GPU to 72 status...
 

Just for your information.

As of midnight tonight (2011.11.27), the GPU to 72 project has found 129 factors for LL candidates by way of GPU factoring which would not have otherwise been found, saving more than 258 needed LL/DC tests.

It has also found 40 factors for DC candidates which would not have otherwise been found, saving approximately 40 DC tests.

It has also found 31 factors for LL candidates by way of P-1 testing which might, or might not, have been found based on the amount of memory available to the LL testers which are supposed to do the P-1 work before the LL work. Saving approximately 62 LL/DC tests.

[URL="http://www.gpu72.com/reports/overall/"]200 factors found in total[/URL].

If you have a GPU and are not yet working with us, please [URL="http://www.gpu72.com/"]consider joining...[/URL]

I found another P-1 factor yesterday morning, but only reported it 20 mins ago...

*2 factors that were discovered earlier this week. Just figured out the other factor was in fact from GPU272...

Chalsall, moderators, this deserves a sticky.....in a very good way. :smile:
Might want to handle this like the severe problems thread, and just keep the handful of posts that remain relevant.

Now, as for what GPU 2 72 offers the GPU owner...the *BEST* exponents with the easiest control to TF, P-1, or LL, as you may desire. Unless you are quite diligent, you will find more factors with GPU to 72, and help more with LL milestones, than with any other method of getting assignments.

:bow::bow::bow::bow::bow::bow::bow::bow::bow::bow:

Overall "GPU to 72" Status
 

OK, it's been almost exactly a month since development started on the system, and approximately three weeks since Workers started using it.

I think we all believed were having a positive effect on the overall GIMPS effort, but this belief was based on subjective opinion rather than hard data. This has now changed.

I have updated the [URL="http://www.gpu72.com/reports/overall/"]Overall System Status[/URL] report to include the amount of LL/DC and P-1 work being saved. And I have to say, even I'm surprised...

[CODE] Factors Work Saved Average GHz/GPU Days per... Total GHz/GPU Days...
[U]Work Type Found P-1 LL/DC Factor Work Saved Factoring Saved[/U]
DC TF 43 0 43 1.964 24.110 84.493 1,036.735
LL TF 148 87 296 4.055 181.016 600.269 26,790.486
[U]P-1 35 70 57.800 176.849 2,023.006 6,189.743[/U]
[B]Total 226 87 409 2,707.769 34,016.965[/B][/CODE]

Those GPUs are just [B][I][U]amazing[/U][/I][/B]! Thanks for all the work everyone!!! :smile:

[QUOTE=chalsall;280437]Those GPUs are just [B][I][U]amazing[/U][/I][/B]! Thanks for all the work everyone!!! :smile:[/QUOTE]

And thanks for all of your work, chalsall! You, too, are amazing!:bow::bow::bow:

[QUOTE=kladner;280438]And thanks for all of your work, chalsall! You, too, are amazing![/QUOTE]

Thank you for the very kind words kladner. You are most welcome -- it's nice being able to give back to the community.

But, could I ask you (and others) to please stop with all the bowing? I know it's just a way of communicating respect through an emotionless channel, but it actually makes me a little uncomfortable...

I am simply standing on the shoulders of giants (George, Scott, Oliver, James, Linus, Monty, all the GPU and P-1 workers, et al...).

[QUOTE=chalsall;280472]
But, could I ask you (and others) to please stop with all the bowing? I know it's just a way of communicating respect through an emotionless channel, but it actually makes me a little uncomfortable...[/QUOTE]

The feeling is understood. Won't do it again. But the thanks remain.

[QUOTE=kladner;280473]The feeling is understood. Won't do it again. But the thanks remain.[/QUOTE]

Thanks. And the "you're welcome" remains as well... :smile:

BTW, Dubslow and I started a discussion here which I felt would be better under it's own thread. Please have a look at the [URL="http://www.mersenneforum.org/showthread.php?p=280468"]How should we compare GPUs to CPUs?[/URL] thread.

Feedback, thoughts, etc very much appreciated.

Welcome to the 100 P-1 club delta_t.

See you at 500???

[QUOTE=petrw1;280681]Welcome to the 100 P-1 club delta_t.

See you at 500???[/QUOTE]

At first I had no idea what you were talking about, my first reaction was that I've done thousands of P-1 over the years, but then I realized I hit 100 P-1 turn-ins on the GPU-to-72 site.

So yes, I'll see you there at 500.

[QUOTE=chalsall;280152]

If you have a GPU and are not yet working with us, please [URL="http://www.gpu72.com/"]consider joining...[/URL][/QUOTE]

Is possible to run both the GPU and my Quad at the same time? Will this overheat or shorten the life of my computer in any way?

[QUOTE=Primeinator;280710]Is possible to run both the GPU and my Quad at the same time? Will this overheat or shorten the life of my computer in any way?[/QUOTE]

Honestly, I think taking part in any distributed project probably shortens the life of your PC at least a tiny bit (but maybe not).

