New Optimized Seti@Home apps

[Assimilator1]

Originally posted by: dajeepster
my 3000+ is a S939.... i bought it a week before the X2 were launched.

Unlucky, must of been one of the earlier (90nm?) cores kicking about.

 

[bryanW1995]

yes, that's extremely unlucky b/c the v8.0 app doesn't work with sse2 or older.

 

[dajeepster]

Originally posted by: bryanW1995
yes, that's extremely unlucky b/c the v8.0 app doesn't work with sse2 or older.

well... I guess i'll just have to turn it into a hardware firewall (it only gets @250ppd anyways).. and build another computer to take it's place, like with the 2 E5405, and the Asus "skulltrail" MB wanna be... which will be delivered on monday.

 

[bryanW1995]

:envy;

 

[Assimilator1]

lol

Doesn't v8.0 include SSE2? or is that in the std SETI client anyway?

Dajeepster
I look forwards to your numbers from that!

[edit]
I never mentioned the WU times I'm getting with my quad on v8.0.
At 3GHz it's doing most WUs in about 1hr 5mins, awesome!

No idea what it was doing with other clients as it had been nearly 6mths since this rig last ran SETI & I can't remember what times it was getting.

 

[bryanW1995]

it's hard to measure performance just by wu time. RAC is much more effective. Speaking of RAC, I FINALLY got all of my quads up to speed a little while ago...should be over 20k in a few days

 

[Assimilator1]

For the short to medium term RAC is useless for measuring output as it takes so long to match actual output, this is one thing I hate about SETI2 ,trying to compare performance of any rig in the near term (upto 1month ish?) is almost impossible.

What we need to do, as was done for SETI1 is to 'capture' a 'typical' WU & use that to benchmark SETI2. Then we can all benchmark rigs quickly, accurately & easily. It would be ideal for comparing new clients & rig tweaks etc too.
AFAIK no WU has been captured for SETI2, though admittedly I haven't done a thorough search about that in some time [edit] just done a quick ish one & nothing turns up.
For SETI1 the most common WU was one with an AR of 0.417 ,I've no idea what it is for SETI2 nor do I know how to find out (I don't mean just WUs on my own PC), any ideas?

BTW alot of the WUs on my rig that I saw that were done in ~1hr 6mins were 50 credit ones.

Awesome output you've got BTW .

 

[bryanW1995]

well, I've run the v 8.0 long enough to get a pretty good idea of performance.

my 2mb c2d (1mb/core) gets almost exactly 450ppd average per 1000mhz of clock speed (hovering around 1800ppd for a while now). this uses ssse3 for core 2.

both of my 12mb c2q's (3mb/core) average 500ppd per 1000mhz of clock speed. this uses sse4.1.

my Q6600 is too new to have accurate numbers, though from looking around at others it seems likely to be closer to the 500ppd than the 450.

 

[Assimilator1]

both of my 2mb c2q's (3mb/core)

A?, I guess you mean 12MB?

Btw I found out that you can look at the AR of WUs you've completed in your account stats, maybe if I looked through at least several hundred results I could get some idea of an average AR............ sounds like fun .

[edit] Just checked 70 results (that'll do for 1 night!) & about 90% are ~0.39 AR ,I'll average them properly in excel when I have more results.
Do you think 200 will be enough to at least get a rough idea??

 

[bryanW1995]

I just give all of my high-producing units 4-5 days' worth of wu's. after a couple of weeks their RAC is typically spot-on with the amount that they're producing per day. Just take that # and divide by total # of ghz and you're gtg. theoretically.

btw, I just checked at lunatics.kwsn.net and they now have an sse2 optimized client for the v8.0 science port! good times ahead...

 

[Assimilator1]

Cool ,I thought I'd be able to use it on my old Skt A crunchers but they don't even support SSE2!, doh!.

2 weeks is too long to wait for a benchmark .

