New 2665 WU's

Just as a teaser, I got 3 2662 WU with the A2 core and the output is about 4600 ppd with a single client on a Q6600 at 3.0 GHz...
OK, it seems that my PPD on this unit is actually pretty good, about 1700PPD. Mind you this is really an arbitrary figure given that everyone's system specs vary considerably, but it's better than what I have seen for the P3065s which are the worst so far, IME. As a quick comparison, the same machine is also folding on a P3064, but that WU is only producing ~1400PPD. Stanford did well to increase the value for the P2665.

Still some bugs to kill. The other side of the medal is that anytime you try to stop the client or restart the computer, it destroy the WU :(
Even with usually dependable network disconnect method?

 
OK, it seems that my PPD on this unit is actually pretty good, about 1700PPD. Mind you this is really an arbitrary figure given that everyone's system specs vary considerably, but it's better than what I have seen for the P3065s which are the worst so far, IME. As a quick comparison, the same machine is also folding on a P3064, but that WU is only producing ~1400PPD. Stanford did well to increase the value for the P2665.

Even with usually dependable network disconnect method?


It's not the same issue as stalling. When I said "destroy", it's litterally that and when it happen, some file got corrupt so it delete it and get a new one...

The ppd I gave is with 1 single client on Ubuntu 8.04 host (no VM, no tricks, anything). This is what it should run with full efficiency compared to A1 core which is inefficient.

 
The ppd I gave is with 1 single client on Ubuntu 8.04 host (no VM, no tricks, anything). This is what it should run with full efficiency compared to A1 core which is inefficient.
I misread your post. I thought you wrote 2665, when in fact you wrote 2662. I never received a 2662 WU yet, only 3062s. If it's one of those beta WUs then it would explain it.

I have received 2665s almost exclusively today. Most of my SMP clients are folding them now. When it rains it pours I guess...
 
Picked up my first two p2665's on my VM's.
FahMon gives me ~1,800 PpD.
So I'm only dropping ~250 PpD per VM vs p2605 so thats not to bad.

Luck ............ :D
 
I misread your post. I thought you wrote 2665, when in fact you wrote 2662. I never received a 2662 WU yet, only 3062s. If it's one of those beta WUs then it would explain it.

I have received 2665s almost exclusively today. Most of my SMP clients are folding them now. When it rains it pours I guess...

2662 is basically the same as 2665 but run on the A2 core instead of the A1. it's still beta (mostly due to the core issues, not WU issues).

 
Is this Stanford's way to "encourage" 1 smp client per quad? My production has tanked when I get the 2665's.

 
I got a 2665 and on my quad @ 3.0ghz with dual smp under windows I get a little over 1500 ppd and on the other client a 2653 I get around 1420 ppd.



 
2662 is basically the same as 2665 but run on the A2 core instead of the A1. it's still beta (mostly due to the core issues, not WU issues).
Thanks for the info. Do you have any idea when the A2 core will be out of beta?

Is this Stanford's way to "encourage" 1 smp client per quad?
I doubt it since the deadline for these WUs is substantially longer.
 
I am currently on vacation and had the opportunity to build a new monster rig (not for me):
Q9450 on an Asus Rampage motherboard, with 4Gig corsair XMS2 Cl4 ram, Geforce 9800 GTX (which is just as much a toaster oven as the 8800 GTX) in a enormous coolermaster cosmos case. I loaded on Vista 64 bit and ran prime 95 for about an hour and it looked stable. You really need to fold for a while to see if the computer is going to go the distance. With only one SMP client running it is cranking through a 2665 at 12 min 27 sec per frame (2191ppd) There is a bunch of monitoring software running so I don't know if that might slow it down much. If it did, it probably wouldn't be too much.

In case you are interested in what the diference in temp is on the cores between stock 2.66GHz@100% and 3.2 GHz@100% load.....it is only about 1-2 degrees C using a arctic freezer pro 7. I hope my drool doen't casue the system to short out.... now we just need to get some faster ram to really push that CPU! I hope this new computer becomes a permanent folder for team 33....maybe.
 
I am currently on vacation and had the opportunity to build a new monster rig (not for me):
Q9450 on an Asus Rampage motherboard, with 4Gig corsair XMS2 Cl4 ram, Geforce 9800 GTX (which is just as much a toaster oven as the 8800 GTX) in a enormous coolermaster cosmos case. I loaded on Vista 64 bit and ran prime 95 for about an hour and it looked stable. You really need to fold for a while to see if the computer is going to go the distance. With only one SMP client running it is cranking through a 2665 at 12 min 27 sec per frame (2191ppd) There is a bunch of monitoring software running so I don't know if that might slow it down much. If it did, it probably wouldn't be too much.