But as long as everything is adequately cooled, it probably doesn't shorten it all that much. I would try it out and take a look at your temps. Anything over 80 is probably alarming (though I prefer all mine under 70).

[QUOTE=delta_t;280709]At first I had no idea what you were talking about, my first reaction was that I've done thousands of P-1 over the years, but then I realized I hit 100 P-1 turn-ins on the GPU-to-72 site.

So yes, I'll see you there at 500.[/QUOTE]

I'm aiming for year end....

[QUOTE=Primeinator;280710]Is possible to run both the GPU and my Quad at the same time? Will this overheat or shorten the life of my computer in any way?[/QUOTE]

yes, of course...that's what most of us do....

And, as long as the heat is managed, your computer's life won't change much...it will probably be replaced by upgrade before it dies....

CPU-Z and GPU-Z are your friends on Windows, sensors on Linux....

[QUOTE=KyleAskine;280712] Anything over 80 is probably alarming (though I prefer all mine under 70).[/QUOTE]
Hehe. My 2600k stock heatsink running at 3.5 GHz (3.4 stock) is running about 80C. That's why I've been asking so much about the Hyper 212+ :P

[QUOTE=Christenson;280717]

CPU-Z and GPU-Z are your friends on Windows, sensors on Linux....[/QUOTE]
The only thing I've found that works was something like GKrellM, and that didn't pick up my 2600k when I installed it... (although my Linux has been down for a few months.)

[QUOTE=Dubslow;280731]Hehe. My 2600k stock heatsink running at 3.5 GHz (3.4 stock) is running about 80C. That's why I've been asking so much about the Hyper 212+ :P[/QUOTE]

Wow, you only have a slight overclock over stock and you're at 80C? Yeah perhaps an upgrade to your heatsink would be wise. Seems anything better than the stock heatsink would provide much better cooling. As a comparison, I have a 2500K at 4.5 GHz at 60C using an H80 at lowest fan speeds.

[QUOTE=delta_t;280735]Wow, you only have a slight overclock over stock and you're at 80C? Yeah perhaps an upgrade to your heatsink would be wise. Seems anything better than the stock heatsink would provide much better cooling. As a comparison, I have a 2500K at 4.5 GHz at 60C using an H80 at lowest fan speeds.[/QUOTE]

And I have my 2500k at 4.3 GHz at around 60C using a hyper 212.

The stock intel heat sinks are a joke. They finally removed them completely from the 3930 and 3960.

[QUOTE=Christenson;280717]yes, of course...that's what most of us do....

And, as long as the heat is managed, your computer's life won't change much...it will probably be replaced by upgrade before it dies....

CPU-Z and GPU-Z are your friends on Windows, sensors on Linux....[/QUOTE]

Okay. Well I will definitely consider adding my GPU to the effort. Will this also slow down the LL testing?

[QUOTE=Primeinator;280792]Okay. Well I will definitely consider adding my GPU to the effort. Will this also slow down the LL testing?[/QUOTE]

Not sure how your quad compares to mine (Intel Core2 Quad Q8200 @ 2.33GHz), but mine runs 'faster' with the GPU running. To explain a bit, the Core 2 Quad seems to share something between core 'pairs' that slows them down depending on the work being done. I started off using P95 with 4 LL's running, and all 4 had around 90ms per iteration. I found out if I switched core 2 and 4 to TF that the LL's now were at around 60ms per. So, using batch files and the start command to run mfaktc on cores 2 and 4 while setting P95 to 1 and 3 keeps the LL's around 60ms. My other GPU machines are i5's and I see a slight change with mfaktc running (2-4ms slower).

[QUOTE=bcp19;280797]Not sure how your quad compares to mine (Intel Core2 Quad Q8200 @ 2.33GHz), but mine runs 'faster' with the GPU running. To explain a bit, the Core 2 Quad seems to share something between core 'pairs' that slows them down depending on the work being done. I started off using P95 with 4 LL's running, and all 4 had around 90ms per iteration. I found out if I switched core 2 and 4 to TF that the LL's now were at around 60ms per. So, using batch files and the start command to run mfaktc on cores 2 and 4 while setting P95 to 1 and 3 keeps the LL's around 60ms. My other GPU machines are i5's and I see a slight change with mfaktc running (2-4ms slower).[/QUOTE]

Mine is an i7 Q720 1.6 Ghz.

[QUOTE=lycorn;280801]Back to the thread subject...
[/QUOTE]

Should we move my question(s) on this thread to another? I am sorry for moving the thread off topic; I merely had a couple of questions regarding the GPU project and potentially adding my efforts to it.

[QUOTE=Primeinator;280806]Mine is an i7 Q720 1.6 Ghz.
[/QUOTE]

I'd reccommend doing like I did then, set P95 to use core 1 and 3 and set up mfaktc/o to run on core 2 and 4. Doing this you will see a significant speed up in your LL processing per core if you currently have all 4 cores doing LL. 2 core = 6 exp in the same time 4 core = 8 exp with this setup.

