🚀 go-pugleaf

In article <1993Jan22.233827.11919@convex.com>, rosenkra@convex.com (William Rosenkranz) writes:
|>while SUNs might not be the way to go (SPARC is way behind the RISC curve),
|>i would put HP up against DEC any day. the PA-7100 chip can easily outperform
|>the alpha running at a much higher clock speed. considering that an under $50k
|>HP 735 turns in SPECfp92 of 150 while the $316k DEC 10000 (at 200 MHz) does
|>only 200 seems like the 21064 is a weak implementation of the basic Alpha
|>architecture (which does have some interesting aspects). HP is even better
|>with DP Linpack (40.8 MFLOPS vs the 10000's 42.5 or so). also the 1000x1000
|>Linpack for alpha is actually lower than HP's (based on DEC's numbers in
|>the november announcement). this tends to tell me that the tiny on-chip
|>alpha caches and high latency off-chip cache is weak. also the fact that
|>I/O busses off the chip run at most half the system clock speed does not
|>help much either. i would expect a 200 MHz chip to get at least 180 MFLOPS
|>on 1000x1000 linpack, if it were designed with technical (floating point)
|>computing in mind. i also think HP has a better graphics story than DEC,
|>though i am far from an "expert" in graphics.

Ooops, you bit off on the HP marketing hype. Yes the initial benchmarks
I have seen indicate that the HP outperforms the Alpha (but just edges
out the $35K DEC 3000/500). As DEC RISC compiler techinology for
the Alpha is still young. As the benchmarks using the KAP preprocessor
have pointed out, you can squeeze as much as a %40 speed increase out
of the Alpha. I predict you will seem some improvements on the 
benchmarks as DEC learns more about optimizing their code. HP has done
a great job, though, of getting more bang out of their architecture.

|>-bill
|>rosenkra@convex.com

How is Morgenstern? (and Hamlet?)
----------------------------------------------------------------------------
Dwight D. Moore
Geosciences Computing Network
University of Oklahoma

dwight@geohub.gcn.uoknor.edu (129.15.40.10)
These opinions do not necessarily represent the opinions of OU or the GCN.
----------------------------------------------------------------------------