The Next Big Development for GIMPS

[Uncwilly]

Originally Posted by storm5510

Race, as in which gets there first. As far as I can tell, I have not gotten any more certification work in the past 12 hours.

We might have more certifiers available than certs that need running at the moment. I haven't gotten any in about a day.

[kriesel]

Originally Posted by storm5510

As far as I can tell, I have not gotten any more certification work in the past 12 hours.

I had 16 CERTs complete in the past 24 hours, upon 9 systems. I've turned off a few from doing CERTs for now (download rate = 0/day) to displace some CERT work toward those who seem to want it.
I may need to do so on my Colab mprime instances too, since they may be doing little else in their brief runs. (Three of the 9 systems were colab accounts, and I get Colab sessions as short as 20 minutes.)

[chalsall]

Originally Posted by kriesel

... and I get Colab sessions as short as 20 minutes.)

Really? That's weird.

When I'm given a CPU (no GPU) instance, they always last exactly twelve (12#) hours.

[kriesel]

Originally Posted by chalsall

Really? That's weird.

Yes it is. When it's both cpu and gpu, it's rare to go 10 hours or even 2 hours. I've seen them down to around 18 minutes, maybe 15. CPU-only are usually around 90 minutes but can be shorter. Exact same setup will very occasionally go CPU-only 12hr 1 minute to 12hr 8 minutes. All durations indicated by last TOP output's system up time. The mprime run on cpu(s), and mfaktc or gpuowl on gpus if available, are background tasks, with top in the foreground at 180-second loop interval. Notebook code similar to https://www.mersenneforum.org/showpo...5&postcount=16

[Uncwilly]

Originally Posted by kriesel

I had 16 CERTs complete in the past 24 hours, upon 9 systems. I've turned off a few from doing CERTs for now (download rate = 0/day) to displace some CERT work toward those who seem to want it.

I just did a forced comms and got a Cert run.
The 97,000,000 range has shown a noticeable amount of exponents that proven composite above the surrounding ranges.

[Prime95]

Originally Posted by intelfx

Is there really any merit in putting an artificial limit here?

Well, no. My super-primitive Linux menus require me to apply some limit and I've upped it to 999999.

In reality 40MB/day and infinite MB/day will result in the same amount of CERT work. The option is there primarily for people to turn off or severely curtail the amount of CERT work they get.

[Prime95]

Originally Posted by kriesel

I had 16 CERTs complete in the past 24 hours, upon 9 systems.

Right now, there is a disproportionate number of proofs coming from gpuowl which cannot do CERT work. This leaves the CERT work to the relatively small number of prime95 clients.

This will change drastically should Ben Delo upgrade to v30.3. I plan on asking him to try phasing in soon.

[kriesel]

Originally Posted by Prime95

... upgrade to v30.3.

Hmm, v30.3, where and when?

[intelfx]

Originally Posted by Prime95

Well, no. My super-primitive Linux menus require me to apply some limit and I've upped it to 999999.

In reality 40MB/day and infinite MB/day will result in the same amount of CERT work. The option is there primarily for people to turn off or severely curtail the amount of CERT work they get.

Hm. Aren't those proofs ≥100 MB in size? With a 40 MB/day limit, how am I going to download even a single proof to CERT it?


I'm probably misunderstanding something.

[chalsall]

Originally Posted by intelfx

I'm probably misunderstanding something.

Yup.

Don't worry. During work like this few can keep up.

[chalsall]

Originally Posted by chalsall

Don't worry. During work like this few can keep up.

Sorry (sincerely).

Reading that reconfirms that I can be an unbelievably pretentious little prick.