PrimeGrid Autumnal Equinox Challenge September 23 - October 6

[biodoc]

No work for me yet. Web site is unresponsive for me.

 

[biodoc]

No work for me yet. Web site is unresponsive for me.

Helps to check "use CPU" on PrimeGrid Preferences. :whiste:

 

[sangyup81]

Really? Your Phenom II, should be able to hand in more than four workunits within 13 days. Hell you should be getting (depending on if you get 30 hrs/workunit) roughly 8-10 workunits per core, for a maximum of 32-40.

Again depends if you can get complete each workunit within 30 hours

OK, yeah I'm over 2% after 50 minutes so that should come out to 41:40 hh:mm per WU instead of the 180 hours I was getting before in the estimate

 

[waffleironhead]

turned on and crunching

 

[GLeeM]

Your PrimeGrid Preferences with only "Woodall Prime Search (LLR)" selected in the Projects section.

Does this mean that I have to crunch a different BOINC project other than Primegrid on my GPU if I want to participate in the race?

 

[Ken g6]

[*]Your PrimeGrid Preferences with only "Woodall Prime Search (LLR)" selected in the Projects section.

Does this mean that I have to crunch a different BOINC project other than Primegrid on my GPU if I want to participate in the race?

No, it doesn't. I hadn't thought about that before. I'll change it to "the only CPU project selected."

On the other hand, realize that any GPU project is going to use some CPU too.

 

[Ken g6]

I'll be in, but not with so many cores as usually, I have Correlizer (a project I really like) also running.

Say, if there's anyone who wants to run both PrimeGrid and another project at the same time, and who is running on Core 2 quads...

Then that person would do better running two instances of PrimeGrid and two instances of Correlizer that other project on each of two such machines than they would running one machine entirely on PrimeGrid and the other entirely on Correlizer that other project.

The reasons are a little mysterious, but it's known that Core 2 Quads run a little slower, per-thread, when running four long LLR WUs as opposed to one or two. It seems to have something to do with Core 2's being bottlenecked on RAM bandwidth.

 

[petrusbroder]

I agree. I run - due to temperatures - 3 or 5 or 7 instances of PrimeGrid LLR WUs and one instance of Correlizer.
Om my i920 PrimeGrid runs 7ºC hotter (all 8 threads) than g.e. Correlizer (all 8 threads). Cutting down one thread (7 threads PrimeGrid and 1 thread Correlizer) makes it "only" 5ºC hotter ...

 

[VirtualLarry]

how do I join team anandtech on PrimeGrid? I'm a newb to BOINC, just installed it and choose PrimeGrid project.

Edit: Ok, I joined the TeAm, and changed my PrimeGrid project preferences to Woodall, but it's currently running two "pps_sr2sieve_xxx" projects, and two "llrWOO_xxx" projects. ETA is 78 hours for the PPS, and 62 hours for the llrWOO. Guess I screwed up, although the BOINC project auto-started projects for me, before I even set up my preferences.

Edit: Ok, I discovered that, based on the default project settings, most of the WUs are CUDA. I tried re-enabling the project prefs to allow NV GPU usage, and resumed the CUDA tasks, but none of them are running. Any idea? I would like to run the GPU tasks, while doing Woodall on the CPU. Feasible?

Ok, cannot figure out how to get CUDA WUs to run at all.

Edit: Figured it out. Had to go into Tools/Computing Preferences, and select to enable GPU.

Hmm, now I want a six-core machine! Thuban OC to 4Ghz would be nice. (6x4.0)

Or perhaps BullDozer. If nothing else, I really hope BD is good for DC, even if it sucks for games.

Edit: Shutting down the GPU. My desktop PC cases just cannot handle the heat load of both the CPU and GPU going full-bore. CPU temps hit 83C, which is pretty close to the alarm temp on this mobo, and GPU temps were 73C, which is acceptable on the GPU, but it causes the CPU temps to rise. The PSU and top of case gets so hot you can barely hold your hand to it. I don't want to stress these rigs too bad, until I get a new set of cases for them.

Edit: Hmm. Unchecking "Use GPU when computer in use" doesn't shut off the GPU entirely. I set the screensaver to BOINC, and when the computer is idle, it still uses the GPU 100%. CoreTemp shows that my CPU temps got up to 86C! Ouch. Oh well.

Edit: I think I found a more permanent way to disable the GPU. Under "Activity" - "Suspend GPU". Now I think I finally shut the GPU off. Computer was starting to smell, I think it was overheating.

