Quick Question about assignments

[bcp19]

Originally Posted by Mr. P-1

Thank you. Are you just giving me benchmark info, or are you offering to take on some of the work we've been discussing?

It was just benchmark info in case anyone else later wanted to know what the 450 could do. With CUDALucas LLing I only have 8% GPU left and if I try to TF it takes 3-4 times as long.

[Chuck]

Originally Posted by Mr. P-1

I think the most useful unit of work here is a TF68-69. Everything else will be in multiples of that.

Based upon what people have said in this thread so far:

Dubslow ??? 15-16 minutes (about 90 per day)
Christenson GTX440 36-37 minutes (about 40 per day)
Christenson GT480 ???
garo ??? 13-14 minutes (est.) (about 100 per day)
Chuck GTX580 ???

68 to 69 is about 16.5 minutes

Chuck

[Mr. P-1]

Last night's grab yielded:

40 assignments under 50M TFed to 68 (No P-1)
59 to 68 (P-1 done)
51 to 69 (No P-1)

I also grabbed the 50Ms and 51Ms. This yielded 38 to 69 at 50M (No P-1). There were no assignments in the 51M range meeting our criteria.

[Mr. P-1]

Responding here to a remark I made in another thread about the leading edge of the first-time test wave:

Originally Posted by Mr. P-1

Test= assignments are in the low 55Ms. All appear to have been P-1ed. Most are TFed to 71 bits, but a few are only at 69.

It would make sense to prioritize these exponents for TF 69-70 over lower P-1ed exponents at the same bit level because it takes less time to do them, and a success saves more LL effort.

Question: Does it make sense to prioritize these over smaller exponents which have not been P-1ed? If the answer is "no", then we probably don't currently have the resources to do them. If "yes" then we do for at least some of them, and we should.

I've also seen some 55M Test=assignments which were not P-1ed. There were all factored to 71, but it raises the possibility that there are some out there which are TFed to 69 and not P-1ed. We should try to do these to 70 if we find any. Only TF 68-69 is higher priority, and I'm pretty certain we have more than enough capacity to do all of these.

[Dubslow]

Originally Posted by Mr. P-1

Responding here to a remark I made in another thread about the leading edge of the first-time test wave:



It would make sense to prioritize these exponents for TF 69-70 over lower P-1ed exponents at the same bit level because it takes less time to do them, and a success saves more LL effort.

Question: Does it make sense to prioritize these over smaller exponents which have not been P-1ed? If the answer is "no", then we probably don't currently have the resources to do them. If "yes" then we do for at least some of them, and we should.

I've also seen some 55M Test=assignments which were not P-1ed. There were all factored to 71, but it raises the possibility that there are some out there which are TFed to 69 and not P-1ed. We should try to do these to 70 if we find any. Only TF 68-69 is higher priority, and I'm pretty certain we have more than enough capacity to do all of these.

I think regardless of exponent, those without P-1 should be prioritized first, followed by bit levels. I don't think the exponent itself should matter. (Or, all else being equal, the higher exponent being given priority to due LL cost)

[Mr. P-1]

Originally Posted by Dubslow

I think regardless of exponent, those without P-1 should be prioritized first, followed by bit levels. I don't think the exponent itself should matter. (Or, all else being equal, the higher exponent being given priority to due LL cost)

No. Consider the following three options which would (on exponents about the same size) all take about the same amount of processing time.

Option 1. Factor two P-1ed exponents from 68-69. We have 2 * about 1% chance of finding a factor that the P-1 failed to find = 2 * about 1% chance of saving 2 LL tests.

Option 2. Factor one P-1ed exponent from 69-70. We have 1 * about 1% chance of finding a factor the the P-1 missed = 1 * about 1% chance of saving 2 LL tests.

Option 3. Factor one non-P-1ed exponent from 69-70. We have 1 * about 1% chance of finding a factor that a P-1 test would miss = 1 * about 1% chance of saving 1 P-1 test + 2 LL tests. We also have 1 * about 0.3% chance of finding a factor that a P-1 test would also find = 1 * about 0.3% chance of saving a P-1 test.

Option 3 is better than option 2, but only marginally. In fact the difference between them is that option 3 gives us an extra 1.3% chance of saving a P-1 test! Option 1 is clearly better than either, and should be prioritised.

[Mr. P-1]

Last night's grab yielded:

25 assignments under 50M TFed to 68 (No P-1)
18 to 68 (P-1 done)
50 to 69 (No P-1)

Additionally I got 62 assignments in the 50M and 51M ranges TFed to 69 (No P-1)

In view of the preceding analysis, starting from tonight, I will also keep exponents TFed to 69 that have had P-1 done.

Congratulations to Chuck for being the first to report back his completion of the work I had given him.

[Chuck]

Originally Posted by Mr. P-1

Congratulations to Chuck for being the first to report back his completion of the work I had given him.

I will start the next batch Monday after I get home and don't have to use remote access software to control my machine. Cutting/pasting/editing is annoying in that mode (using Teamviewer).

Chuck

[Mr. P-1]

Originally Posted by Chuck

I will start the next batch Monday after I get home and don't have to use remote access software to control my machine. Cutting/pasting/editing is annoying in that mode (using Teamviewer).

I use TightVNC to do, I guess, the same job. I've never tried to cut@paste through it though, instead I pass data via a share on the "remote" machine I have mounted locally. I put "remote" in quotes, because it's physically right next to my main box, and connected via the LAN. Couldn't you do something similar with a genuinely remote system using VPN?

[Chuck]

Originally Posted by Mr. P-1

I use TightVNC to do, I guess, the same job. I've never tried to cut@paste through it though, instead I pass data via a share on the "remote" machine I have mounted locally. I put "remote" in quotes, because it's physically right next to my main box, and connected via the LAN. Couldn't you do something similar with a genuinely remote system using VPN?

I don't have to use it very often. I'm not smart enough to get involved with anything more complicated.

[kladner]

I am running a Gtx 460, at a 15% OC: 776MHz. I can handle two PF's in mfaktc. Currently finishing up Primenet assignments in the 59M range, taking them from 69-72. Timings are:

69-70 ~35-40min
70-71 ~1h 15m
71-72 ~2h 30m

I would be glad to take up any more productive assignments which a more knowledgeable person would like to throw my way.