[rc5] Speed freeks with Cyrix, AMD, PPro or RISC machines ?

Remi Guyomarch rguyom at mail.dotcom.fr
Tue Aug 19 02:01:49 EDT 1997


Marc Sissom wrote:

[...]
> Have they(bovine) contacted you yet?

Yes, I'm now working on the 2.x series.

> At 08:14 PM 7/12/97 +0200, you wrote:
> >
> >Now the main cracking routine checks two keys at once. I hope the C
> >version will be significantly faster on superscalar RISC-Unix machines.
> 
> Have you gotten any reports on the c version and sparcs?
> 
> >I've also made separate ASM version for 486, Pentium, Cyrix 6x86 and
> >AMD-K5. I will include the PPro ASM version later.
> 
> Please? We have several ppros here in the office and your version 2
> 486 or p5 code is still the fastest for them(except for the cyberian
> client). The bovine v2.4 is faster(except for cyberian) on our p5 machines.

Yes, we have improved the P5 code in 2.004, and it should be faster than
the current (3.04) cyberian client, on a P5. On a PP200 / NT4 the 2.004
P5 code runs at ~263 kkeys/s and the cyberian client runs at ~245
kkeys/s (iddle machine at night, cheking real blocks).

The cyberian P6 code is currently faster than ours, but we will have a
faster P6 core in 2.005 (*don't ask me 'when').

> >Since I can't test the Cyrix and AMD version on real Cyrix and AMD
> >processor, I can't tell for sure if these clients are faster or not. But
> >I hope a *significant* speedup on these processor, compared to numbers
> >posted on the RC5 client speed page.
> 
> At home the 6x86 pr166 gets 270K or better with your code under linux.
> If you want me to beta-test on a 6x86, my home email is marcus at dfwmm.net
> (that's our team name as well).

I've tried to improve both the 6x86 code and the K5 code in 2.004, and
it should be faster than my old v1-based code. I just checked the client
speed page and it seems that 2.004 is checking ~280 kkeys/s under Linux.

-- 
Rémi
----
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.



More information about the rc5 mailing list