Edit: Last night, I left the GPU running, but downclocked it to 405Mhz (lowest Precision 2.04 slider will go). Temps on GPU were 65-66C, rather than 70-73C. CPU temps were 81-83C, rather than 86C.

But when I woke up the machine today, it was a tad sluggish, so I went to Tools/Computing Preferences, unchecked use GPU when computer in us, hit OK, and then the screen basically stopped updating. The mouse kept moving for a while, but nothing was redrawing or updating on the screen. GPU got hung or something. I dunno. I had to hit my Reset button on the PC. Hopefully I'm not having hardware problems.

 

[petrusbroder]

Welcome to the PrimeGrid crunching, Virtual Larry.
I really like your pitching-in for the race.
When the race is done, and you still need my help, I'll be happy to help with one 8-thread CPU and one GTX460 GPU.!

 

[somethingsketchy]

Only 43%, 43%, 40% on the first three work units. The anticipation is killing me. Does anyone else notice that there are already six teams on the board with credit, yet we are not in the first 24 hours of the competition?

http://www.primegrid.com/challenge/2011_7/top_teams.html

 

[RobertE]

Only 43%, 43%, 40% on the first three work units. The anticipation is killing me. Does anyone else notice that there are already six teams on the board with credit, yet we are not in the first 24 hours of the competition?

http://www.primegrid.com/challenge/2011_7/top_teams.html

A good number of those are bogus invalid results. The first real ones should start showing up anytime now though.

For me, my best ones are a little over 63% at this point.

 

[Ken g6]

Hm, I was going to post 1-day stats, but nobody on our TeAm has completed any work yet!

As for me, I won't have any work done 24 hours from now either.

 

[sangyup81]

Give me about 16 hours or so and I'll have my first 4 WUs in!

 

[Rudy Toody]

40% on my gimpy (has bad temperature sensor on chip, so it arbitrarily throttles back) pc.

46% on my good pc.

 

[RobertE]

About 11.5 hours from now for my first 4. Coming in at about 35ish each.

 

[Ken g6]

CoreTemp shows that my CPU temps got up to 86C! Ouch.

Ouch is right! Worse, if your computer is even slightly unstable, the result will be wrong and it won't count!

So, I suggest you back off your overclock and your voltage. Then run Prime95 for at least an hour or two to make sure you're stable. (Don't do blend test; just do large FFTs for now.)

 

[GLeeM]

The reasons are a little mysterious, but it's known that Core 2 Quads run a little slower, per-thread, when running four long LLR WUs as opposed to one or two. It seems to have something to do with Core 2's being bottlenecked on RAM bandwidth.

I think I remember you saying that turning off hyperthreading helps.

Can you estimate how much better my i7 920 at 4.0 will do if I turn off HT?

When running eight WOOs I am at about 41% after 25 hours

 

[Ken g6]

It would be more accurate to say that turning off hyperthreading helps if you're not juggling and you're close to the end of the race. This is a long race; pace yourself!

I would guess that with hyperthreading off each of four WUs would be at no more than 82%.

 

[VirtualLarry]

Ouch is right! Worse, if your computer is even slightly unstable, the result will be wrong and it won't count!

So, I suggest you back off your overclock and your voltage. Then run Prime95 for at least an hour or two to make sure you're stable. (Don't do blend test; just do large FFTs for now.)

It's a really mild overclock, actually. 2.5 -> 3.0, 333FSB -> 400FSB. No vcore increase at all.

OCCT stable using 64-bit linpack with 90% of 8GB RAM for an hour. Don't recall if I tried Prime95 on this rig.

Edit: 1.152v under full CPU load in CPU-Z. 45nm C2Q Q9300.

Edit: My computers run so hot (even in an AC-controlled room set to 69F!), because my case cooling is so poor. They are in CoolerMaster Elite 330 cases, which only have a single rear fan, and they are in cubby-holes, so they heat up. I'm hoping that by upgrading the cases to Rosewill Blackhawk cases (5 fans), that temps will improve.

All the heat from the GPU rises, and then heats the northbridge (burning hot!), and then heats the intake air for the CPU cooler, which heats the air going into the PSU.

Heatsink is a HyperTX2. I did blow the dust out when I upgraded them to quad-cores. Haven't checked the dust since, but I don't live in a very dusty place.

