Getting others to do the work on exponents I like (was: Trial Factoring Progress)

[tuckerkao]

Originally Posted by Uncwilly

Remember this post?
Real random things can be quite lumpy.
Patience grasshopper.

I also remember you've told me that every exponent that I'm going run the PRP or LL tests if not an already known Mersenne Prime will turn out composite.

By the way, how come increase the P-1 bounds by only 60% almost double the entire factoring time as showing on the screenshot below?

[tuckerkao]

Originally Posted by kriesel

"Two years ago, after 20 years of searching, a volunteer identified the 51st Mersenne Prime, 2^82,589,933" (which is actually extremely composite).
It's not true that the lucky finder of 2^82589933-1 has been at it for 20 years.
"After less than 4 months and on just his fourth try, he discovered the new prime number." https://www.mersenne.org/primes/?press=M82589933

2^168,202,123 is also extremely composite. 2^168,202,123 - 1 will be my 4th PRP of this type.

M168202123: NF up to 2^77 and NF-PM1 as of today.


Out of all the M168,***,*23 exponents, I've tried M168377323, M168476323, M168433723.

Ben Delo has tested M168308323, M168485323, M168548323.

[Viliam Furik]

Originally Posted by tuckerkao

2^168,202,123 - 1 will be my 4th PRP of this type.

Where are the first 3?

[tuckerkao]

Originally Posted by Viliam Furik

Where are the first 3?

They were already mentioned on the lower part of the previous post.

If Ben Delo shows up again, he can take M168174323, M168314323, M168412723, M168424723, M168704323, M168779323. These are all PRP ready.

That Radeon RX 6900 XT still cost $1450. I want it to be at most the MSP price. I still don't know exactly which M168,***,*23 is the real hidden Mersenne Prime. I need both AMD Threadripper 5970X and Radeon RX 6900 XT to be able to run through all of them efficiently if no others help.

I'll get M168037423, M168133123, M168175123, M168196123, M168731923, M168931123, M168932123, M168934123, M168937123 trial factoring up from 2^77 to 2^78, will cost my GPU nearly 1 day each. They've all reached the NF-PM1 stage.

[Viliam Furik]

Originally Posted by tuckerkao

They were already mentioned on the lower part of the previous post.

Oh, so you meant that they were the ones you proved composite. I thought you were saying you somehow found 3 PRP numbers - PRobable Primes.

[tuckerkao]

Originally Posted by Viliam Furik

Oh, so you meant that they were the ones you proved composite. I thought you were saying you somehow found 3 PRP numbers - PRobable Primes.

I don't have a P-PRP result of a new exponent or the first time test as of now. If I get 1 someday, I'll ask for the LL tests from both Kriesel and you.

For several weeks, I thought M168433723 was the exponent I looked for, no it turned out composite.

I understand it's my own fault, if I want others to help me on the PRP tests, I have to get more exponents PRP ready at first before calling out on everybody. I thought I had like 10 of them, 4 of the M168,***,*23 already lost to F-PM1.

I've done the TFs all by myself for nearly half a year and UncWilly can view these records transparently, the only range I'd ask you for help would be the 2^77 to 2^78.

I'm currently running PRP on M168202123 using my CPU and TF M168202123 from 2^77 to 2^78 using my GPU, so I'm all booked for now.

[Viliam Furik]

Originally Posted by tuckerkao

I don't have a P-PRP result of a new exponent or the first time test as of now. If I get 1 someday, I'll ask for the LL tests from both Kriesel and you.

Stop it or I'll get a heart attack from laughing too hard... Even though I'm 18. (that is a bad joke, don't take it seriously)

If you somehow get a P-PRP result, "GIMPS prime discovery protocols" will come into action and a bunch of undisclosed trusted individuals will be tasked with running an LL test to verify the discovery.

[Uncwilly]

Originally Posted by tuckerkao

I'm currently running PRP on M168202123 using my CPU and TF M168202123 from 2^77 to 2^78 using my GPU, so I'm all booked for now.

If you are so sure about one of your special numbers being prime, you would be ahead of the game to use some of the online resources to run them all quickly. You can rent GPUs and CPUs in the cloud. If you spend the money to test them all and find one, you get all the glory.

[tuckerkao]

Originally Posted by VBCurtis

You quoted my post and my questions, but didn't address why your machine overheats; you merely confirmed that it does. Why does it do so? "it's old" isn't a cause of overheating.

So, I'll repeat myself- if you aren't able to grasp why a machine would overheat, and aren't able to solve the overheating problem, why will a new machine not overheat? That fancy GPU you list is going to make more heat than the GPU you have now, right? Why will that heat not matter?

