I've grabbed 616 that should be finished in the next five days or so.

I can do 42 per day.....so if Mark provides the buffer I can take it from there and provide the 40 required for steady state.

[QUOTE=petrw1;403078]I can do 42 per day.....so if Mark provides the buffer I can take it from there and provide the 40 required for steady state.[/QUOTE]

Coolness. Thanks guys. :smile:

[QUOTE=chalsall;403072]Just so everyone knows, we need a little bit of a surge in DCTF to 72 in order to satisfy the churners with a bit of a buffer.

Anyone so inclined are asked to take some candidates from 71 bits to 72. Preferably from the lowest exponent.

Not yet critical, but welcomed.[/QUOTE]

Can "Let GPU72 decide" take care of this?

[QUOTE=Chuck;403102]Can "Let GPU72 decide" take care of this?[/QUOTE]

Generally "LGPU72D" doesn't change the worktype.

However, if you (or anyone willing) give me explicit permission (either publicly here or privately via PM) it can be done.

If it's an easy thing to do on your end, you have my permission to change what "LG72D" gives me.

EDIT: I should clarify that I mean this as a blanket statement as opposed to specifically for this event (since it appears to already have been dealt with).

[QUOTE=chalsall;403103]Generally "LGPU72D" doesn't change the worktype.

However, if you (or anyone willing) give me explicit permission (either publicly here or privately via PM) it can be done.[/QUOTE]

Do it.

I think we discussed this once: people who want to keep their work type chose "what makes sense", people who don't, chose "let gpu72 decide". That is why we had two in the beginning. Or we can talk Scott into adding a "I don't care" work type to Misfit. In fact, what is the [U]actual, practical[/U], (as opposed to theoretical) difference between WMS and LGPU72D? (on the server side, i.e. what the server does for any of the requests?). We used to discuss this in the past and it was somehow established, at a time when I was caring about. I don't especially care now, because I am "picky" with my choices, each is particularly done to suit the card is fetching for. But it would be nice to be stated again.

Edit: on the bright side, I will bring back about 900-950GHzD/D to LLTF/DCTF/Whatever, immediately after I get rid of Madpoo's triple checks in 24M (about 110 from 119 assignments done already, so this is 2-3 days left). On the dark side my Titan let the magic smoke out today. It was DC-ing Manfred's assignment in 332M (we did it in parallel to check the residues, and coordinate on the PM on this forum). It was close to 150M iterations (we had a match to 145M, he is a bit more advanced, as he is the assignee, I periodically had to stop and switch to other job, to let him lead the way). Now, I will have to look for some power IC which exploded (that is Q16 and L12 coil, for who knows the PCB layout of the EVGA Titan classical design).

[QUOTE=LaurV;403118]On the dark side my Titan let the magic smoke out today. It was DC-ing Manfred's assignment in 332M (we did it in parallel to check the residues, and coordinate on the PM on this forum). It was close to 150M iterations (we had a match to 145M, he is a bit more advanced, as he is the assignee, I periodically had to stop and switch to other job, to let him lead the way). Now, I will have to look for some power IC which exploded (that is Q16 and L12 coil, for who knows the PCB layout of the EVGA Titan classical design).[/QUOTE]

:(

From what I understand, What Makes Sense honours bit level settings and Let GPU72 Decide honours the work type.

I like the idea of a "I Don't Care" option, where GPU72 has complete control.

[QUOTE=LaurV;403118]In fact, what is the [U]actual, practical[/U], (as opposed to theoretical) difference between WMS and LGPU72D? (on the server side, i.e. what the server does for any of the requests?).[/QUOTE]

The difference internally is WMS honours the Pledge level (e.g. 73), while LGPU72D sets it (usually to 75, sometimes to 74).

[QUOTE=chalsall;403074]For a steady state, about 40.[/QUOTE]

Good job I'm still doing 48 a day then.

[QUOTE=chalsall;402816]George asked me to bring in those candidates in the 332M range which are not yet appropriately TF'ed. I brought in approximately 3,500 candidates between 332192831 and 332.6M which had not been assigned to registered users, and which were TF'ed to less than 78 bits.[/QUOTE]
Can you have "Fetch Spider" acquire all expos above 332792600 (up to 333M) that are at 72?
That, with what I have queued, will finish off the bit level from 332192831 out to 333M.

Not that I am doing much, but I have switched to DCTF once my current assignments complete.

What option do I choose to let GPU72 decide whether to do LL TF or DC TF? I can't figure it out.

edit: on misfit

[QUOTE=Uncwilly;403216]Can you have "Fetch Spider" acquire all expos above 332792600 (up to 333M) that are at 72?
That, with what I have queued, will finish off the bit level from 332192831 out to 333M.[/QUOTE]

Done. Actually, I've brought in everything below 78 bits not already assigned.

[QUOTE=casmith789;403228]Not that I am doing much, but I have switched to DCTF once my current assignments complete.[/QUOTE]

Thanks

[QUOTE=casmith789;403228]What option do I choose to let GPU72 decide whether to do LL TF or DC TF? I can't figure it out. edit: on misfit[/QUOTE]

For legacy reasons, there isn't a way to have GPU72 assign work types different than what is requested (i.e. LLTF or DCTF). This is probably something which should be retired -- my original thinking was that some definitely wanted to do LLTF, while others definitely wanted to do DCTF.

