of my
>information. 

I'll second that, I guess there's only one way to find out the GFLOPS value
of D.net and that's by adding an FPU intensive test into the clients... 

>The big thing you can extrapolate from the RC5 numbers is some aggregate
>MegaShift/sec number, which is fairly meaningless.

Well, I wouldn't call it meaningless, if we had the bandwith we could be
doing realtime rendering solely on integer code with this 'monster'. :-)





Cu
Roby.
--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list