Strategic Double Clicking

[Madpoo]

Originally Posted by ATH

One of Curtis Cooper's results is bad, though it is 7 years old so that machine probably don't exist anymore: http://mersenne.org/M41425513

It just made me wonder if any of his current machines have mostly bad results? and if yes do you report it to him?

On the whole, his machines are generally decent... across all of them they're about average, because, let's face it, the number of machines he's had over the years basically defines the average.

There were only one or two older machines (no longer producing) that were more bad than good, if I'm remembering it right (I looked at his stats a while ago).

Let's see... doing a fresh look...

He's had nearly 10,200 cpu ids since Primenet version 5 went live in 2008'ish. Not to say there were that many unique machines (although it wouldn't surprise me), since reinstalling Windows or other changes make Primenet generate a new id.

Out of those 10,200, only 146 have at least 1 bad result, and only 107 of them have Bad >= Good.

Of course, 5418 of them have zero bad, zero good. Just haven't had any confirmations yet. 15 of those have a suspect result which mismatched on a double-check, so... probably bad? Might have to do some speculative checks on systems like that and see which way they go.

[Madpoo]

Anyone interested in settling an interesting quad check?
http://www.mersenne.org/report_expon...3500677&full=1

Both AirSquirrels and I did a check on this already, and I have my fingers crossed that my result is correct, but... we'll see?

We've been trying to do triple-checks of each others results when needed, but in this case we need someone else to weigh in. LOL

The worktodo line would be:
DoubleCheck=73500677,75,1

[Mark Rose]

Originally Posted by Madpoo

Anyone interested in settling an interesting quad check?
http://www.mersenne.org/report_expon...3500677&full=1

Both AirSquirrels and I did a check on this already, and I have my fingers crossed that my result is correct, but... we'll see?

We've been trying to do triple-checks of each others results when needed, but in this case we need someone else to weigh in. LOL

The worktodo line would be:
DoubleCheck=73500677,75,1

Sure. I'll have it done next week.

By the way, are you going to be queuing all the low triple checks again sometime soon? If so, I'll keep grabbing double checks. If not, I'll start grabbing triple checks from this thread to get a positive answer sooner.

[airsquirrels]

Originally Posted by Mark Rose

Sure. I'll have it done next week.

By the way, are you going to be queuing all the low triple checks again sometime soon? If so, I'll keep grabbing double checks. If not, I'll start grabbing triple checks from this thread to get a positive answer sooner.

I think Madpoo has a couple months more on his current hopper, I just pulled in another batch of TCs however I'm trying to hit the AMD GPU FFT sweet spots so it's mostly in the 50s, only 20 left in the 3x and the 4x are for the churners right now.

[Mark Rose]

Originally Posted by airsquirrels

I think Madpoo has a couple months more on his current hopper, I just pulled in another batch of TCs however I'm trying to hit the AMD GPU FFT sweet spots so it's mostly in the 50s, only 20 left in the 3x and the 4x are for the churners right now.

Can you tell me which range specifically is best for you? I'm using CPUs, so I'll stick to what's outside of it. In the <50M range, I'm chewing through about 100/month now.

[airsquirrels]

Originally Posted by Mark Rose

Can you tell me which range specifically is best for you? I'm using CPUs, so I'll stick to what's outside of it. In the <50M range, I'm chewing through about 100/month now.

Based on loads of data, these are the FFT sizes and corresponding ranges that do well on AMD clFFT:

2048K 37.13 - 39.53
2560K 46.17 - 49.25
3072K 55.18 - 58.85
3200K 58.85 - 61.22
3456K 64.23 - 65.95
4000K/4096K 73.18 - 77.99

The 39-46M range is not great for my GPUs, 61-64M and 66M-73M are the other big dead spots.

[Mark Rose]

Originally Posted by airsquirrels

Based on loads of data, these are the FFT sizes and corresponding ranges that do well on AMD clFFT:

2048K 37.13 - 39.53
2560K 46.17 - 49.25
3072K 55.18 - 58.85
3200K 58.85 - 61.22
3456K 64.23 - 65.95
4000K/4096K 73.18 - 77.99

The 39-46M range is not great for my GPUs, 61-64M and 66M-73M are the other big dead spots.

The below 46M for this thread is basically currently assigned or already done. There are 40 unassigned triple checks needed below 49.25M if you would like to queue them:

DoubleCheck=48014177,72,1
DoubleCheck=48019087,72,1
DoubleCheck=48021059,72,1
DoubleCheck=48027143,72,1
DoubleCheck=48042073,72,1
DoubleCheck=48044081,72,1
DoubleCheck=48050389,72,1
DoubleCheck=48062081,72,1
DoubleCheck=48074977,72,1
DoubleCheck=48099251,72,1
DoubleCheck=48104813,72,1
DoubleCheck=48122021,72,1
DoubleCheck=48122471,72,1
DoubleCheck=48140119,72,1
DoubleCheck=48178093,72,1
DoubleCheck=48214139,72,1
DoubleCheck=48222241,72,1
DoubleCheck=48223573,72,1
DoubleCheck=48322763,72,1
DoubleCheck=48348847,72,1
DoubleCheck=48367097,72,1
DoubleCheck=48372391,72,1
DoubleCheck=48410573,72,1
DoubleCheck=48427231,72,1
DoubleCheck=48434123,72,1
DoubleCheck=48517037,72,1
DoubleCheck=48531139,72,1
DoubleCheck=48538409,72,1
DoubleCheck=48572059,72,1
DoubleCheck=48573319,72,1
DoubleCheck=48648199,72,1
DoubleCheck=48657031,72,1
DoubleCheck=48665891,72,1
DoubleCheck=48685183,72,1
DoubleCheck=48760711,72,1
DoubleCheck=48761711,72,1
DoubleCheck=48762059,72,1
DoubleCheck=48805763,72,1
DoubleCheck=48151063,72,1
DoubleCheck=49149887,72,1

I'll focus on work below 55.18, which should keep me busy for a few months.

[airsquirrels]

Dumb question, but since I am doing this work on GPUs how, other than contacting madpoo, do I "reserve" these?

[Mark Rose]

Originally Posted by airsquirrels

Dumb question, but since I am doing this work on GPUs how, other than contacting madpoo, do I "reserve" these?

Get a new mprime install, configure your PrimeNet user id in prime.txt, add the lines above to worktodo.add and run `./mprime -c`. If they are not reserved, you'll end up with lines like:

DoubleCheck=DEADBEEFDEADBEEFDEADBEEF,12345678,72,1

and if they're already reserved,

DoubleCheck=N/A,12345678,72,1

in worktodo.txt.

[Madpoo]

Originally Posted by airsquirrels

Dumb question, but since I am doing this work on GPUs how, other than contacting madpoo, do I "reserve" these?

If you want, I can create the assignments for these like the others. In the batch I made previously I picked all of the 57M and larger in those ranges, plus a few in the sub 39M. There are a few hundred more in those other more-ideal ranges, (and a couple thousand more triple checks to go, in general).

We'll get to them all eventually.

If anyone else is interested in doing large* batches of triple-checks, I can create the assignments for you and PM the worktodo. The benefit of doing that is I can make sure you're not one of the previous testers.

* - by large # of assignments I mean in excess of maybe 40-50 exponents, for the simple reason that it's a manual process to modify my scripts for a particular user and pick out a good batch of exponents. I have scripts setup for AirSquirrels and myself which is nice when doing 300+ assignments at a time.

[Mark Rose]

I do wonder how many assignments you two do on a monthly basis.


Also, Madpoo, how many machines have now completed an LL test without having a DC currently assigned or completed? That would be a nice number to have a countdown of.