[Shaktai]

Reader50 already posted the link on the RC-5 page, but for those of you who are crunching on RC-5, there is a new optimized version. http://www.distributed.net/download/prerelease.html It is altivec enabled and from reports gives about a 4x performance boost on G4's. This latest prerelease version also seems to boost G3 performance as well.

While we really need help on projects like Folding@Home or distributed folding, if you are going to crunch on RC5, you might as well use the fastest client. Get it NOW!


UPDATE: Here are benchmark comparisions between the original Mac client that used only core #0 and the new client core #4 on an iBook 600 with G3. A big improvement.

RC5-72: Benchmark for core #0 (ANSI 4-pipe)
0.00:00:16.54 [1,183,761 keys/sec]
RC5-72: Benchmark for core #4 (KKS 2-pipe)
0.00:00:16.20 [1,814,733 keys/sec]

Maybe I'll benchmark my PowerMac 800 after it finishes with its current 73 point Folding@Home protein unless someone else posts G4 benchmarks first.

[Jsnuff1]

Wow these G4 cores are more optimized than rc5-64. These are benchmarks of my 500mhz 7400
Core #0 930,436 keys/s
Core #6 4,320,428 keys/s

Since this system is dual it will be doing about 8.5 mkeys/s, hehe my friends 3ghz p4 tops out at 3 mkeys/s.

[chris v]

Cool. Now, maybe we can exhaust the keyspace in only 1000 years!

Downloading now.

CV

[chris v]

Cube 450 = 3,672,964 keys/s

Was 900,000 +/- before.

Much better. I'll get the dual gig on it on Monday.

CV

[Shaktai]

Well curiosity got the better of me. I stopped my 73 point protein just long enough to benchmark.

PM 800 G4 single w/256k L2 and no L3

Benchmark for core #0 (ANSI 4-pipe)
1,684,191 keys/sec

Benchmark for core #7 (KKS 7450)
7,865,735 keys/sec

There are seperate cores optimized for different G4 processors. For instance, #6 is optimized for the 7400.

[Toast]

This is great to hear, with the new core we should be able to move up pretty quickly vs the other teams...

For comparison:
My AMD Duron 850@912 gets 2,356,285 keys/sec

How about a new core for the AMD?

[reader50]

This refers to the prerelease clients, check the release candidate version numbers to see if these changes have been included. Or just download the prerelease clients.

To post the more complete recent changes, which might be of interest:

2.9003-481 fix: smp via fork: 2nd process died after switching to project
ogr (added workaround for #3006)
chg: ppc: use OGR-scalar cruncher for PowerPC 7450 (#3035)
fix: ppc: manual selection of OGR cores is now possible
new: macosx: CPU temperature reading by Karakoussis Apostolos
new: ppc: two additional rc5-72 cores for ppc by Didier Levet

2.9003-480 new: x86: rc5-72 core by Gnatiuc Ianos (Athlon optimized)
chg: all: -numcpu 0 disables use of shared memory (if any)
fix: macosx: changed -install to work on Mac OS X 10.2 (#1618)

2.9002-479 new: ppc: rc5-72 core by Malcolm Howell (603e optimized)
new: ppc: two rc5-72 cores by Didier Levet (AltiVec optimized)
new: amigaos: optimized 68k rc5-72 cores by Malcolm Howell
fix: amigaos: PowerUp compiler linklib no longer trashes memory
imp: amigaos: added G3/G4 cpu detection

[chris v]

Cube 450 in >console mode:
3,848,537 keys/sec. Comperable to the RC5-64 client on this machine.

The GUI overhead of 10.2 is responsible for a 200,000 key/sec loss in power. Console mode overnight or on work machines over the weekend could make up a lot of work.

CV

[Shaktai]

Well it looks like a few folks took my suggestion, as the small RC-5 team saw a 50% boost in production and was #32 for the day. Unfortunately it looks like xlr8yourmac.com got wind of it sooner. They just flew past our team and was #8 on the day yesterday, actually knocking Team MacNN down a notch overall. Still it looks like our guys will probably gain that back in a day or two when they pass German OC-Maniacs und Hardwarefreaks. Hopefully those few who are crunching on RC-5 will all convert to the latest version. My power is staying with Folding@home for now.