[QUOTE=Primeinator;280806]
Should we move my question(s) on this thread to another? I am sorry for moving the thread off topic; I merely had a couple of questions regarding the GPU project and potentially adding my efforts to it.[/QUOTE]

No worries. Thread hijacking is part of the thrill. It happens all the time.

Milestones
 

>100,000 Ghz Days
> 300 Factors found
>1,000 P-1 completions

Wow! Congrats for the hundred terahertz trial factoring done!

[QUOTE=petrw1;280991]>100,000 Ghz Days
> 300 Factors found
>1,000 P-1 completions[/QUOTE]
Primes found?
TFed to 72?

[QUOTE=davieddy;281067]Primes found?
TFed to 72?[/QUOTE]

Sorry I'm more into "round" numbers.

Unlikely this work will find Primes since it is almost all factoring.
But I'll let you know if we find one anyway.

TF'd to 72+: 3,604

[QUOTE=petrw1;281095]TF'd to 72+: 3,604[/QUOTE]

Factors found: 317.
LL/DC tests saved: 574.
P-1 tests saved: 128.
Total GHz Days of work saved: 48,116 (132 GHz YEARS!!!). :cool:
Candidates worked and returned to PrimeNet: 6,818.

I've added some more data on the [URL="http://www.gpu72.com/reports/overall/"]Overall System Progress[/URL] page.

Trying to be as open and transparent as I can be....

Who had the second LL completion?

[QUOTE=Dubslow;281127]Who had the second LL completion?[/QUOTE]

[URL="http://www.gpu72.com/reports/workers/"]Worker Progress[/URL]

[QUOTE=chalsall;281100]Factors found: 317.
LL/DC tests saved: 574.
P-1 tests saved: 128.
Total GHz Days of work saved: 48,116 (132 GHz YEARS!!!). :cool:
Candidates worked and returned to PrimeNet: 6,818.

I've added some more data on the [URL="http://www.gpu72.com/reports/overall/"]Overall System Progress[/URL] page.

Trying to be as open and transparent as I can be....[/QUOTE]

And this is only since October?? With fewer than three dozen people assisting in the GPU project?

Something like that. GPU's for the win!

[QUOTE=Primeinator;281140]And this is only since October?? With fewer than three dozen people assisting in the GPU project?[/QUOTE]

Only since NOVEMBER... The last 30 days... :smile:

GPUs are serious game changers.... :cool:

[QUOTE=chalsall;281143]Only since NOVEMBER... The last 30 days... :smile:

GPUs are serious game changers.... :cool:[/QUOTE]

:explode:

Wow...Remarkable! I couldn't find the correct emote for "flabbergasted" so the above one will have to do. I checked my GPU but I do not think that it is of the correct type to work towards the GPU effort.

Work saved
 

For some time I've thought that the future DC
saved shouldn't be valued as highly as the
initial LL.

OTOH, it has recently occurred to me that [B]in practice [/B]
the typical fate of first time assignments is to get
returned unfinished about 4 times before they
get completed, thereby increasing "work" saved.

David

[QUOTE=Primeinator;281151] I checked my GPU but I do not think that it is of the correct type to work towards the GPU effort.[/QUOTE]

What kind is it? Any recent AMD or NVIDIA should be able to contribute.

I remember he posted in another thread; if memory serves it's a (really) old nVidia.

[QUOTE=KyleAskine;281170]What kind is it? Any recent AMD or NVIDIA should be able to contribute.[/QUOTE]

[QUOTE=Dubslow;281173]I remember he posted in another thread; if memory serves it's a (really) old nVidia.[/QUOTE]

It is an ATI Mobility Radeon HD 4670 according to my device manager.

Wait till the first of the year, then you can get a lot of GPU for just a few bucks!!! :smile:

[QUOTE=Primeinator;281174]It is an ATI Mobility Radeon HD 4670 according to my device manager.[/QUOTE]

The good news is that you do support OpenCL 1.0! The bad news is that you have to have OpenCL 1.1 to use mfakto :down:

I don't know why this is required, and would ask bdot why 1.0 is no good.

Apparently memory does not serve.
[QUOTE=Christenson;281176]Wait till the first of the year, then you can get a lot of GPU for just a few bucks!!! :smile:[/QUOTE]
What do you mean?

[QUOTE=Dubslow;281188]Apparently memory does not serve.

What do you mean?[/QUOTE]
My Galaxy GTX440 was about $100 last June. I'm expecting some after-Christmas sales on not quite state of the art hardware that will include plenty fast GPUs....allowing you to pick up a decent one for a good price.

[QUOTE=Dubslow;281127]Who had the second LL completion?[/QUOTE]

It was mine.

Cross quoted from a thread in GPU Computing:

[QUOTE=Dubslow;281145]It's [URL]http://www.newegg.com/Product/Product.aspx?Item=N82E16814127519[/URL]
Hopefully that helps.[/QUOTE]

I'm surprised that all the GTX 460 cards at NewEgg are still that pricey. Cheapest one is $159.99 after rebate, free shipping. I guess I caught Microcenter in close out mode when I got mine at $99.95 after rebate. Still waiting on the rebate. This was a local purchase so I did have to pay a little over $10 in sales taxes.

