Newsgroups: comp.arch,comp.sys.super,comp.parallel,comp.sys.powerpc
From: 881045m (Richard Muise)
Reply-To: 881045m@dragon.acadiau.ca
Subject: Re: Is IBM planning to discontinue POWER2 development?
Organization: Acadia University
Date: Tue, 10 Jan 1995 18:52:02 GMT
Message-ID: <1995Jan10.183042.24360@relay.acadiau.ca>

krste@ICSI.Berkeley.EDU (Krste Asanovic) writes:

>I'm willing to wager that being 64b didn't help it's SPEC performance
>at all. In fact, if the same code was run as 64b (under some currently
>hypothetical 64b PowerPC OS) I'd also wager it would run measurably
>slower. Larger longs and pointers only make the memory hierarchy
>perform worse.

I don't think I actually said that 64-bits were responsible for the
speed up (the improved FP and Int units do, as well as the much faster clock).
However, I am rather curious now. Is there no speed from using 64-bits
rather than 32-bits? What if parts of the Spec benchmark code did something,
like a quad-word integer addition. Would not executing this code on the
64-bit chip be faster than on the 32-bit chip?


>The POWER2 does 2 mul-adds per cycle, or 2*2*71.5 = 286MFLOPS. The 620
>can do a FP mul-add every cycle, 2*133 = 266MFLOPS (I'm fairly sure
>this is true for double precision). This isn't a great difference,
>especially given the shorter absolute latency on the 620FPU (same
>number of clock ticks (?) but shorter cycle time).

I think IBM claims the fastest MFLOPS rating for the 71.5 Mhz machine was
266 MFLOPS. Why it is less than the projected maximum, I'm not sure. But
the SpecFp92 benchmark does show the PowerPC 620 as being faster than
the POWER2. So with only one unit, it does beat the POWER2 on at least
one benchmark.

And the PowerPC 620 can easily be increased in clock speed, while increasing
the POWER2 would seem to be more difficult, what with the board latencies
of having 9 chips to form one CPU.



-- 
Friends : Don't Let Friends Do DOS !	
	881045m @ dragon.acadiau.ca
	richard@admin.acadiau.ca
moderator of Comp.Parallel.    Comp.Parallel address: rmuise@dragon.acadiau.ca