Remember, it is the "pre-release" version that is needed to get the most bang for your Mac.
http://www.distributed.net/download/prerelease.html

[Shaktai]

Always curious, I decided to benchmark my Athlon 2400 to see how my Macs compare.

Athlon 2400 (2.0 ghz)
Benchmark for core #5 (DG 2-pipe)--6,100,555 keys/sec

iBook 600 (G3)
Benchmark for core #4 (KKS 2-pipe) --1,814,733 keys/sec


PM 800 G4 single w/256k L2 and no L3
Benchmark for core #7 (KKS 7450) --7,865,735 keys/sec

[chris v]

My dual gig is now doing 18.2 on average, and had only been geting 4.8 or so with the Dec. 3 client. Looks like about a 4-fold increase in productivity with the altivec core.

Calling Gary L. Grey!

I wonder what happened to the guy who had the lab at Harlingen going last year? He hasn't turned up this go-round.

CV

[OpenStep]

Wow, i installed it on my G4's and its running a LOT faster. The new core makes a huge difference

[Shaktai]

Found an interesting tidbit tucked away in the well known official IBM pdf regarding the 970 chip. Page 13 listing performance. Down at the bottom under "Additional Performance" for the 1.8 ghz chip they listed:

RC-5 "estimated" performance at 18M keys/sec.
Peak scalar GFlops = 7.2
Peak SIMD GFLOPs = 14.4

Wonder what client version and OS as it is dated October 2002. Just found it a bit interesting that they used RC-5 as a measure. Not sure what to compare it against though.

[chris v]

Okay, I don't know if it's just me, but every time I go back to crunching RC5, stats break within a day or two. Sorry to the rest of you if it's me breaking their stats (murphy's law sort of thing), but I'm back on SETI until it's fixed, anyway.

CV

[Shaktai]

Originally posted by chris v:
Okay, I don't know if it's just me, but every time I go back to crunching RC5, stats break within a day or two. Sorry to the rest of you if it's me breaking their stats (murphy's law sort of thing), but I'm back on SETI until it's fixed, anyway.

CV

Seems to me that I recall they have long had that problem, even with the last set. Stats reporting just goes down for a day or two every once in a while. Doubt you have anything to do with it other then bad timing.

[Shaktai]

Kind of a boring rainy day, so I was poking around the various DC websites and found an interesting GUI utility for RC5.

http://casbahsoft.com/products.html

Might be helpful for those who aren't comfortable with the terminal. Looks like it would be pretty easy and straightforward to use, but I haven't tried it.

[Shaktai]

Some basic Mac team comparision stats.

--Team MacNN is #18 for yesterday and #52 overall and still climbing due to the new altivec optimized client.

--xlr8yourmac.com is #5 for yesterday and and #39 overall

--Team Evangelista (Macs Rule!) is still the top Mac team, at #21 overall and was #16 for yesterday. Looks like xlr8your mac.com will pass them before long though.

[macdude22]

RC5-72: using core #7 (KKS 7450)

0.17:34:14.22 - [5,750,636 keys/s]


I'm using an emac 700, is this good? I just got my Ibook 800 and it just started crunching I will post when it finishes. I'm not so familiar with RC5, I've always been doing seti. I'm not sure if this is a good rate or not. Thanks.



-The RadioSpy-

[macdude22]

The iBook 800's stats are

RC5-72: using core #4 (KKS 2-pipe)

[2,372,388 keys/s]

[Shaktai]

Originally posted by macdude22:
RC5-72: using core #7 (KKS 7450)
0.17:34:14.22 - [5,750,636 keys/s]

I'm using an emac 700, is this good?

Not sure. My PowerMac 800 does around 7,865,735 keys/sec. The best way to benchmark is to use the benchmark function built into RC5.

Start the program with " ./dnetc -bench" This will run a series of benchmarks for each core, and will only take a couple of minutes. The numbers produced will provide a good example of what you can expect for maximum performance. When you actually run the program, it automatically selects the fastest core.

Originally posted by macdude22:
The iBook 800's stats are

RC5-72: using core #4 (KKS 2-pipe)

[2,372,388 keys/s]

Those looks about right when compared to my iBook 600 scores posted above.

[macdude22]

For reference my athlon XP 1700, with a clock speed of a little over twice my emac is crunching away at

[4,380,228 keys/s]

a quite amount less than the emac. An interesting point about how efficient the G4 can be.