I would concur in Christenson's expectation of post-holiday sale prices.
:down:
Maybe this beastie ([url]http://www.newegg.com/Product/Product.aspx?Item=N82E16814130626[/url]) would even get more affordable. Not that I would suggest it seriously. Dual GTX 460 SLI-on-a-card, minimum 46 amps on 12v. Worse, it looks sure to dump most of the heat inside the case. Bunch of hard luck stories in the reviews, too.

I don't know about what's there now, but the one I linked to I got for $149.99 (no rebate) way back in April.

[QUOTE=kladner;281233]
:down:
Maybe this beastie ([URL]http://www.newegg.com/Product/Product.aspx?Item=N82E16814130626[/URL]) would even get more affordable. Not that I would suggest it seriously. Dual GTX 460 SLI-on-a-card, minimum 46 amps on 12v. Worse, it looks sure to dump most of the heat inside the case. Bunch of hard luck stories in the reviews, too.[/QUOTE]

This has a name: "Bleeding Edge". Those heat pipes are trying to tell you something about how much heat you are going to get, and the need to manage it somehow (see liquid versus air cooling thread!)

Can you say "jet engine sounds coming from fans?" :smile:

[QUOTE=kladner;281233]
Maybe this beastie ([url]http://www.newegg.com/Product/Product.aspx?Item=N82E16814130626[/url]) would even get more affordable. Not that I would suggest it seriously. Dual GTX 460 SLI-on-a-card, minimum 46 amps on 12v. Worse, it looks sure to dump most of the heat inside the case. Bunch of hard luck stories in the reviews, too.[/QUOTE]

Yeah, the HD 6990 and the GTX 590 have begun putting their fans in the middle, pushing half of the hot air back into the case. I cannot understand why they would do that, other than that they simply could not cool it sufficiently anymore with their intake at the end of the card.

[QUOTE=bcp19;280797]Not sure how your quad compares to mine (Intel Core2 Quad Q8200 @ 2.33GHz), but mine runs 'faster' with the GPU running. To explain a bit, the Core 2 Quad seems to share something between core 'pairs' that slows them down depending on the work being done. I started off using P95 with 4 LL's running, and all 4 had around 90ms per iteration. I found out if I switched core 2 and 4 to TF that the LL's now were at around 60ms per. So, using batch files and the start command to run mfaktc on cores 2 and 4 while setting P95 to 1 and 3 keeps the LL's around 60ms. My other GPU machines are i5's and I see a slight change with mfaktc running (2-4ms slower).[/QUOTE]
The memory controller for core 2 quads is on motherboard rather than the cpu. They all have huge problems with memory bandwidth and quads. The core 2 quads are actually two dual core cpus bolted together.

[QUOTE=henryzz;281382]The memory controller for core 2 quads is on motherboard rather than the cpu. They all have huge problems with memory bandwidth and quads. The core 2 quads are actually two dual core cpus bolted together.[/QUOTE]

I find the same with my Q9550. It really slows down when I have all 4 "cores" working. I have all 4 doing P-1 right now and even though I limit it to 3 High Mem it still completes at half the pace of my i5-750 OC'd to 3.20. I am trying to run two P-1 workers with 2 cores each to reduce the memory bandwidth wars; so far it seems to be faster.

[QUOTE=petrw1;280681]Welcome to the 100 P-1 club delta_t.

See you at 500???[/QUOTE]

And monst makes 3.

Daily completions to 72...
 

Hey all... To show to a certain individual who continues to believe that this sub-project is somehow "hurting" GIMPS, I did a quick query to see how many candidates we were completing to 72 "bits" or higher. I'll probably put this into yet another report shortly, but so it's "on record" now:

[CODE]mysql> select date(Completed),count(*) from Assigned where FactTo>=72 and Completed>0 group by date(Completed) limit 100;
+-----------------+----------+
| date(Completed) | count(*) |
+-----------------+----------+
| 2011-11-05 | 42 |
| 2011-11-06 | 45 |
| 2011-11-07 | 28 |
| 2011-11-08 | 35 |
| 2011-11-09 | 47 |
| 2011-11-10 | 38 |
| 2011-11-11 | 27 |
| 2011-11-12 | 49 |
| 2011-11-13 | 30 |
| 2011-11-14 | 36 |
| 2011-11-15 | 53 |
| 2011-11-16 | 70 |
| 2011-11-17 | 31 |
| 2011-11-18 | 54 |
| 2011-11-19 | 19 |
| 2011-11-20 | 76 |
| 2011-11-21 | 124 |
| 2011-11-22 | 105 |
| 2011-11-23 | 116 |
| 2011-11-24 | 127 |
| 2011-11-25 | 99 |
| 2011-11-26 | 155 |
| 2011-11-27 | 207 |
| 2011-11-28 | 311 |
| 2011-11-29 | 234 |
| 2011-11-30 | 246 |
| 2011-12-01 | 148 |
| 2011-12-02 | 387 |
| 2011-12-03 | 187 |
| 2011-12-04 | 445 |
| 2011-12-05 | 177 |
| 2011-12-06 | 423 |
| 2011-12-07 | 216 |
| 2011-12-08 | 143 |
+-----------------+----------+[/CODE]

Also, so you all know... As it appears we've saturated the available P-1 fire power, I've begun returning to PrimeNet candidates TFed to 72 or above but without P-1 work completed; I'm still currently holding the lowest 250 for P-1 workers.

