A chunk of the log would be helpful, showing a few percent of the SMP unit is all that is needed (I can reconstruct it from there). Thanks!
Printable View
A chunk of the log would be helpful, showing a few percent of the SMP unit is all that is needed (I can reconstruct it from there). Thanks!
here is a log file. There are different projects included so I thought you might want more than just one slice.
thanks :)
*edit* it won't open, tells me it is invalid :(
try renaming to .gz suffix and see if that helps. I'll work on a redo in the meantime
OK that worked, thanks :)
http://fahinfo.org/index.php?avgscores=true
Angra, all chips aren't equal so frequency is not a good measure of the return of PPD. new C2D based quads do more instructions per clock cycle than those older xeons for example.
UnG - I had the idea in my head that my xeons should do about the same clock for clock as a Q6600. They are based on the same generation die and have the same cache. Are there other factors?
I tried to use that website to guesstimate what 5450s might do, but I have to know what projects are likely to be assigned, don't I? Anyway, there is no option for "Harpertown" which is the relevant core name, I think.
OIC , new xeons, my bad, yeah they should do about the same per GHz per core set. Most of the quads are using higher memory bandwidth configurations on the consumer boards. Best guess is within 5% per GHz.
sparky,
so what's the answer? :)
Phenom 9600 2.3GHz and Q6600 2.4GHz are very similar, Q6600 is a little faster (partially due to the extra 100MHz but that's not all of it). Once I get my 9600 BE put in and clocked up I'm going to see what similarly clocked Q6600 chips are doing and compare again to see - for at least F@H - how Phenom scales in comparison.
^^
Yea but who has their E6600/Q6600 at stock clock?:D
I know, that is only part of it. I just wanted a starting point for the comparisons.
Quote:
Originally Posted by angra
Angra, I looked at project 3060 in your case. I can't compare to a Q6600 because I haven't seen the logs for a single 3060 project folding on one yet.
However: your machine returns 1% of project 3060 and an average of about 885 seconds. That's an average of 88500 seconds or 24.58 hours for a complete run of project 3060. Project 3060 is worth 2539 points for a PPD average on project 3060 of 2487 PPD.
at 2.66GHz that's ~935 PPD per GHz . or 233.75 PPD per GHz per core which is better than the 185.68 PPD per GHz per core reported by the few Kensfields results I've seen.
Unfortunately I have nothing better to compare it to as most units that get those WU's are running Linux/Unix/OSX variants with Quads.
Based on what we know about memory bandwidth on servers like those vs tweaked systems like ours and how F@H responds to better memory bandwidth, I'd say you are within 5% of PPD per GHz average of what a Q6600 can do with the same OS. Given margins of error, it's a push (toss up).
Does that help?
It helps :) mostly I am just curious. That log is from a 5355 (2.66GHz). There are actually 2 on that machine (2 sockets, 8 cores, 2 installs). 5400 series question is a hypothetical.
hey, whadya know, I caught you guys in ranking again! :) chop, chop!
haha, 2 days until we overtake the littlewhitedog, and then we move two more positions within a week and a half. Neener neener come catch us Angra :p
BTW THIS TEAM IS REALLY STEAMING ALONG.....
:wtf: :wtf: :wtf: :wtf: :wtf:
Here's the bench for a 3060 wu and a Q6600 at 9x372. I know that's not stock, but it might help with the comparison.Quote:
Originally Posted by Ugly n Grey
I have been folding for XS since about the last week of January. I now have more points in XS's bank in these few weeks then for over 10 months at overclock. woot!
Edit: watch out sparky ;)
wow, you must have more than one instance running? because that works out to 875 PPD per Ghz , so somewhat slower than what Angra is putting up on the xeons. Of course if you're doing that twice per day per GHz with two clients running, that's smokin fast (almost twice as fast per GHz than Angra)Quote:
Originally Posted by [XC] 4X4N
Good work 2long :up: look out though, I've built a shelf to stack my quads on (I'm not kidding).
No, that's only one client running. The 3060's take forever, but don't give you many more points. That same machine does the 2653's in 7 min 50 sec per frame. I hate the 3060's. It's all I have been getting for a few days.
Well I know they take more crunching, I often wonder how well they compile this stuff given the advancements in instructions available. Especially with quads/SSE3/4 and the average RAM install being so high.
Anyhow, it looks like your xeons are slightly faster than a consumer grade quad per cycle Angra (10% give or take my 5% error rate). Absolute speed is lower because of the higher clocks but per clock is good. :up:
Could just be the difference in OS's, - not really an apples to apples hardware comparison.
OS's can make a difference. I was running linux with my other quad. 2653 wu's would take just over 6 min per frame. Same set-up with windows was about 7 1/2 min per frame.
I can definitely see that.Quote:
Originally Posted by [XC] 4X4N
I think the problem lies with the use of MPI, which is actually a Unix port to Windows, not windows native multithreading, so there's a huge hit there until the experts get that sorted out. Personally, I think it's the worst way to proceed, Windows makes up the bulk of systems in the known universe (good or bad , it's the truth) and here we are resurrecting a message passing interface based on old fortran junk. But it's really popular with anyone trying to track movement in space (outer or inner such as molecules). In conclusion, it sucks and we're stick with it.
/rant
Anyhow, the fact that Linux as a Unix clone does better at MPI is no surprise at all :) so let the Linux boxes rock on :up:
BTW: Detected number of cores F@H assigns project 3060 to Quads and p2653 to C2Ds. Depending on project's priority Quads can have p2653s too (and have had), but p3060 shouldn't be assigned to C2D.
FYI, as far as Project 2653 goes, my parents stock Q6600 at 2.4 GHz gets almost exactly 2100 PPD.
Edit: Very weird, I just got the Affinity changer going on my overclocked Q6600. I have never received a Proj 3060 until yesterday, and I got one on each pair of cores. Took over 24 hours to complete both of them for an avg. of 1650 PPD instead of the usual 2100 PPD on the 2653's. Disappointing for points, but we all know its not all about the points anyway.