 

[RobertE]

4 WUs reported, and we are now on the board in 16th. Crunch on!

 

[biodoc]

4 WUs reported, and we are now on the board in 16th. Crunch on!

:thumbsup:

I've got 14 WUs in the 84%+ range so it'll be a few hours...

 

[rabrittain]

Edit: Last night, I left the GPU running, but downclocked it to 405Mhz (lowest Precision 2.04 slider will go). Temps on GPU were 65-66C, rather than 70-73C. CPU temps were 81-83C, rather than 86C.

I'm concerned about your cpu temps. (That's 177.8-181.4F!! (That last sentence is more of a note to myself, for reference. I realize that everybody on our forum knows how to convert temperature.))

I'm reasonably certain that you know more than I do, so I won't say anything about thermal insulating material or heat sink fins clogged with dust.

My cpu temps range from 50-78C during this current Primegrid race. The 50C machine is powered by a AMD Athlon 64 FX60 dual core cpu. There is an nVidia GeForce 9800 GT which has been running Primegrid Collen-Woodall gpu tasks for several months. I'm running Debian 64 linux on that machine, and the subroutine which displays the temperature actually displays the temps for core 1 and core 2, and there has been some discussion regarding the placement of the associated sensors, and many think that the reported temps for the cores are inaccurate for that cpu.

I have another machine, and except for the video card and associated tasks it is identical to the first, but the OS is Microsoft Win XP Pro sp3 (32 bit). I am not running any gpu tasks on that machine. Everest reports that the cpu temperature is 63C.

I have a machine powered by an Intel Pentium D 820+ and there is an nVidia GeForce 8800 GT video card (that is not very powerful, but it gives me an extra couple thousand Primegrid Collen-Woodall gpu task points per day.) The OS is Win XP Pro sp3 (32 bit). Everest reports the cpu temp to be 78C, and it always runs above 75C, no matter what project the cpu is dedicated to. When "Smoke" first sent it to me 4 or 5 years ago, I thought that it would burn itself up for sure. However, not only does it keep going (slow and steady wins the race), but it also gives me much less trouble than my other machines.

Rudy Toody provided me with all the essential components necessary to build a machine based on An AMD Phenom II X4 940 Black edition. None of the other computers are overclocked, but the Phenom is OC'd 10%. There are no gpu tasks associated with this machine. Speedfan 4.40 reports the cpu temp to be 70C. This is the first time that it has run so hot; it normally runs ~63C.


EDIT: next paragraph --

So Larry, do your cpu's normally run so hot? How many computers dedicated to crunching do you have; and what sorts of cpu's do you have? How are the temperatures reported, and have you ever had any instability or other problems due to such high temps? Since I originally wrote this, I looked at your signature. I am able to see what computers and cpu's you have.

(I'm not very curious, but I want to know anyway.) Thanks --

Rex

 

[petrusbroder]

My temps are running higher too: I have a range of 65ºC (AMD Phenom II ???) - 85ºC (Intel i7 920 OC'ed to 3.3Ghz) among my processors. I am not worried - they have been at these temps for "ages" (= at least 1.5 - 2 years) without issues.

I have invalid WUs reported in PrimeGrid, but those were (all except one) GPU-WUs ... and mostly from one of my ATI-GPUs, and a few from my GTX260. The one which error'ed out on a CPU errored out after less than 10 seconds of crunching ...

 

[rabrittain]

OK, I read it more carefully. They have to send it to you after the beginning time and you have to return it before the end time. Looks like I may only get 4 WUs in on my Phenom II x4 system since the second iteration won't finish in time =(

I don't think that we should assume that the estimated time to completion is correct. On 2 of my dual core machines (the ones with the Cuda tasks), the time was ~400 hours for each of those 4 tasks when they first started. Now they've been crunching for ~44 1/2 hours and the time to completion ranges from 119 to 345 hours.

The initial times to completion were much shorter on the other 2 machines (perhaps the tasks are a different type.) On the FX60 machine they started at ~65 hours, and after 43 1/2 they are 59 1/2 and 60 hours to completion.

On the phenom, the initial times to completion were ~39 1/2 hours. One of those tasks has been finished. Two of the other tasks have ~4 1/2 hours to go; the other task has ~6 hours to go.

Who knows how many tasks we'll get done.

I say, "Increase the 'Additional Work Buffer' (Advanced - Preferences - Network Usage tab) to 10 days, and go for it!!