 

[Assimilator1]

» AK V8 Windows x32 SSSE3x Intel only

Only just noticed , what's the SSE3x about?

 

[bryanW1995]

ssse3 was new for the core architecture. I don't know about ssse3x. I've looked all over, and it seems that most of the references refer back to seti optimized apps. I THINK that the "x" was unnecessary as there is only one version of ssse3.

edit: if you click on the ssse3x link at lunatics.kwsn.net, it takes you to the actual download page. that page correctly identifies the instruction set as ssse3 with no "x" at the end of it. I think now that it was just a typo.

 

[Assimilator1]

No I don't think it is, if you click on the SSE3 & SSE3x links at the KWSN site they take you to 2 different files with different file descriptions. I don't know where you were looking .

SSE3x

For Intel Core 2 and Xeon Processors that support SSSE3 (65nm, or 45nm under RAM/FSB/Cache Pressure)
Also the file size is different.

SSE3
Supporting Intel and AMD processors that have SSE3

I still don't know what the difference is though, wth do they mean by 'cache pressure'?, sounds like a bad translation.

 

[bryanW1995]

http://en.wikipedia.org/wiki/SSSE3

 

[Assimilator1]

Ahhh now I see! I'd missed the extra 'S' !, so yes you are right the 'x' is a typo or maybe they put it in to make it stand out.

That explains why my Q6600 is slower than the others as I'm running SSE3 not SSSE3 (stupid Intel naming!).
So basically anyone running Intel Core 2 CPUs & on want to be running SSSE3, & for all recent(ish) AMDs & older Intels SSE3.

Time to upgrade my client! [edit] done.

Maybe you could highlight the differences between SSE3 & SSSE3 in your 1st post & include that useful wiki link? ,save others from making my error .

 

[Smoke]

Ass1,

Now your Q6600 will be as fast as mine. :laugh:

 

[Assimilator1]

I hope so , & then soon I hope to over overclock it some more.........mind you temps are hitting 59C now,hmm....

 

[petrusbroder]

I have compared the seti@home standard application with the v.8 optimized application for Intel, SSE4.1. Here is the app I used.

The CPU is a C2D E8200 (2.66 GHz) OC'ed to 3.4GHz. The core temps are 58-59ºC (room temperature 31ºC). All WUs have been validated.

All the WUs result in 19.07 credits.

Standard application: _________________ 1 471 seconds (average of 11 WUs) +/- 9 seconds (standard deviation).
Optimized SSE4.1 application: ________ ___ 694 seconds (average of 9 WUs) +/- 6 seconds (standard deviation)

I am quite happy with this result. It more than 2.1 times as fast, i.e. I'll get twice as many credits with the SSE4.1 application compared to the standard one. And for 2 cores ...
This computer replaces my old trused AMD XP 2600+ ...

 

[Assimilator1]

Awesome improvement & upgrade .
What would also be intersting is to see how SSSE3 compares to SSE4.1, & maybe SSE3 too.

BryanW
Btw you missed my question 2 posts up.

 

[danzigrules]

anyone care to help a ubuntu n00b with installing this?

I use the good ole apt-get install method to get the boinc client and manager, so I assume I need to put the files in /var/lib/boinc-client/projects/setiathome.berkeley.edu ?

Of course do sudo /etc/init.d/boinc-client stop first then copy the files, then sudo /etc/init.d/boinc-client start, and I should be good to go?

Thanks

 

[Assimilator1]

Bump for new SETI crunchers & new owners of relevant CPUs.

 

[Assimilator1]

Worth 1 more bump I think.

 

[dajeepster]

I've noticed that any computer that i have that has an nvidia based card and running the optimized seti@home apps... that it doesn't run the CUDA apps... anyone else experiencing this?

 

[lane42]

Looks like the only machine running the video card is your Q6850 machine, and it's the only one with the 6.4.7 client. I'am sure you have the latest video card drivers, and version 6.4.5 should work, so try the 6.4.7 on the other machines and see what happens.

Gregg