mersenneforum.org PRP testing service
 Register FAQ Search Today's Posts Mark Forums Read

2019-03-29, 20:43   #12
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

32·509 Posts

Quote:
 Originally Posted by M344587487 The limited testing I've done for this card puts the scaling closer to 2 but I wasn't exhaustive so the figure could be off. Encouraging bigger tests makes it easier to fill capacity but really I just went for simple. You could also min-max by requesting exponents near the end of an xxM block.
The 2.095 I gave was for prime95 or CUDALucas. Gpuowl has tested lower than 2 for me, and variable. I think it hasn't reached asymptotic yet in the range I've tested, and the opencl compile overhead of some seconds per start is a relatively constant overhead that lowers the power. https://www.mersenneforum.org/showpo...6&postcount=10

Quote:
 That's a funny rule, what happens if the result is automatically sent from an unattended PC and a database mod sees the result before the owner of the hardware? I'd already planned on sending the results to the user and having them report to primenet, it wouldn't take much to have the results sent to the user by email with no interaction on my part.
I think I have not stated it quite right. As I recall, Woltman knew before Pace, but Pace is listed first. Something to be careful and clear on though. I know how the printout was read has affected the order of discovery of primes.

Paypal would be a good early choice.

Quote:
 Thanks. I am considering a UPS and didn't know efficiency was a major factor. A UPS isn't a strict requirement as the supply is pretty stable and the workload not time critical, but it would be nice.
Laptops contain internal power backup. Some laptops have gpus too. I'm using memory sticks for redundant work in progress backup. If the old system dies, and the HD is toast, and the primary backup failed, I still have yesterday's daily xcopy to stick.

Last fiddled with by kriesel on 2019-03-29 at 20:44

2019-03-29, 21:29   #13
GP2

Sep 2003

29·89 Posts

Quote:
 Originally Posted by M344587487 That's a funny rule, what happens if the result is automatically sent from an unattended PC and a database mod sees the result before the owner of the hardware?
This has literally happened.

And once before, except that time it wasn't a database mod.

2019-03-31, 10:41   #14
preda

"Mihai Preda"
Apr 2015

3×421 Posts

Quote:
 Originally Posted by M344587487 It's an idea people have toyed with for years but it's not something that's materialised at least not publicly. Someone should do it at least to undercut the cloud which seems ill-suited to PRP yet is currently being used to some degree. I'm willing to to make a proper go of it and ramp up to whatever the demand is.
Nice idea, I wish you good luck, though I'm a bit pessimistic about how high the demand might be.

Did you consider offering fast-turnaround PRP for a premium? i.e. optimize for low-latency vs. throughput, and start the test right away. "Get a wavefront PRP fully done in under 24h, for a price"

2019-03-31, 15:50   #15
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

32×509 Posts

Quote:
 Originally Posted by preda Nice idea, I wish you good luck, though I'm a bit pessimistic about how high the demand might be. Did you consider offering fast-turnaround PRP for a premium? i.e. optimize for low-latency vs. throughput, and start the test right away. "Get a wavefront PRP fully done in under 24h, for a price"
"(Help M344587487 pay for his Radeon VII)"
Approx 20 hours for 84M.
A definite contender for verification of the next GIMPS prime discovery.
At $700 new, 5 year straight line depreciation, ~$1.15 cost/85M PRP at my current electrical costs, best I've seen yet, by a lot.

Last fiddled with by kriesel on 2019-03-31 at 15:51

2019-03-31, 18:12   #16
M344587487

"Composite as Heck"
Oct 2017

22·163 Posts

I've given the problem some thought and there are really two phases, the breakeven point and what is done after that. Attached is a PDF of the breakeven point of various configs of GPU count and time to breakeven with my numbers. I'm fine with any of them but need the funds to hand to buy the hardware and guarantee they can run to the breakeven point. After that my obligations are complete and I have some hardware to play with, the hardware is paid for but I probably can't afford to run them. If there is demand I could then offer something like electricity cost + £0.50 per test to account for the opportunity cost of not selling the hardware before it depreciates to nothing.

Quote:
 Originally Posted by GP2 This has literally happened. And once before, except that time it wasn't a database mod.
Nice. Nothing like old threads for a bit of nostalgia.

Quote:
 Originally Posted by preda Nice idea, I wish you good luck, though I'm a bit pessimistic about how high the demand might be. Did you consider offering fast-turnaround PRP for a premium? i.e. optimize for low-latency vs. throughput, and start the test right away. "Get a wavefront PRP fully done in under 24h, for a price"
I'm pessimistic about demand too, if it ends up doing nothing or prompting someone with better electric prices to undercut to the point there is demand so be it. Kriesel said about premium queue jumping too, is there really any benefit to be had?

Quote:
 Originally Posted by kriesel "(Help M344587487 pay for his Radeon VII)" Approx 20 hours for 84M. A definite contender for verification of the next GIMPS prime discovery. At $700 new, 5 year straight line depreciation, ~$1.15 cost/85M PRP at my current electrical costs, best I've seen yet, by a lot.
Hey that has a nice ring to it ;)

Yes I fully recommend R7 for PRP testing, it's the only thing that makes sense on a large scale if you're buying hardware.
Attached Files
 r7breakeven.pdf (22.2 KB, 60 views)

 2019-04-04, 20:31 #17 M344587487     "Composite as Heck" Oct 2017 22·163 Posts I've made a Kickstarter to try and crowdfund a PRP testing farm made of Radeon VIIs. It's along the same vein as what's suggested in this thread but translated into something that makes it viable for a non-enthusiast audience to participate: https://www.kickstarter.com/projects...ng-server-farm I don't know if it's wildly optimistic or even deluded to think that it could get funded but I'm giving it a go. Time will tell I guess, if you have any suggestions let me know.
 2020-02-03, 01:08 #18 JCoveiro   "Jorge Coveiro" Nov 2006 Moura, Portugal 2·13 Posts Hey! Nice topic! Paying for PRP tests! Any update on this project?
 2020-02-03, 08:49 #19 M344587487     "Composite as Heck" Oct 2017 10100011002 Posts Long dead due to lack of interest. This was an idea that sounded plausible in my head but reality disagreed.
2020-02-03, 08:51   #20
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

32·509 Posts

Quote:
 Originally Posted by JCoveiro Hey! Nice topic! Paying for PRP tests! Any update on this project?
The Kickstarter page says "Funding unsuccessful". That page seems to me quite well written. More about why someone might want to contribute might have helped Something along the lines of new discoveries, providing data for the number theorists, something like that.

Last fiddled with by kriesel on 2020-02-03 at 08:54

 Similar Threads Thread Thread Starter Forum Replies Last Post Nick Factoring 8 2016-03-18 03:37 rover Sierpinski/Riesel Base 5 6 2006-10-02 13:43 eepiccolo Math 6 2006-03-28 20:53 E_tron Software 1 2003-10-05 03:55 PageFault Software 2 2002-09-26 01:35

All times are UTC. The time now is 09:33.

Thu Oct 22 09:33:02 UTC 2020 up 42 days, 6:44, 0 users, load averages: 1.70, 1.77, 1.64