EulerMaxell's newbie questions

[kriesel]

Originally Posted by EulerMaxell

The machine has a webpage: https://www.marshall.edu/it/services...green-cluster/

It's described there as a cluster.
Ooh, 6 TFLOPS. A pair of Radeon VIIs beats that at FP64 3.36 TFLOPS each; trounces it at SP at 13.44 TFLOPS each. https://www.techpowerup.com/gpu-specs/radeon-vii.c3358
Web search for 'Nvidia Tesla 448 core' comes up as ~10-year-old c2050 to c2075 or m2070 GPU https://en.wikipedia.org/wiki/Nvidia_Tesla

550GB ram. My personal LAN tops that too, and probably so do some other GIMPS participants'.

Compare to UW-Madison, in which at least some individual departments have their own or multiple HPC clusters
https://hpc.chem.wisc.edu/using-the-clusters/
https://chtc.cs.wisc.edu/uw-research...g/hpc-overview
https://stat.wisc.edu/users-guide/
https://www.ssec.wisc.edu/news/articles/11671/
...
Also Condor was originated there. https://research.cs.wisc.edu/htcondor/description.html
https://htcondor.org/
https://kb.wisc.edu/sscc/page.php?id=102178

[Uncwilly]

Originally Posted by EulerMaxell

I am wondering if there is a way to run GIMPS on a supercomputer and if there are any guides for installing the program on a program that runs Cent OS.

Mlucas will run on all sorts of systems.
https://www.mersenneforum.org/mayer/README.html
Which is written by user ewmayer.
His email address is at the bottom of the readme.

[kriesel]

Originally Posted by EulerMaxell

I am part of a university that has a supercomputer called Big Green and I would like to run GIMPS on it.
Unfortunately, the program requires access to the Internet and the supercomputer does not have access to it.
I am wondering if there is a way to run GIMPS on a supercomputer and if there are any guides for installing the program on a program that runs Cent OS.
I am wondering where I can find resources for running GIMPs on a big computer.

GIMPS is not a program, it is an organized search composed of volunteers running various software on various hardware. Consult the reference info for available software for various types of CPUs and GPUs. And for your other questions. Use the forum search function. Do your own due diligence.

[Dr Sardonicus]

Originally Posted by EulerMaxell

I am part of a university that has a supercomputer called Big Green and I would like to run GIMPS on it.
<snip>

This may be a silly question, but have you talked to the University's IT people about this?

[retina]

Originally Posted by kriesel

My personal LAN tops that too ...

I'm not sure how that is relevant to the OPs questions.

[Dr Sardonicus]

Originally Posted by kriesel

<snip>
Ooh, 6 TFLOPS. A pair of Radeon VIIs beats that
<snip>
550GB ram. My personal LAN tops that
<snip>
Compare to UW-Madison,
<snip>

Disparaging other folks' computer resources is not in keeping with the spirit of this Forum.

[kriesel]

Originally Posted by retina

I'm not sure how that is relevant to the OPs questions.

Originally Posted by Dr Sardonicus

Disparaging other folks' computer resources is not in keeping with the spirit if this Forum.

The point about tone / disparagement is noted. Consider this second try post an apology for that aspect.

Characterizing "Big Green" as a supercomputer occurs to me as inaccurate. By orders of magnitude. I felt that needed addressing. These days supercomputers are measured in Petaflops not teraflops. https://top500.org/lists/top500/2021/11/

"Big Green's" specs are reachable or exceedable by a committed hobbyist of modest means. To me that means "Big Green" does not now qualify as a supercomputer.
These days, a hobbyist willing to amortize $200/month hardware expenditure over a few years can assemble used hardware to achieve specs that will dwarf "Big Green" specs as given.

Also note "For comparison, a desktop computer has performance in the range of hundreds of gigaFLOPS to tens of teraFLOPS." https://en.wikipedia.org/wiki/Supercomputer "Supercomputer" is a rapidly moving target, as a chart there shows several orders of magnitude advance per decade of time. "Big Green" at 6. TFLOPS may have qualified a decade ago. And may be that old, judging by the GPU specs. Perhaps Marshall U has plans to replace it. Or perhaps there's a typo in the specs stated on their web page.

If various individual participants of GIMPS were to configure their own current personally owned hardware, into Linux networked clusters, like "Big Green's" configuration is described, it's likely GIMPS would contain (if it does not already) at least several such exceeding "Big Green" posted specs. Without including AWS, Google Colab, etc such as employed by Ben Delo and others.

And as always, contributions of all sizes are welcome. The GIMPS project is inherently highly data-parallel, with many millions of exponents to factor and test independently. Close coupling of hardware may streamline management but would have little benefit on performance near the wavefront or well above. Networked clustering probably provides most of the management benefit and allows use of economical commodity hardware.

[EulerMaxell]

I don't have much experience with manual uploads, how complicated would doing it manually be?

[EulerMaxell]

I am wondering what is the meaning behind the massive jump in teraflops a few days ago
https://www.mersenne.org/primenet/graphs.php.
The graph jumps at 7.5 exaflops on March 7, 2022.

[Uncwilly]

I think that firejuggler turned in some P-1 results that broke the credit system.

[kriesel]

New results/day at that graph location shows a ~36% drop in LL/PRP and DC ~ 2-3 Feb (which I think is Ben Delo's PRP throughput dropping by ~200/day, and the corresponding low-cost PRP-proof assignments absent so not performed or counted as DC).