New Fermat factors

[Dr Sardonicus]

Originally Posted by houding

I don't go to the fermatsearch website that often, and I see there has a new factor been found.


Congratulations

Thanks for posting the news!

From the Fermat Search News page:

Nov 24th, 2021
New Fermat factor from Gary Gostin and FermatSearch!
205071338665 . 2¹³⁸² + 1 is a Factor of F₁₃₇₉!!!
Gary Gostin discovered the fifth Fermat factor of this year!
This was discovered on an Intel i9-9900K CPU with an RTX 2080 GPU running a new GPU-based factoring program.
We now have 360 Fermat factors known, one for each degree of the circle!
Congratulations to Gary from FermatSearch!

[rogue]

Originally Posted by Dr Sardonicus

Thanks for posting the news!

From the Fermat Search News page:

What is this "new program"? Can it be downloaded? It it open source?

[ET_]

Whops SORRY!!!

Nov 24th, 2021
New Fermat factor from Gary Gostin and FermatSearch!
205071338665 . 2¹³⁸²+1 is a Factor of F₁₃₇₉!!!
Gary Gostin discovered the fifth Fermat factor of this year!
This was discovered on an Intel i9-9900K CPU with an RTX 2080 GPU running a new GPU-based factoring program.
We now have 360 Fermat factors known, one for each degree of the circle!
Congratulations to Gary from FermatSearch!

[ET_]

Originally Posted by rogue

What is this "new program"? Can it be downloaded? It it open source?

Both Erwin Doescher and Gary Gostin are testing new codes, but neither is available for production at the moment.

[Gary]

The program is called “gfs” for GPU Fermat Search. It is a “Sieve + Trial Division” program that supports integers (trial divisors) up to 2046 bits and can search for factors of either Fermat numbers only or of Fermat numbers and GFNs (a^2^n + b^2^n). The sieve is performed by all cores of the CPUs, in parallel with the trial division calculations which are performed on the GPU. The program has 62 CUDA assembly libraries, one for integers of each multiple of 32 bits in size from 96 to 2048 bits. The libraries are generated by a separate program and contain a total of > 150K lines of CUDA assembly. As Luigi mentioned, gfs is still in development with a long list of tasks / issues to address. So far it has only been tested on Ubuntu with Pascal and Turing GPUs. Performance is currently 10x – 2x slower than mmff for 40 <= n <= 200, 40% to 20% faster than Feromant-cuda for 200 <= n <= 600, and 30x to 6x faster than pmfs for 600 <= n <= 2000. (for each comparison performance vs n is ordered: 10x @ 40 increasing to 2x @ 200, etc).

Originally Posted by houding

I don't go to the fermatsearch website that often, and I see there has a new factor been found.

Congratulations

Thanks!