In case you are interested in what the diference in temp is on the cores between stock 2.66GHz@100% and 3.2 GHz@100% load.....it is only about 1-2 degrees C using a arctic freezer pro 7. I hope my drool doen't casue the system to short out.... now we just need to get some faster ram to really push that CPU! I hope this new computer becomes a permanent folder for team 33....maybe.

Assuming that ram is DDR2-800, you can probably loosen the timings up and run it much faster. Hopefully its a 2x2gig set and not 4x1, the 2x2gig set is easier to OC most of the time. Anyway, the point is, loosen timings and up the fsb, as the cpu mhz gained is more important for folding than tighter ram timings. He'll never notice 5-5-5 timings over 4-4-4 while gaming or doing day to day stuff.

edit: I'd like to see what kind of time you're getting on the 2665 with 2 SMP's running if possible. Seems like 12.27 a frame is kinda slow for 3.2ghz and only one client. But who knows, maybe that's right.
 
Assuming that ram is DDR2-800, you can probably loosen the timings up and run it much faster. Hopefully its a 2x2gig set and not 4x1, the 2x2gig set is easier to OC most of the time. Anyway, the point is, loosen timings and up the fsb, as the cpu mhz gained is more important for folding than tighter ram timings. He'll never notice 5-5-5 timings over 4-4-4 while gaming or doing day to day stuff.

edit: I'd like to see what kind of time you're getting on the 2665 with 2 SMP's running if possible. Seems like 12.27 a frame is kinda slow for 3.2ghz and only one client. But who knows, maybe that's right.

You're right, it is the 2x2gig DDR2-800 ram (that is why I stopped his OC at a lowly 400 fsb on that mobo. I was worried about his stability with OC'd ram. If I drop down to 5-5-5-18 from the 4-4-4-12 spec speed (at 2.1volts) how much do you think the ram would take for an OC...DDR2-860? This systems primary use is for flight simulation so the extra CPU speed would be a big benefit over the increased memory latency. I haven't had much experience with ram overclocks...but this ram even has the crazy big (and sharp) heat sinks.

Back on topic 12:27 did seem a little low to me, but dual SMPs are only for deadicated folders and I don't want to scare this one away with the elaborate shutdown/start up routines to keep them working (and not lose data).

I'll report back on frame times with a higher OC later....
 
at 5-5-5-18, you can technically go in the DDR2-950 area fine. The 2.1 volts is a maximum value and the actual voltage might be much less.

 
Ok, I dropped a little, still managing 3100+ ppd w/ the 2665's.

When the A2's are released, it looks like 1 SMP per quad is the way to go.
 
My whole farm has got the 2665 WU.

Getting about 1700PPD x 2 = 3400 per boxen

I did noticed that one SMP = 13 per frame , while 2 x SMP = 16 per frame.

So, it is worth running two SMP clients with this work unit.

 
argh....one of these hit my athlon x2 4800 today....couldn't believe they would assign it to an older dual core. dropped me from ~1000 PPD on the general run that I was getting on there to 500 PPD during day-time usage. the quickest frame so far where I wasn't at the pc was 45:23 for 610 PPD. it should speed up a bit when I'm not using the fileserver, but still....this is brutal for an X2

further argh.....got one on my brisbane borg....that being said, while I was sleeping, inactivity on my fileserver and borg brought up the PPD. the Toledo 4800 is going for 770 PPD while the Brisbane is going for 745 PPD. This one looks pretty cache happy, with the Toledo having 2x1024 L2 cache vs the Brisbane's 2x512, leaving the Toledo faster despite the 100 mhz lower clock speed.

 
My OC-ed 5200 got one last night...... and will have it for quite a few :(

 
My 4400+ picked one up this morning.......26 minutes a frame and showing 1024ppd.....

 
16 of 18 smp's are chug'n them right now. sitting at 30k PPD.. not great not as bad as it was..,

 
I am currently on vacation and had the opportunity to build a new monster rig (not for me):
Q9450 on an Asus Rampage motherboard, with 4Gig corsair XMS2 Cl4 ram, Geforce 9800 GTX (which is just as much a toaster oven as the 8800 GTX) in a enormous coolermaster cosmos case. I loaded on Vista 64 bit and ran prime 95 for about an hour and it looked stable. You really need to fold for a while to see if the computer is going to go the distance. <snip>.

I agree, I've got several computers that I fold on until the warranty runs out before I'm sure they are stable and I'll give them to my family :)
Would not want to pass out an unstable computer.
 
Definitely. My quad was P95 stable for 24 hours and a chewy WU knocked it off it's square.



 
Remember push Prime95 stable and then knock it back a touch. My farm has been stable for nearly a month now without a hitch... Even converting for the Chimp Challenge was pretty easy... I largely attribute that to two things:

1. TRUE's
2. Static IPs



 
I don't know if it's my slow network or what not, but has anyone else noticed that the P2665 WUs take a lot longer to upload results? I mean way longer on the order of twice as long or even longer??