[QUOTE=chalsall;281513]I did a quick query to see how many candidates we were completing to 72 "bits" or higher. I'll probably put this into yet another report shortly[/QUOTE]

Actually, I realized this was better plotted on a graph. Please see the bottom of the [URL="http://www.gpu72.com/reports/overall/"]Overall System Progress[/URL] report.

Note that the right most data points on the graph are what has been completed on the day the graph is viewed.

So long as we average more than approximately 200 for the red line on the graph, we're not holding back GIMPS at all....

(Deeper drill-down (read: more graphs) to be made available over the next few days.)

Both the charts and the graph are great. It's interesting that there appear to be one or more heavy hitters in the 72 and above range who report every other day. I'm guessing it's just one, since the "more than" case might imply synchronization. I'm by no means a statistician,but the pattern seems very dramatic, at least at the moment.

GPU to 72 makes Top 10
 

in the year-to-date TF team report.

[url]http://www.mersenne.org/report_top_teams_TF/[/url]

[QUOTE=petrw1;281558]in the year-to-date TF team report.

[url]http://www.mersenne.org/report_top_teams_TF/[/url][/QUOTE]If I made myself a team, I would be in the top 25.

[QUOTE=chalsall;281571]Five seconds.... :smile:[/QUOTE]

Took me 30 seconds... I was focusing on the numbers, based on the thread context...

[QUOTE=Uncwilly;281581]If I made myself a team, I would be in the top 25.[/QUOTE]
I would be 17. And I've only been doing this around 4-5 months.

(nucleon? xyzzy? well actually, that's why GPU272 has made it so high in <1 month...)

[QUOTE=kladner;281541]It's interesting that there appear to be one or more heavy hitters in the 72 and above range who report every other day. I'm guessing it's just one, since the "more than" case might imply synchronization. I'm by no means a statistician,but the pattern seems very dramatic, at least at the moment.[/QUOTE]

Yeah... it makes the [URL="http://www.gpu72.com/reports/overall/graph/month/"]Monthly LL chart[/URL] rather hell to read at the moment, but I'm not going to complain about all the work being provided... :smile: (And you're correct: it's primarily one of our largest contributors.)

BTW, on this new page, simply click on the charts to switch between Weekly and Monthly views. This is accessible from the menus (subs of the OSP), or by clicking on the chart on the Overall System Progress page.

[QUOTE=chalsall;281698]Yeah... it makes the [URL="http://www.gpu72.com/reports/overall/graph/month/"]Monthly LL chart[/URL] rather hell to read at the moment, but I'm not going to complain about all the work being provided... :smile: (And you're correct: it's primarily one of our largest contributors.)

BTW, on this new page, simply click on the charts to switch between Weekly and Monthly views. This is accessible from the menus (subs of the OSP), or by clicking on the chart on the Overall System Progress page.[/QUOTE]

A few months ago someone asked why PrimeNet would almost double every few days. I think xyzzy said he only uploaded results every couple days, and was big enough to do that.

[QUOTE=chalsall;281698]Yeah... it makes the [URL="http://www.gpu72.com/reports/overall/graph/month/"]Monthly LL chart[/URL] rather hell to read at the moment, but I'm not going to complain about all the work being provided... :smile: (And you're correct: it's primarily one of our largest contributors.)[/QUOTE]

Nothing a run through some sort of low-pass filter won't fix, if you're looking for more feature requests.

Workers' progress graphs...
 

Just a quick note... Some of you might find the (temporary) links from the [URL="http://www.gpu72.com/reports/workers/"]Workers' Progress[/URL] page interesting...

And, yup, it's mostly Xyzzy who's causing the [URL="http://www.gpu72.com/reports/workers/overall/graph/"]Sawtooth pattern on the LL Range graph[/URL]....

In the range of 40M to 50M there are 128k expos TFed to 68 bits only. On the range 45M to 56M where GPU272 has its den, there are 41k of them too. Actually, they are all in 45M to 48M range, as there is none left to 68 for higher expos, but that's not the point. The point is [B]why GPU272 can't get its claw on ANY of these exponents???[/B]

I am trying since yesterday evening to grab some TF work at the LL-front, but I only want to TF from 68 (or lower, but this is not possible, as there are no expos with lower TF limit in the range) up to 69. That is one bit only, and the reason is I am testing some GPU settings and I want the results to come out fast, but I don't want to waste the time re-doing known results (my experiments could influence the speed, but not the correctness of the results, so I don't want to re-do known results to see if I get the same things. In fact, I am interested in some "best compromise" in running mfaktc/cudalucas together with some graphic-intensive application as AutoCad or AltiumProtel - of course you guessed I did not invest that huge amount of money in my computer only to crunch prime numbers :P).

