Subject: Re: Port benchmarks
To: Toru Nishimura <port-alpha@netbsd.org>
From: John Maier <johnam@kemper.org>
List: port-alpha
Date: 03/25/2000 23:46:15
So the chip is exactly the same as the 21064, but with some bad changes to
memory access.  Thus internally the 21066 can process instructions (int and
fp) just as fast as a 21064.  But there is a big performance penalty in
cache and memory access.

I would assume the 21066 isn't even pin compatable with the 21064, which
make a upgrade out of the question.

So basically the 21066 was big mistake by DEC.

jam

----- Original Message -----
From: Toru Nishimura <nisimura@itc.aist-nara.ac.jp>
To: <port-alpha@netbsd.org>; John A. Maier <johnam@kemper.org>
Cc: <nisimura@itc.aist-nara.ac.jp>
Sent: Sunday, March 26, 2000 05:24 AM
Subject: Re: Port benchmarks


>
> Curt Sampson <cjs@cynic.net> wrote;
>
> >> Why is the Alpha so much slower that the P-75?
> >
> > It's a 21066, not a 21064. The 21066 is very, very slow when
> > accessing the memory and second level cache, and this cripples its
> > performance.
>
> According to Dr. Bhandarkar book "Alpha -- Implementations and
> Architecture" a table on page 119,
>
> 21064 21066
> -------- ----------
> [ two have the same frabrication technoglogy ]
>
> Freq. 200MHz 166MHz
> SPECint92 138 76
> SPECfp92 200 87
> Linpack 100x100 Mflops 43 15
> Linpack 1000x1000 Mflops 156 55
> Livermore Loops Mflops 28 17
>
> The auther tells the addtional cost of external PCI chipset (21070)
> could be justified by performance gain, which means there is little
> win of 21066.  The chip was defeated by Pentium.
>
> Tohru Nishimura