Strategic Double Clicking

[rudi_m]

Originally Posted by Madpoo

Here's a good old fashioned list of strategic double-checks below 60M. Been a while since an updated list and I thought it was due:

I took all below 50M.

[storm5510]

Originally Posted by rudi_m

I took all below 50M.

May I politely inquire as to how long it will take to run thig group?

[rudi_m]

Originally Posted by storm5510

May I politely inquire as to how long it will take to run thig group?

I don't know exactly because I have also queued a lot other work and it depends whether my users will run many corporate jobs or not next days. But I'm pretty sure that these 33 exponents I took will finish within about one month.

Currently I run mprime on 11 recent quad-core i5/i7 CPUs, but using only 3 cores per system and most of my machines run mprime only during the night and weekend (88 of 168 hours per week). If I wouldn't feel guilty about wasting energy and corporate money I could increase my throughput by factor 3 I guess. :)

[storm5510]

Originally Posted by rudi_m

I don't know exactly because I have also queued a lot other work and it depends whether my users will run many corporate jobs or not next days. But I'm pretty sure that these 33 exponents I took will finish within about one month.

Currently I run mprime on 11 recent quad-core i5/i7 CPUs, but using only 3 cores per system and most of my machines run mprime only during the night and weekend (88 of 168 hours per week). If I wouldn't feel guilty about wasting energy and corporate money I could increase my throughput by factor 3 I guess. :)

I understand. Thank you for the reply. From what I gather, there are several who run these in a network server environment. My last I/T job was over a decade ago. The company had eight Dell servers, all Xeon's. I wasn't the admin so I could not have gotten away with doing this.

I only took one because I don't know the time-frame in which MadPoo expects these to be completed. There are others who can run them faster than I. In the low 40M range, I can DC one every 36 hours with Prime95, more or less.

[Mark Rose]

I regularly queue up to a month of assignments. Right now I have only 2 assignments, because I've been busy with other things. With the current 40M assignments, I'm finishing about 7 per day.

Maybe I'll get bored tonight and queue a month, if there's enough work remaining for that.

[storm5510]

Originally Posted by Mark Rose

I regularly queue up to a month of assignments. Right now I have only 2 assignments, because I've been busy with other things. With the current 40M assignments, I'm finishing about 7 per day.

Maybe I'll get bored tonight and queue a month, if there's enough work remaining for that.

You've got to be using more than a single machine to do this? I wouldn't want to pay your electric bill!

[Mark Rose]

Originally Posted by storm5510

You've got to be using more than a single machine to do this? I wouldn't want to pay your electric bill!

At home I have 5 x i5-6600 underclocked to 3.3 GHz. I also have 2 x i3-4170. The mprime cost is about 400 watts, or about $55/month. Four of those i5-6600 run off a single power supply consuming around 260 to 270 watts. That cluster produces about 2.66 results per day, or about $0.48 per 40M DC in electricity.

At work, an i7-4770k, 2 x i7-4770, an i7-3770, an i5-2500k, and an i3-2120 run full time. I also run on another i7-4770 part time that's temporarily offline.

[Madpoo]

Originally Posted by storm5510

I only took one because I don't know the time-frame in which MadPoo expects these to be completed. There are others who can run them faster than I. In the low 40M range, I can DC one every 36 hours with Prime95, more or less.

Oh, I'm not picky about these. It is nice to see results from these lists within a few weeks (30-45 days I guess) but that's just me.

When I started the big project to independently verify all of the self-verified work, I frequently queued up 2+ months at a time... of course those were without any assignments but I didn't want too much queued up at once anyway.

I think as these strategic double-checks get harder to identify, it's probably just polite to not grab more than you could finish in a month or so, allowing others to partake in the fun.

[Madpoo]

Originally Posted by Mark Rose

I regularly queue up to a month of assignments. Right now I have only 2 assignments, because I've been busy with other things. With the current 40M assignments, I'm finishing about 7 per day.

Maybe I'll get bored tonight and queue a month, if there's enough work remaining for that.

With a single dual 14-core Xeon server, I could do about 4 of those per day (well, a little longer, I think 40M takes ~ 14 hours each using 14 cores per worker).

I'm a big fan of using servers with as many cores as you can get. It makes perfect sense for a virtual server which is their primary purpose, and as a bonus they can breeze through LL work pretty quick.

[storm5510]

Originally Posted by Madpoo

Oh, I'm not picky about these. It is nice to see results from these lists within a few weeks (30-45 days I guess) but that's just me.

When I started the big project to independently verify all of the self-verified work, I frequently queued up 2+ months at a time... of course those were without any assignments but I didn't want too much queued up at once anyway.

I think as these strategic double-checks get harder to identify, it's probably just polite to not grab more than you could finish in a month or so, allowing others to partake in the fun.

I looked at the single exponent I took on mersenne.org. I was surprised when I saw that it had remained unverified since 2009. It's almost like this one has been passed over by the server for some reason. Others had factored it to 72 bits since.

Originally Posted by Madpoo

I think as these strategic double-checks get harder to identify, it's probably just polite to not grab more than you could finish in a month or so, allowing others to partake in the fun.

I know I can do 20 a month on this i7 rig, but I didn't want to be greedy either. I have four others in my queue, all assigned by the server. It would be difficult now to go back to your list and pick out a few more without stepping on another's toes. I should try to take one from the upper end just to see how it times out. M59522479 is unassigned. I'll see if I can grab that one.
.....
Sucessfully registered http://www.mersenne.ca/exponent/59522479 with the server. Prime95 'estimates' four days to run. I don't think it will take that long. Note: I don't know how to make link appear as the exponent only.

[nofaith628]

Last time I queued up the exponents I had at least 60 assignments in the 50-68M. The single PC is still working hard churning through all these double checks.

Quite a few exponents have finished since I last posted here (and many mismatches were found too!), but the amount I grabbed should suffice for another month or so.

Current status with my dual 14 core Xeon machine is (running two workers with prime95):
~1.5ms/iter on 55M exponents (Worker 1)
~1.8ms/iter on 59M exponents (Worker 2)

The machine runs draws around 330W of power according to my UPS display.