The Factoring Report at mersenne.ca

[Gordon]

I am nearing the end of a project to TF all exponents below 1M to at least 64 bits (is there an old version of mfaktc that do exponents < 100K anywhere?)

I pick up the ranges from this report


You will see that the 0.13M range has 143 exponents left to test. If you drill down into the report...

1. click on 0.13m
2. on next page put factored through to 63 bits and get data
3. get a long list of all those candidates
4. click on output TF worktodo and enter 64 bits as the upper limit
5. Get just a SINGLE exponent back

See screen caps.

This report is fundamentally broken, often you will get a "complete" list, factor them out of the way then rerun the same report, you would expect no candidates right? Wrong, on many an occasion it "finds" more candidates that need checking.

[Anonuser]

Originally Posted by Gordon

I am nearing the end of a project to TF all exponents below 1M to at least 64 bits (is there an old version of mfaktc that do exponents < 100K anywhere?)

I pick up the ranges from this report


You will see that the 0.13M range has 143 exponents left to test. If you drill down into the report...

1. click on 0.13m
2. on next page put factored through to 63 bits and get data
3. get a long list of all those candidates
4. click on output TF worktodo and enter 64 bits as the upper limit
5. Get just a SINGLE exponent back

See screen caps.

This report is fundamentally broken, often you will get a "complete" list, factor them out of the way then rerun the same report, you would expect no candidates right? Wrong, on many an occasion it "finds" more candidates that need checking.

The report is probably not broken. The 142 exponents in question do not show up in the second report because they are currently assigned. If you tick the "Output TF worktodo format" box, then it will tick the "Exclude currently assigned exponents" box automatically. I suspect that's some sort of "safety mechanism" to avoid accidental poaching.

[Gordon]

Originally Posted by Anonuser

The report is probably not broken. The 142 exponents in question do not show up in the second report because they are currently assigned. If you tick the "Output TF worktodo format" box, then it will tick the "Exclude currently assigned exponents" box automatically. I suspect that's some sort of "safety mechanism" to avoid accidental poaching.

Exponents must be on a merry-go-round of assignment/freeing up as the reports can change daily, by dozens of exponents.

[Anonuser]

Originally Posted by Gordon

Exponents must be on a merry-go-round of assignment/freeing up as the reports can change daily, by dozens of exponents.

Indeed. It seems that some of the resources were redirected to ECM.

http://www.mersenneforum.org/showpos...&postcount=498

[VBCurtis]

Originally Posted by Gordon

I am nearing the end of a project to TF all exponents below 1M to at least 64 bits (is there an old version of mfaktc that do exponents < 100K anywhere?)

How many factors did you find, out of how many exponents? Your project is a nice empirical way to check on the ECM status report: with a t25 marked done on every number below 1M, one would expect a 19-digit factor to be missed something like 1/e^9 times; that is, your project should have been fruitless.

For every 8000 factors of 19 digits, one would be (expected to be) missed by a t25. But only ~5% of exponents have a factor of 19 digits length (or are mersennes different from the usual heuristic?), so you'd expect a missed factor something like 1 out of 160,000 tests.

Note that even 25% of a t30 would halve the chances of a missed factor, and exponents at 130,000 level have a t30 complete.

[Madpoo]

Originally Posted by Anonuser

Indeed. It seems that some of the resources were redirected to ECM.

http://www.mersenneforum.org/showpos...&postcount=498

ECM work for a single exponent can be assigned to multiple users, but only for another ECM test. Which basically means that if it's assigned to someone for ECM you can *only* get another ECM assignment on it... no TF. :(

[lycorn]

Originally Posted by Gordon

I am nearing the end of a project to TF all exponents below 1M to at least 64 bits (is there an old version of mfaktc that do exponents < 100K anywhere?)

If memory serves me, 100K is the lowest we can gousing mfaktc.. Former versions wouldn´t factor exponents lower than 1M.

[LaurV]

Originally Posted by lycorn

If memory serves me, 100K is the lowest we can gousing mfaktc.. Former versions wouldn´t factor exponents lower than 1M.

The versions of mfaktX that sieve with the GPU are like that (i.e. all since/after v0.18, for mfaktc). Older versions of mfaktc which sieve only with the CPU exist, where you can go as low as 2k for the exponent, but they are about 5 times slower (still 40 times faster than the CPU) than the last version of mfaktc, and they also keep the CPU busy (you can not run P95 in the same time, or you can, but the output is lousy). I just used one to factor all expos between 2k and 100k from 57 to 58 (sometime higher) for the "skipped bitlevels" project from Mark.

[lycorn]

You´re right|
Your memory is better than mine. DDR4 vs DDR2?...

[lycorn]

While we´re at it: the report isn´t updating. The latest version is from the 7th Feb.

[lycorn]

Stuck again . Last update on the 22nd.