So, we are talking about at least 40000 exponents, if not 128000 of them. The search on GPU272 "get new assignment" page always comes out empty handed.

edit: I use the optional range as default, 0-55M, select TF to 69. It can't fulfill even the request for a SINGLE exponent. It always responds with "sorry, no available exponents which match your criteria". C'mon! Get your claw on few of them.

@ LaurV: Those are the sort of assignments I look for, though I am happy starting out from 69.

Of course, that is what I have to do, too. I was just wondering....

[QUOTE=Dubslow;281586]

(nucleon? xyzzy? well actually, that's why GPU272 has made it so high in <1 month...)[/QUOTE]

If I was my own team, I'd be ranked 32 in 24hrs (ish), 14 in a week (ish), 5 in a month (ish), ranked 1 in about 150days (ish).

These GPUs are thundering a long. No wonder if you look at things like top500.org you see GPUs in the higher ranked clusters.

-- Craig

[QUOTE=LaurV;281746]Of course, that is what I have to do, too. I was just wondering....[/QUOTE]

Most of those exponents already have one LL test done so are not grabbed.

Response presumably deleted
 

[QUOTE=garo;281751]Most of those exponents already have one LL test done so are not grabbed.[/QUOTE]:coffee:

Question and comment, in reverse order:
In the Cost per Factor reports, there are infinities shown in some of the GHz Days cells, even when GHz Days should be very low, less than 100.

Do exponents either assigned for LL-test or completed LL-test count as 'Returned to PrimeNet'?

[QUOTE=Dubslow;281813]In the Cost per Factor reports, there are infinities shown in some of the GHz Days cells, ...[/QUOTE]

I believe it is a case of division by zero. :)

[QUOTE=RichD;281815]I believe it is a case of division by zero. :)[/QUOTE]

Indeed. Then the question is, did chalsall program the infinity, or did it appear on it's own? If the latter, that's awesome.

[QUOTE=Dubslow;281813]In the Cost per Factor reports, there are infinities shown in some of the GHz Days cells, even when GHz Days should be very low, less than 100.[/QUOTE]

Indeed. But as RichD pointed out, for the particular ranges and bit-levels in question, one or more test was run but no factor was found. With additional samples, the values should converge on the theoretically expected value. That table is based entirely on samples we've seen -- it does not draw from the PrimeNet data set.

And to answer your follow up question, no, the infinity symbols had to be hand coded, or else Perl would have thrown a divide by zero exception.

[QUOTE=Dubslow;281813]Do exponents either assigned for LL-test or completed LL-test count as 'Returned to PrimeNet'?[/QUOTE]

They should, but they don't. The quantity is so small I don't think it's worth bothering.

Well there's 100+ assigned, and that's not insignificant as far as 'exponents been fully factored', where that's one possible interpretation of 'returned to PrimeNet'. And really, they were returned, because they're no longer assigned to "GPU Factoring".

Xyzzy broke the graph....
 

[url]http://www.gpu72.com/reports/overall/graph/[/url]

BAD Xyzzy, bad...just kidding

[QUOTE=petrw1;282252][url]http://www.gpu72.com/reports/overall/graph/[/url][/QUOTE]

I think that when something breaks it's usually the work of the trolls, usually the fish get blamed.

These GPU thingies are just amazing. We are averaging a GHz-day every 53 seconds. How insane is that?

:max:

so, about 1272 GHz/day by 24H? this is scary

[QUOTE=Xyzzy;282255]These GPU thingies are just amazing. We are averaging a GHz-day every 53 seconds. How insane is that?[/QUOTE]

Ever heard of the "glass ceiling"?

Tinkle, tinkle, tinkle.... :smile:

How can we be sure your percentages are only 6 digits?

[QUOTE=firejuggler;282256]so, about 1272 GHz/day by 24H? this is scary[/QUOTE]It would be (60*60*24) / 53 = 1630 GHz-days/day. Which is like 5x GTX 580s... :cool:

If I run my 8800GT for a month and a half, I can do what Xyzzy does in a day. :surprised

[QUOTE=James Heinrich;282264]It would be (60*60*24) / 53 = 1630 GHz-days/day. Which is like 5x GTX 580s... :cool:

If I run my 8800GT for a month and a half, I can do what Xyzzy does in a day. :surprised[/QUOTE]

What scares me a little bit more is Craig Meyers appears to have even more fire power!!! :smile:

[URL="http://www.gpu72.com/reports/workers/overall/graph/"]Top 10 Workers[/URL]

[QUOTE=chalsall;282267]What scares me a little bit more is Craig Meyers appears to have even more fire power!!! :smile:

[URL="http://www.gpu72.com/reports/workers/overall/graph/"]Top 10 Workers[/URL][/QUOTE]

It amazes me that I am in that exalted group (Top 10) with an overclocked GTX 460.:whistle:
Good thing TheJudger (or many others) are not players in this field.

