Trial Factoring with gpuOwl

[valeriob01]

For everybody to know: the old TF branch of gpuOwl currently works under the following conditions:

• ROCm version: 4.2.0
• Hardware: Radeon VII and Radeon Pro VII
• OS: Ubuntu 20.04 / kernel 5.4.0-73-generic
• Compiler: g++ (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0

However the results can't be submitted to Primenet:

Found 2 lines to process.
Done processing:
* Parsed 1 lines.
* Found 0 datestamps.
	GHz-days
Qty 	Work 	Submitted 	Accepted 	Average
0 	- all - 	0.000 		

    Did not understand 1 lines.
    Recognized, but ignored 0/0 of the remaining lines.
    Skipped 0 lines already in the database.
    Accepted 0 lines.

These lines were unknown:

{"exponent":"333730849", "worktype":"TF", "status":"NF", "program":{"name":"gpuowl", "version":"4.7-5b01b65-mod"}, "timestamp":"2021-06-05 08:14:07 UTC", "aid":"715A1D17BA9C63A6CCC66D09A3D3790B""bitlo":76, "bithi":77, "begink":"113201797068360", "endk":"226403807589960", "rangecomplete":true}

Problems in the result line:

1. Lacks username;
2. Lacks a comma before bitlo;
3. Datestamp not found;
4. GHz-days not found.

[preda]

Two questions:

1. does it find factors?
2. with a competitive speed, to be worth it?

[valeriob01]

Two questions:

1. does it find factors?

I would need an already factored exponent that is known to have factors to test...

2. with a competitive speed, to be worth it?

absolutely, it is very fast, faster than my RX Vega 64.

and I have tested running both PRP and TF on the same GPU at the same time and there was very little performance dropdown !

[valeriob01]

I found an exponent to test which is known to have factors because I have already computed it:

• Exponent: 199999673
• Bitlevel: 74-75
• Factor: Factor: 20677946129698001050849

will post the result here.

[valeriob01]

I found an exponent to test which is known to have factors because I have already computed it:

* Exponent: 199999673

* Bitlevel: 74-75

* Factor: Factor: 20677946129698001050849

will post the result here.

2021-06-05 18:29:46 {"exponent":"199999673", "worktype":"TF", "status":"F", "program":{"name":"gpuowl", "version":"4.7-5b01b65-mod"}, "timestamp":"2021-06-05 16:29:46 UTC", "aid":"715A1D17BA9C63A6CCC66D09A3D3790B""bitlo":74, "bithi":75, "begink":"47223742025460", "endk":"94456847945460", "rangecomplete":false, "factors":["51694949845488"]}
2021-06-05 18:29:46 Bye

[valeriob01]

I found an exponent to test which is known to have factors because I have already computed it:

* Exponent: 199999673

* Bitlevel: 74-75

* Factor: Factor: 20677946129698001050849

will post the result here.

2021-06-05 18:29:46 {"exponent":"199999673", "worktype":"TF", "status":"F", "program":{"name":"gpuowl", "version":"4.7-5b01b65-mod"}, "timestamp":"2021-06-05 16:29:46 UTC", "aid":"715A1D17BA9C63A6CCC66D09A3D3790B""bitlo":74, "bithi":75, "begink":"47223742025460", "endk":"94456847945460", "rangecomplete":false, "factors":["51694949845488"]}
2021-06-05 18:29:46 Bye

That is another factor, what it means?

[preda]

Have a look here:
https://www.mersenne.ca/exponent/199999673
the factors are of the form 2* k *exponent -1, and what is reported above is the "k" for the factor. Can easily be changed between the two.

[valeriob01]

Have a look here:
https://www.mersenne.ca/exponent/199999673
the factors are of the form 2* k *exponent -1, and what is reported above is the "k" for the factor. Can easily be changed between the two.

So the program is working, it only needs adapting the result line so that Primenet accepts it...

[valeriob01]

Have a look here:
https://www.mersenne.ca/exponent/199999673
the factors are of the form 2* k *exponent -1, and what is reported above is the "k" for the factor. Can easily be changed between the two.

So the program is working, it only needs adapting the result line so that Primenet accepts it...

Ok, and also primenet.py needs the TF (2) worktype to fetch trial factoring work.

[valeriob01]

Here's some benchmark for trial factoring exponent 218812621, bitlevel 73-74:

Radeon VII:

• GHz: 710-790
• ETA: 17 minutes

Radeon Pro VII:

• GHz: 780-890
• ETA: 16 minutes

[valeriob01]

Here's some benchmark for trial factoring exponent 218812621, bitlevel 73-74:

Radeon VII:

* GHz: 710-790

* ETA: 17 minutes

Radeon Pro VII:

* GHz: 780-890

* ETA: 16 minutes

For comparison: benchmark for mfakto on RX Vega 64, same exponent, same bitlevel:

• GHz/day: 634.55
• ETA: 19m44s

[valeriob01]

@preda Can you tell me just where is the code for the output line so that I can try to fix it?

[valeriob01]

@preda Can you tell me just where is the code for the output line so that I can try to fix it?

Tired of waiting for your reply, I went ahead and found it.
For contribution I share the command I used to find the code for output line:

find . -type f -exec grep -H 'bitlo' {} \;

which produces:

./Results.cpp:   snprintf(buf, sizeof(buf), "\"bitlo\":%u, \"bithi\":%u, \"begink\":\"%llu\", \"endk\":\"%llu\", \"rangecomplete\":%s%s",
Binary file ./openowl matches

This allowed me to add the comma before bitlo. Now I only need to understand why the timestamp is not accepted by Primenet.

[valeriob01]

I will open a discussion for this topic.