 
I don't know if it's my slow network or what not, but has anyone else noticed that the P2665 WUs take a lot longer to upload results? I mean way longer on the order of twice as long or even longer??


Just looked at my log from the last client that had one.. now my connection is not that great at only 384kb up but this is nuts...

Project 2653 snip...
[01:13:07] Folding@home Core Shutdown: FINISHED_UNIT
[01:13:10] CoreStatus = 64 (100)
[01:13:10] Sending work to server
[01:13:10] + Attempting to send results
[01:15:54] + Results successfully sent
[01:15:54] Thank you for your contribution to Folding@Home.

Project 2665 snip...
[19:29:56] Folding@home Core Shutdown: FINISHED_UNIT
[19:30:01] CoreStatus = 64 (100)
[19:30:01] Sending work to server
[19:30:01] + Attempting to send results
[19:46:35] + Results successfully sent
[19:46:35] Thank you for your contribution to Folding@Home.

That is insane :eek:
 
^ Yeah, I thought I wasn't imaging things. I just uploaded three of these back-to-back in the last half hour and it was at least 5 minutes UL for each...about 5x longer than a 2653! :eek:
 
[06:06:44] Sending work to server


[06:06:44] + Attempting to send results
[06:07:52] + Results successfully sent
[06:07:52] Thank you for your contribution to Folding@Home.
[06:07:52] + Number of Units Completed: 49


Not sure of your delays... took me a hair over a minute...

[05:21:02] + Attempting to send results
[05:25:33] + Results successfully sent
[05:25:33] Thank you for your contribution to Folding@Home.
[05:25:33] + Number of Units Completed: 48

That one took a bit more at 4 1/2 minutes...

I'm at 640Kb/s up..

[06:13:01] + Attempting to send results
[06:17:31] + Results successfully sent
[06:17:31] Thank you for your contribution to Folding@Home.
[06:17:31] + Number of Units Completed: 35

I pulled those form my last 6 downloads... Maybe run a speed test and make sure your modem doesn't need to be unplugged and sit for 30 seconds and reintialize.

 
The log also show the size of the results file, paste them here.

 
I just woke up to find both clients stuck at 100%, at "will end MPI now". It's been like that for 6 hours!



 
[06:06:44] I pulled those form my last 6 downloads... Maybe run a speed test and make sure your modem doesn't need to be unplugged and sit for 30 seconds and reintialize.
I ran a speed test. My UL speed is around 400kb/s.It's only with the 2665s that I see long times when sending results, not the other WUs. The 2653s take about a minute by comparison.
 
I remember seeing a thread at F@H about compression for upload... there must be something funky with the 2665... Maybe it is larger? Or is it just not compressing well prior to upload?



 
Just finished one.....

[21:04:14] + Attempting to send results
[21:04:14] - Reading file work/wuresults_06.dat from core
[21:04:14] (Read 22070464 bytes from disk)
[21:12:30] Posted data.
[21:12:30] Initial: 0000; - Uploaded at ~43 kB/s
[21:12:30] - Averaged speed for that direction ~43 kB/s
[21:12:30] + Results successfully sent

 
Yep.....that's what I thought....And the thing is, it didn't really take much longer than a 2605 normally does..... I was quite pleased actually.

 
Exactly.

TBH, if there's one thing that I have more concern than controversial point assignments is WU size. The bigger the farm, the greater the effect of these large WUs slowing the network down. When I had three 2665s UL one after the other last night, my network was at the mercy of the results servers for about 20 minutes.
 
Exactly.

TBH, if there's one thing that I have more concern than controversial point assignments is WU size. The bigger the farm, the greater the effect of these large WUs slowing the network down. When I had three 2665s UL one after the other last night, my network was at the mercy of the results servers for about 20 minutes.

I've been dealing with that problem here.

The new version of untangle is going to have QOS built in to it. As soon as that happens I'm going to put it in as a bridge and have it throttle that port.

Anyone know what port it uses?

 
Anyone know what port it uses?

Fah uses port 80 & 8080.
Port 80 tends to be the asignment server.
Port 8080 for uploading-downloading work.

I only notice the uploads when I play online games.
At which point I lag like hell .......... :mad:
But then 99.9% of my protiens are p2605 which are only 5 Meg upload size.

Luck ........... :D
 
Running dual 2665's on dual smps, combined they take up almost 1gb of ram. Is that normal or do I need to restart smp clients?


 
Running dual 2665's on dual smps, combined they take up almost 1gb of ram. Is that normal or do I need to restart smp clients?



That's normal. You are lucky WU 2619 isn't around anymore because 2x of them will eat over 1600 MB :rolleyes:

 
Back
Top