[QUOTE]so, about 1272 GHz/day by 24H? this is scary[/QUOTE]Not enough to catch nucleon, though.

:sad:

[QUOTE]It would be (60*60*24) / 53 = 1630 GHz-days/day. Which is like 5x GTX 580s...[/QUOTE]We must have messed up our math somewhere. We are seeing around 1275 GHz-days/day. How about 68 seconds?

:ouch1:

[QUOTE=Xyzzy;282272]Not enough to catch nucleon, though.

:sad:[/QUOTE]

Yeah?? Watch me! :P
(edit: I was doing DC's up to now, switched to TF yesterday morning. I will let it on its own till after the weekend, that means no other things running on the computer, as I would be out of town)

[QUOTE=firejuggler;282256]so, about 1272 GHz/day by 24H? this is scary[/QUOTE]
[QUOTE=James Heinrich;282264]It would be (60*60*24) / 53 = 1630 GHz-days/day.[/QUOTE]
[QUOTE=Xyzzy;282273]We must have messed up our math somewhere. We are seeing around 1275 GHz-days/day. How about 68 seconds?[/QUOTE]
86400 seconds / 1275GHz-days = 67.8 seconds/1GHz-day. Still impressive!
What puzzles me is how [i]firejuggler[/i] got the right value out of the wrong math. :unsure:

[QUOTE=kladner;282270]It amazes me that I am in that exalted group (Top 10) with an overclocked GTX 460.:whistle:[/QUOTE]I shall have to go out today and buy me a Sandy-E and a GTX 570 to try and compete. :flex:
My 8800GT is barely keeping me in the Top-30, and I'm only there cause of P-1 work.

I did (53/60)*60*24 instead of (60/53)*60*24

[QUOTE=kladner;282270]It amazes me that I am in that exalted group (Top 10) with an overclocked GTX 460.:whistle: [/QUOTE]

What the hell? How overclocked is it? When I was in Windows mine was at 850 MHz but I'm barely inside top 20...

[QUOTE=Dubslow;282304]What the hell? How overclocked is it? When I was in Windows mine was at 850 MHz but I'm barely inside top 20...[/QUOTE]

Currently at 875MHz. It will run higher but started having errors in MemtestG80. It ran clean at 880 or a little more, but I backed it down some to leave a margin. It passes the long self-test in mfaktc.

I should have said which Top 10 I was referring to. It is this one:

[URL]http://www.gpu72.com/reports/workers/overall/graph/[/URL]

As to "barely in" the same can be said of my position at 9. I'm sure this too shall pass. Another super-cruncher or two like Xyzzy would shove me right out, but so would two more KyleAskine's or Chuck's.

I got crossed up when I joined the forum, so I'm kladner here, and ktony in Primenet records.

I'm 19 in chalsall's list. Like 47 in PrimeNet. Currently around 2.9K credit reported to GPU272.

Yes I'm Craig Meyers on the gpu-to-72 list, nucleon on the forums and the website [url]http://www.mersenne.org[/url].

As far as I know I have the fastest gpu farm on gimps. I think Xyzzy is number2.

I'm responsible for about 3% ish of the entire GHz-days metric across the entire project at the moment. That's an interesting thought :) I'm not sure my percentage on the sub project.

I cracked number 1 for the last 12 months yesterday. Woo :)

My farm consists of:
- 4x GTX580s
- 2x GTX560TI
- 1x GTX460
- 1x GT440

The 460 and 440 are doing LL DCs though. One of the GTX580 is running at 80% capacity. Only because I have run out of CPUs to supply them. :)
You were would need more than 5x GTX580s to beat me.

I only live in a single bedroom high rise unit. So power is probably an issue. I'm not to keen to buy more CPUs. I am keen to upgrade to Ivy Bridge E when it comes out. If it's released at 8cores, I'm there.

My 15-day average is 1421GHz-days/day. I think peak is around 1467GHz-days/day. I'm lower than peak for the average as I had a large outage window with one of my nodes. Heat is a major issue.

I'm keen to do a write up with more details and photos of my farm if anyone is interested.

To keep in line with Xyzzy's stat - I'm 1GHz-day per 60.8secs.

-- Craig

I think 'interested' qualifies as an understatement. Yes please :smile:

[QUOTE=nucleon;282377]Yes I'm Craig Meyers on the gpu-to-72 list, nucleon on the forums and the website [URL]http://www.mersenne.org[/URL].

As far as I know I have the fastest gpu farm on gimps. I think Xyzzy is number2.

I'm responsible for about 3% ish of the entire GHz-days metric across the entire project at the moment. That's an interesting thought :) I'm not sure my percentage on the sub project.

