Would P-1 been better than PRP on exponent 332,646,233 ?

[drkirkby]

I'm doing a PRP test on the large number exponent 332,646,233.

https://www.mersenne.org/report_expo...2646233&full=1

It's been shown by others to have no factors up to 2^81, but I'm wondering if the PRP test was the next logical test or not. Would P-1 been more sensible? If so, can I stop the PRP and start a P-1? If P-1 would have been more sensible to do P-1 first, then can i change to P-1? Of course, Sod's Law will probably leave a P-1 inconclusive, so I need to do a PRP after all. But I can always come back and complete the PRP. I tried requesting the manual assignment of P-1, but got the message

Error text: No assignment available meeting CPU, program code and work preference requirements, cpu_id: 2399764, cpu # = 0, user_id = 244634

My PC is not the fastest in the world, but it has 64 GB RAM and 26 cores. For some reason I was able to get PRP, but not P-1 manually.

I've only done 6.9% of the PRP test, but have an estimated 77 days to go. This is the first, (and almost certainly the last) time I tackle a PRP tests on a possibly 100 million digit number. The chances of getting it are too slim for the huge computational work needed to check it. Someone else gave up on a PRP test of this exponent some years ago. I wonder why?

That said, I'm hoping some upgrades to the computer will help. I bought a couple of faster CPUs yesterday. They still have the same number of cores as the CPU in my workstation (26), but having two CPUs will double that to 52 cores. I still need to resolve the issue of getting some more RAM, as the CPUs are not optimally configured - there should be 4 DIMMs per CPU for best performance, but I only have two DIMMs.

My puppy typed A3WQ2 when she jumped on the laptop keyboard - I thought I would leave it. I might see if the ASCII codes for that would form a sensible exponent, and if so try that!

Dave.

[LaurV]

That exponent has about 4.2% P-1 done from kriesel, which is enough for the actual front, but is somehow low for 332M, considering that the PRP test takes a lot of time in that range. You are good, but if you have a really good GPU, like a R7 or so, and want to try your look, you can do some more. You may want to ask kriesel if he kept the stage 1 checkpoint file, so you won't need to start from scratch. But my advice is just to continue the PRP.

edit: Maybe you forgot to check the "show full details" button, and you don't see the P-1 history?

[Runtime Error]

As LaurV has noted, yes it has had adequate P-1 done already by Krisel. You are right to PRP next. Make sure you are running a version of Prime95 that can create a proof file so that someone can double check it for less than 1% of the original effort.

Originally Posted by drkirkby

That said, I'm hoping some upgrades to the computer will help. I bought a couple of faster CPUs yesterday. They still have the same number of cores as the CPU in my workstation (26), but having two CPUs will double that to 52 cores. I still need to resolve the issue of getting some more RAM, as the CPUs are not optimally configured - there should be 4 DIMMs per CPU for best performance, but I only have two DIMMs.

The FFT length for a huge 332M+ exponent probably can't fit into the CPU cache, so having only two DIMMs is severely bottlenecking your throughput. You might cut your job time by almost half by adding two more so it can run in quad-channel mode.

In my experience of running 332M+ tests on dual-socket motherboard machines, it does not make sense for the two CPUs to share a single workload. In my case, the test actually runs slower on two CPUs than on one CPU (with 8 DIMMS total). Relatively, the machine can get more than double throughput by simultaneously running two 332M+ exponent tests, one on each CPU.

Originally Posted by drkirkby

My puppy typed A3WQ2 when she jumped on the laptop keyboard - I thought I would leave it. I might see if the ASCII codes for that would form a sensible exponent, and if so try that!

That prediction method will work just as well as anything anyone else has proposed!!! Perhaps put some peanut butter on the number pad to generate test candidates?