I bought a cheap $40 CPU Air Cooler from Best Buy, 2 of my neighbors came to take the old liquid cooler out and install the new air cooler in.

It seemed like the speed of the CPU is functioning smoothly, see the screenshot below.

[drkirkby]

Originally Posted by tuckerkao

I bought a cheap $40 CPU Air Cooler from Best Buy, 2 of my neighbors came to take the old liquid cooler out and install the new air cooler in.

It seemed like the speed of the CPU is functioning smoothly, see the screenshot below.

It looks from your screenshot as if you are using 1 worker. Have you actually verified you get best throughput on those crazy big exponents with one worker? I find I get best throughput with 4 workers on exponents around 105 million, but using a bigger FFT, which one would need for 168 million exponents, I see my throughput is best with 3 workers, although there's not much difference between 2 and 3 workers, so I would run a longer test to ensure that the results we more accurate - the default is only 15 seconds. You might find that running 2 workers you can do 2 exponents in 40 days, rather than 1 exponent in 23 days.

I don't know exactly what FFT size you should test with. The output from Prime95 indicates it is using an FFT of 9 M. However, 9 M is 9216 k, but I can't benchmark 9216 k. However, I can benchmark 9600 k. Someone else could no doubt help you find the optimal FFT size to benchmark.

[Worker #1 Aug 11 21:47:02] Your timings will be written to the results.bench.txt file.
[Worker #1 Aug 11 21:47:02] Compare your results to other computers at http://www.mersenne.org/report_benchmarks
[Worker #1 Aug 11 21:47:02] Benchmarking multiple workers to measure the impact of memory bandwidth
[Worker #1 Aug 11 21:47:02] Timing 9600K FFT, 52 cores, 1 worker.  Average times:  2.74 ms.  Total throughput: 364.46 iter/sec.
[Worker #1 Aug 11 21:47:19] Timing 9600K FFT, 52 cores, 2 workers.  Average times:  3.10,  3.47 ms.  Total throughput: 611.13 iter/sec.
[Worker #1 Aug 11 21:47:37] Timing 9600K FFT, 52 cores, 3 workers.  Average times:  6.87,  6.92,  3.06 ms.  Total throughput: 616.53 iter/sec.
[Worker #1 Aug 11 21:48:08] Timing 9600K FFT, 52 cores, 4 workers.  Average times:  6.92,  7.06,  6.91,  6.95 ms.  Total throughput: 574.72 iter/sec.
[Worker #1 Aug 11 21:48:43] Timing 9600K FFT, 52 cores, 5 workers.  Average times: 11.91, 10.55, 10.40,  6.89,  6.85 ms.  Total throughput: 566.07 iter/sec.
[Worker #1 Aug 11 21:49:24] Timing 9600K FFT, 52 cores, 6 workers.  Average times: 11.60, 10.48, 10.56, 11.65, 10.55, 10.45 ms.  Total throughput: 552.59 iter/sec.
[Worker #1 Aug 11 21:50:10] Timing 9600K FFT, 52 cores, 13 workers.  Average times: 32.04, 31.99, 24.63, 24.56, 24.51, 24.64, 24.32, 23.91, 24.01, 24.28, 24.49, 19.31, 19.52 ms.  Total throughput: 534.79 iter/sec.

One worker is particularly poor on mine as I have dual processors, but 2 workers is not optimal.

[tuckerkao]

Originally Posted by drkirkby

You might find that running 2 workers you can do 2 exponents in 40 days, rather than 1 in 23 days.

I'm running the PRP test for M168202123, I may perform some P-1 factoring for other M168,***,*23 exponents during the meantime. It'll take a while before I get more of them to the PRP ready stage, then I can use AMD Threadripper 5970X to run through them with the speed of around 4 days per exponent next year. I'm able to complete 1 P-1 factoring per day if using only 1 worker currently, so this won't require me to have a newer high-end PC. There are still hundreds of those exponents without P-1 in my range, so my current old PC should have its role to finish the preparation tasks.

I only have 1 CPU on my current computer, I don't know why I need to use the dual worker configuration. I tried that on mfcktc with trial factoring 2 exponents at the same time, then it showed half the speed for both of them. I have only 1 GPU and that won't increase my overall output. My 1 CPU and 1 GPU can work simultaneously, but the amount of heat generated has already been a lot, more workers may stress the heat even greater.

My current CPU only has 6 total cores and it was bought in year 2013 with the LGA 2011 socket on the motherboard.

I've asked the PRP test of M168779323 from you before, I don't know how long it'll take you to finish the exponent, or if you run M168479323 with the dual worker configuration along with the other exponent since you probably won't be able to get back to Cat 1 exponents after at least 3~4 months.