I cracked number 1 for the last 12 months yesterday. Woo :)

My farm consists of:
- 4x GTX580s
- 2x GTX560TI
- 1x GTX460
- 1x GT440

The 460 and 440 are doing LL DCs though. One of the GTX580 is running at 80% capacity. Only because I have run out of CPUs to supply them. :)
You were would need more than 5x GTX580s to beat me.

I only live in a single bedroom high rise unit. So power is probably an issue. I'm not to keen to buy more CPUs. I am keen to upgrade to Ivy Bridge E when it comes out. If it's released at 8cores, I'm there.

My 15-day average is 1421GHz-days/day. I think peak is around 1467GHz-days/day. I'm lower than peak for the average as I had a large outage window with one of my nodes. Heat is a major issue.

I'm keen to do a write up with more details and photos of my farm if anyone is interested.

To keep in line with Xyzzy's stat - I'm 1GHz-day per 60.8secs.

-- Craig[/QUOTE]

As I said. I'm amazed to be in this demographic, even for a while.

[QUOTE=nucleon;282377]
You were would need more than 5x GTX580s to beat me.
....
Heat is a major issue.
....
To keep in line with Xyzzy's stat - I'm 1GHz-day per 60.8secs.
[/QUOTE]
How about a tandem of Teslas? :P

They are much faster on CudaLucas, which unfortunately does not give the big credit given by mfaktc. To stay in line with your statements, I am the one who did 24 DC's for the project from the total of 47, so more then half, and I am on the same position as you if we talk about how many exponents we cleared. You have 1+32, I have 24+9 (DC plus factors). I think only kyle, pete and xyzzy beat us.

Unfortunately, with mfaktc the speed compares to 580's, it seems that mfaktc is not able to get all the juice from the Teslas...

To stay more in line: I could get 1GHz-day every 126.31 secs if I run them maxed, as maxed as the CPU's would let me do it, because that is a problem too. Unfortunately they never ran "mfaktc only", I would have no CPU power for it. I pour some CL from time to time and I have other things to do too (CAD software mostly need that power).

The BIGGEST issue is also the heat, and that is more acute for the board in the middle, which is squeezed between the other board below and the CPU's fan above, and never gets enough flow of cool air.

If I ever succeed to go to water-cooling, then we talk... :P

[QUOTE=LaurV;282385]
You have 1+32, I have 24+9 (DC plus factors). I think only kyle, pete and xyzzy beat us.

[/QUOTE]

Not letting you get away with that :)

I'm sorry but that's not fair :)

For the benefit of the project, I cleared a lot of 71-72 TFs. Which has similar chance of factors as 69-70, with 71-72 taking 4x the effort of 69-70.

I also have 4x DCs done outside of the GPU-to-72 project. Give me a week, and I'll have another 3more done.

If you're going to do that, I'll go grab the 69-70 range currently on the system and have it completed in under a week and see how many that generates. :)

I'm sorry you can't use factors found as a strict comparison.

Yes I'm aware of the issues with comparing LL GHz vs TF GHz.

If you want to combine LL, TF, and P-1 work done into one metric that is fair, then I'd like to see a ranking.

I'm thinking you need to find something like average number of TF GHz-days to find a factor (say x). And use that to reduce TF GHz-days down to a work-unit style metric, and reduce LL GHz-days down to a similar metric.

-- Craig

The ultimate goal of the project is to find primes, so therefore, the next-to-ultimate goal is to clear as many expos as possible in as short as possible time. Trial factoring that results into "no factors" does not give us too much gain (don't jump on my head, George said it, and that is the reason they do not really care about checking that "no factors" reports are not lies, see the discussion about adding a checksum (secret key) to mfaktc "no factor" lines, same as P95 is doing, nobody is interested in doing that, and they classified my complain as "overreacting").

So, what we do here? Exactly! We clear exponents. The rest is just talking, arguing, having fun, "my card is better that yours", "my cat can drink more milk" etc :smile:

The rest, I agree with everything you said. It is impossible to compare different worktype on different platforms (CPU, GPU, etc). We are still arguing about that on a parallel thread on this very subforum. And the discussion is and it will be... endless.

[QUOTE=LaurV;282401] We are still arguing about that on a parallel thread on this very subforum. And the discussion is and it will be... endless.[/QUOTE]
Reposted for emphasis. We got nowhere quickly, with lots of arguing. The problem is determining the scaling factors nucleon mentioned is incredibly difficult to do with any sort of accuracy at all.

[QUOTE=nucleon;282377]My farm consists of:
- 4x GTX580s
- 2x GTX560TI
- 1x GTX460
- 1x GT440[/QUOTE]

Wow!!!

[QUOTE=nucleon;282377]I'm keen to do a write up with more details and photos of my farm if anyone is interested.[/QUOTE]

I would be very interested.

I'd also be interested in knowing your KWh stats, and how you cool all of that!!! A/C, or vent outside? (Unless, of course, you also use it to heat your living space during the winter (assuming you live somewhere where it gets cold during the winter -- I don't enjoy that "luxury"... :smile:))