Riesel base 3 reservations/statuses/primes

[gd_barnes]

That must be a fast siever.


I will have to try a range with srsieve on my new Ryzen 9 that I just got 3 months ago.

[rogue]

Originally Posted by rogue

I will revise my estimate to about 32G per day as opposed to what I wrote yesterday. It is easier to get accurate numbers after many hours of runtime. Here is the current line:

p=61209583927, 14.10K p/sec, 83052 factors found at 1.43 sec per factor (last 134 min)

so it has been running a little under 20 hours.

Still running. I had to stop to install Windows updates, but there are about 3.37m candidates remaining so it has removed about 300,000 candidates in about 5 days.

p=308438889503, 28.20K p/sec, 64097 factors found at 3.90 sec per factor (last 1005 min)

[rogue]

Originally Posted by rogue

Still running. I had to stop to install Windows updates, but there are about 3.37m candidates remaining so it has removed about 300,000 candidates in about 5 days.

p=308438889503, 28.20K p/sec, 64097 factors found at 3.90 sec per factor (last 1005 min)

Still running:

p=702338392703, 25.26K p/sec, 8318 factors found at 9.99 sec per factor (last 1125 min)

It is down to about 3.2m candidates. I will stop when the removal rate reaches about 1 factor per 15 seconds.

[gd_barnes]

Originally Posted by rogue

Still running:

p=702338392703, 25.26K p/sec, 8318 factors found at 9.99 sec per factor (last 1125 min)

It is down to about 3.2m candidates. I will stop when the removal rate reaches about 1 factor per 15 seconds.

Why 15 seconds? Tests for R3 at n=75K-80K run about 6-8 secs on a modern I7-9700 with all cores running. Plus there are many tests that will be removed by k's found prime. You could have probably justified stopping at a removal rate of 5 secs

I suppose that maybe you will be testing on slower machines, which maybe makes 15 secs better. But just thought I'd point it out.

[rogue]

Originally Posted by gd_barnes

Why 15 seconds? Tests for R3 at n=75K-80K run about 6-8 secs on a modern I7-9700 with all cores running. Plus there are many tests that will be removed by k's found prime. You could have probably justified stopping at a removal rate of 5 secs

I suppose that maybe you will be testing on slower machines, which maybe makes 15 secs better. But just thought I'd point it out.

Now you tell me! :-)

15 seconds is the time for a single test on this laptop at n=80000.

I'll switch over to testing the range.

[gd_barnes]

My rule of thumb for CRUS: Sieve on my fastest sieving machine until the removal rate equals the test time of a candidate at 50-60% of the n-range of the average of the machines that I will be testing on. Those machines that I will be testing on may or may not include the fastest siever.

I just checked out all that sieving that I did for R3 using my newer Ryzen 9...a machine that I did not have when I did the sieving. Since it's an AMD it should be somewhat faster at sieving than Intels by my experience. It was. Using srsieve at P=35G I was getting a test removed every ~9 seconds. Using the Intel that I had used before it was about every ~10-12 seconds.

With an average test time of 6-8 seconds at n=70K-80K as stated previously, from my perspective P=35G was slightly over-sieved for the entire k=25G-55G range.

I wanted to check it with the Ryzen in case it was significantly faster and I could justify sieving some more. It wasn't enough faster to do so.

From your perspective it looks like you could have stopped sieving at P=~400G somewhere using your super speedy siever assuming that your average test time is similar to mine. Anyway...just food for thought if you sieve k=36G-37G for your next set of testing.

[rogue]

I only have one machine that can sieve 4000 or so sequences (enough for a 1G range of R3) in the GPU. I just kicked off the 36G range.

Since I will be using PRPNet to manage the testing, I can statistics in the database to estimate the average PRP test on the same computer. After a few days (or a week) I should have enough data to determine the optimal factor removal rate, although I believe that your estimate of 400G is very close.

[rogue]

On the computer where I'm doing the sieving tests take about 12.5 each on average.

But since I have multiple computers, the average across all of the cores is 7.5 seconds per test. That is my sieving target removal rate.

In short, if using srsieve2cl, this is going to be based upon the average test time of the various cores you are using for PRP testing.

With my cores a range of 1G will take about 17 days of PRP testing.

[rogue]

Originally Posted by rogue

On the computer where I'm doing the sieving tests take about 12.5 each on average.

But since I have multiple computers, the average across all of the cores is 7.5 seconds per test. That is my sieving target removal rate.

In short, if using srsieve2cl, this is going to be based upon the average test time of the various cores you are using for PRP testing.

With my cores a range of 1G will take about 17 days of PRP testing.

For my setup the optimal sieve depth is between 500e9 and 550e9.

[rogue]

R3 35G is done to n=100000. Primes are attached.

Gary, do you want residues for this range?

[gd_barnes]

Originally Posted by rogue

R3 35G is done to n=100000. Primes are attached.

Gary, do you want residues for this range?

Yes

Are you reserving the next 1G range? Or should I just wait until you reserve each range?