RC5 Pre-release client optimized for P4 ?

[Slacker]

[May 02 01:54:24 UTC]
RC5: Benchmark for core #8 (AK class 7) 0.00:00:17.01 [4,688,545 keys/sec] :Q

This is from my P4 1.6a running at 2320mhz , dont know if it matters but 145fsb and PC2700 mem at 382mhz ddr

The Throughput meter thing is showing a steady 4.4KKS <<thats good, no?

Here is a link to the thread where I found out about the Pre-release client http://subscriber.anandtech.com/messageview.cfm?catid=39&threadid=768727And a link to the Dnet page for the client
http://www.distributed.net/download/prerelease.html

 

[Orange Kid]

Thats good for a P4........about twice what it used to be 🙂

 

[Jay]

Took em long enough. Oh wait, I don't do RC5 anymore and I don't own a P4 machine. Nevermind, ignore me. 😉

 

[Slacker]

(Jay) Oh wait, I don't do RC5 anymore

does that mean that I win the TA NetZero.Net marathon? 😀

 

[Jay]

Well, I'm no longer under the NetZero e-mail addy, so I think you have the lead there. I'm under jay@jator.net if you want to compare stats though. 😛

 

[SpaceWalker]

Yeah, I posted about it back on the 20th...

http://forums.anandtech.com/messageview.cfm?catid=39&threadid=768727

 

[vss1980]

I wonder if they are making use of the P4's SSE2 Integer SIMD functions to get the extra performance. Could be semi-interesting, as if that is a SSE2 optimised core, it could mean less work and time needed to modify the core to get the best performance from newer CPU's as they come out.

 

[ViRGE]

Vss, Moose has said before that there's not much that can be done with SSE 1 or 2. Even though SSE2 features 128 bit ops, for example, it doesn't have the right 128 bit registers to do bitslicing like the G4 can. I'm sure they're using some optimizations there, but most likely, they're just trying to make up for slow(or is it nonexistant? BK?) rotate functions. Then again, they thought they couldn't do anything with MMX either, but they did, and managed to make the PMMX a formidable chip.

 

[sciencewhiz]

It's just slower (4 cycles as opposed to 1, IIRC). If you notice, the old values for the P4 was about half the kkeys/mhz of the P3. When you consider the double clocked ALU and the 4x slower rotate, it makes sense.

With the new core, it appears to be at least 30% faster. While still slow, it is more respectable.