mfaktc: a CUDA program for Mersenne prefactoring

[James Heinrich]

Originally Posted by kriesel

If you cut GpuSieveSize to 1024, how much performance do you lose?
If at GpuSieveSize=2047, you run two instances (in separate folders, very similar work; similar exponent, same bit level), do you gain combined throughput, or lose?

Factor=168999517,76,77
@ 2047 = 13020 Ghd/d
@ 1024 = 12916 Ghd/d
@ 512 = 12814 Ghd/d

Running one instance at GpuSieveSize=64 produced some nasty coil whine so I quickly shut that down.
Running two instances at 2047 produced throughput about 4020 Ghd/d each (so losing nearly 40% throughput compared to a single instance), but also a (softer) coil whine so I also shut that down quickly.

edit: just after I finished getting the numbers for the above post:
M168999517 has a factor: 139882222139204664331687

[Magellan3s]

Originally Posted by James Heinrich

I haven't used mfaktc for years until this morning. I now have a 4090 to test on.
What would you like me to test?

Did you bite the bullet and pay scalper prices? I'm nowhere near a Microcenter (6 hours+) and I can't force myself to pay hundreds of dollars over msrp.

[Mark Rose]

Originally Posted by Magellan3s

Did you bite the bullet and pay scalper prices? I'm nowhere near a Microcenter (6 hours+) and I can't force myself to pay hundreds of dollars over msrp.

They go for about $2500 here in Canada. That's currently about US$1850.

[Magellan3s]

Originally Posted by Mark Rose

They go for about $2500 here in Canada. That's currently about US$1850.

That's not TOO bad.... I'm looking at $2000 + Taxes on Ebay if I get it from a scalper... substantially more than MSRP pricing. I'm hoping sometime in the next few weeks the supply normalizes and I can get one from amazon for retail pricing.

[James Heinrich]

Originally Posted by Magellan3s

Did you bite the bullet and pay scalper prices? I'm nowhere near a Microcenter (6 hours+) and I can't force myself to pay hundreds of dollars over msrp.

Certainly not! BestBuy got a shipment of 250 of them for pre-order that www.droptracker.ca alerted me to so I got my order in. C$2299 for MSI GeForce RTX 4090 Gaming X Trio 24GB. Expensive, but I haven't bought a new GPU since my GTX 670 so I figured it's time.

[kriesel]

Originally Posted by James Heinrich

C$2299 for MSI GeForce RTX 4090 Gaming X Trio 24GB.

Sold out of course.
I don't think there's a version of mmff compiled and shared for CUDA Capability 8.9 yet (SDK 11.8 or higher per https://en.wikipedia.org/wiki/CUDA)
Perhaps someone blessed with a shiny new RTX40xx could take the plunge, and help out a bit with MM31 or MM127.

[rebirther]

Here is a working cuda12 linux app, the lib is static linked.

[kriesel]

Thanks rebirther. (Which Linux distro and version was that from?) Now we just need someone to come forward and do a build for Windows CUDA 12.

[lalera]

hi,
i have ubuntu v22.04 and cuda toolkit 12 installed
i compiled mfaktc v 0.21 and let the selftests -st and -st2 run - works
then i edited the makefile and compiled
then i tried it out on a installation without toolkit installed - works

[James Heinrich]

How does mfaktcv021linux64.zip that lalera posted differ from mfaktc-linux64-cuda12.zip that rebirther posted?

[rebirther]

Originally Posted by kriesel

Thanks rebirther. (Which Linux distro and version was that from?) Now we just need someone to come forward and do a build for Windows CUDA 12.

I have asked the user:

WSL: Kali Linux - Kernel: Linux i7-10700 5.15.79.1-microsoft-standard-WSL2, Cuda: cuda_12.0.0_525.60.13_linux.run, clang-version: Debian clang version 14.0.6

The libfile was included in the compilation (static) where I have got the error which I have posted last time. Answered your latest question.