Date: Thu, 18 May 1995 01:34:42 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: phk@ref.tfs.com (Poul-Henning Kamp) Cc: FreeBSD-hackers@FreeBSD.org (FreeBSD hackers) Subject: Re: Adaptec 2940? Message-ID: <199505180834.BAA13812@gndrsh.aac.dev.com> In-Reply-To: <199505180618.XAA28462@ref.tfs.com> from "Poul-Henning Kamp" at May 17, 95 11:18:24 pm
next in thread | previous in thread | raw e-mail | index | archive | help
>
> > > The drivers are identical, so the interupt time should be the same for
> > > driving either card on the same machine. Your benchmark is not really
> > > valid since they were run on different motherboards.
> >
> > To some extent. It is instesting that a good EISA system can best a
> > poor PCI system. Woe to those buying cheap PCI motherboards.
>
> Oh, it's much worse than that. This is some numbers I have been collecting,
> the represent the bandwidth between CPU and RAM pretty well:
>
> First line is read dominated, second is write dominated:
>
> My PC, en 486DX2/66 32Mb, FreeBSD-2.1, GCC 2.6.3
> flagmose# cc -o ram-speed -O2 ram-speed.c
> flagmose# ./ram-speed
> 49005fb0 0.615 uS/op 1.63e+06 op/S 6.201 Mb/S
> 8938c0df 0.159 uS/op 6.29e+06 op/S 23.981 Mb/S
>
> A "IBM Valuepoint 486DX2/D PC", as above
> myname# cc -o ram-speed -O2 ram-speed.c
> myname# ./ram-speed
> 49005fb0 0.377 uS/op 2.65e+06 op/S 10.116 Mb/S
> 8938c0df 0.150 uS/op 6.66e+06 op/S 25.423 Mb/S
>
> A ASUS P90 Neptune, FreeBSD-2.1, GCC 2.6.3
> phk@time.cdrom.com cc -o ram-speed -O2 ram-speed.c
> phk@time.cdrom.com ./ram-speed
> 49005fb0 0.356 uS/op 2.81e+06 op/S 10.723 Mb/S
> 8938c0df 0.209 uS/op 4.78e+06 op/S 18.235 Mb/S
> phk@time.cdrom.com uptime
> 4:12PM up 4 days, 4:05, 4 users, load averages: 0.54, 0.27, 0.11
> Time wasn't idle, so these numbers are not 100% reliable
Add these to your list:
A ASUS P90 Neptune, FreeBSD-2.1, GCC 2.6.3
hookturn:rgrimes {1002} cc -o ram-speed -O2 ram-speed.c
hookturn:rgrimes {1003} ./ram-speed
49005fb0 0.334 uS/op 2.99e+06 op/S 11.407 Mb/S
8938c0df 0.174 uS/op 5.76e+06 op/S 21.964 Mb/S
hookturn:rgrimes {1004}
Box was idle.
A ASUS P90 Triton, 15ns Async SRAM Cache, FreeBSD-2.1, GCC 2.6.3
[Bios on default autoconfig settings, I can make it go faster :-)]
gndrsh# cc -o ram-speed -O2 ram-speed.c
gndrsh# ./ram-speed
49005fb0 0.303 uS/op 3.30e+06 op/S 12.578 Mb/S
8938c0df 0.124 uS/op 8.09e+06 op/S 30.852 Mb/S
gndrsh#
I'll get EDO numbers to you tomarrow, I am waiting on more MB to
come in since I got cleaned out :-(.
See, 10 to 30% faster than the Neptune chip set, and that is with
out Pipeline Burst Cache!!!
> Here is the source-code, Bruce and Rod use a different piece of code,
> which is probably better, but this is what I have used all the time,
> so I can compare my own results:
Not necessarly better, it measures different things, sequential memory
read and memory write with cache aligned data vs what ever pattern of
access your program uses. (I am use to your older code that had
sequential read/write, random read/write, etc...).
--
Rod Grimes rgrimes@gndrsh.aac.dev.com
Accurate Automation Company Custom computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199505180834.BAA13812>