When I have some cycles I'll update the code such that the "Let GPU72 decide" LLTF requests will assign DCTF if required. Anyone who definitely wants to do LLTF please choose any other option (including "What Makes Sense"), preferably to 75 if you're willing to go that far.

[QUOTE=Gordon;403215]Good job I'm still doing 48 a day then.[/QUOTE]

I should have been clearer. When I said "about 40", I meant about 40 more than we were doing.

[QUOTE=chalsall;403237]Done. Actually, I've brought in everything below 78 bits not already assigned.[/QUOTE]
What I was asking for was above your original 332.6M cut-off.

[QUOTE=Uncwilly;403241]What I was asking for was above your original 332.6M cut-off.[/QUOTE]

And that's what I did, but not just those at 72 bits.

[QUOTE=chalsall;403239]I should have been clearer. When I said "about 40", I meant about 40 more than we were doing.[/QUOTE]

Covered....as I mentioned a couple days ago....my 42/day will start later today once current LLTF completed

[QUOTE=petrw1;403243]Covered....as I mentioned a couple days ago....my 42/day will start later today once current LLTF completed[/QUOTE]

Thanks! To you, and everyone who answered the "surge" request, and/or gave permission for a LLTF to DCTF change for LGPU72D.

Based on what is queued we're looking good for staying ahead of the Cat 4 DC'ers (mostly AKA "the churners"), but probably next week I'll update the assignment code such that LLTF LGPU72D requests may occasionally assign some DCTF work.

This will probably not be needed to be used (unless a new MP is found, and we get a new surge of Newbies like last time). Also, please know that this will only be used to stay ahead of the LL'ing Cat 4 wave front. Or, in other words, only working from 71 bits to 72 at the DCTF'ing wave back.

I've just switched over to DCTF, and I was assigned this exponent from GPU72 (along with some other similar ones):

78986359 DC TF 71 74

As far as I can tell that exponent has never had an LL done, and it is much too high to be the current wavefront for double checks, so I am a bit confused. What am I missing?

[QUOTE=Mark Rose;403136]I like the idea of a "I Don't Care" option, where GPU72 has complete control.[/QUOTE]I would also like to see that option.

[QUOTE=chalsall;403253]Also, please know that this will only be used to stay ahead of the LL'ing Cat 4 wave front. Or, in other words, only working from 71 bits to 72 at the DCTF'ing wave back.[/QUOTE]

A little confused....

If I understand correctly the "LL'ing Cat 4 Wave Front" is about 72.6M
Isn't the "DCTF'ing wave back" about 58M?

I am currently taking the lowest DCTF to 72 Bits. About 40.8M.
Is this not what you were needing for the DC Churners?

[QUOTE=casmith789;403406]What am I missing?[/QUOTE]

You Let GPU72 Decide, and pledged to go to 75 bits.

This doesn't make sense for DCTF'ing.

[CODE]| Chris Smith | 2015-06-02 18:57:52 | DF(9) -- N: 1 G: 150 P: 75 L: 0 H: 100000000 -- n: 100 p: 74 l: 66000000 h: 80000000 -- A: 4[/CODE]

[QUOTE=petrw1;403409]A little confused....

If I understand correctly the "LL'ing Cat 4 Wave Front" is about 72.6M
Isn't the "DCTF'ing wave back" about 58M?

I am currently taking the lowest DCTF to 72 Bits. About 40.8M.
Is this not what you were needing for the DC Churners?[/QUOTE]

I misspoke. s/stay ahead of the LL'ing Cat 4 wave front/stay ahead of the DC[LL]'ing Cat 4 wave front/ should be more correct.

[QUOTE=chalsall;403410]You Let GPU72 Decide, and pledged to go to 75 bits[/QUOTE]If that was me, I'd assume if I selected "Let GPU72 decide" and "75" that it would interpret it as "Give me something to do, whatever is most useful right now, and I'm willing to TF as high as 2[sup]75[/sup] if that's required".

[QUOTE=chalsall;403410]You Let GPU72 Decide, and pledged to go to 75 bits.

This doesn't make sense for DCTF'ing.[/QUOTE]

Fsck me...

Thanks, you've pointed out a bug. LGPU72D for DCTF hasn't had an audit recently. Sorry about that.

[QUOTE=James Heinrich;403412]If that was me, I'd assume if I selected "Let GPU72 decide" and "75" that it would interpret it as "Give me something to do, whatever is most useful right now, and I'm willing to TF as high as 2[sup]75[/sup] if that's required".[/QUOTE]

LOL...

I'm doing the best I can with the time allowed.

Ah okay. MISFIT greys out that option so I just assumed it wasn't important for LGPU72 decide.

[QUOTE=casmith789;403416]Ah okay. MISFIT greys out that option so I just assumed it wasn't important for LGPU72 decide.[/QUOTE]
It wasn't. You just helped Chris fix a bug, read the discussion above, with James, who took up the glove.

curtisc????
 

[QUOTE=chalsall;403072]Just so everyone knows, we need a little bit of a surge in DCTF to 72 in order to satisfy the churners with a bit of a buffer.

Anyone so inclined are asked to take some candidates from 71 bits to 72. Preferably from the lowest exponent.

Not yet critical, but welcomed.[/QUOTE]

I just happened to notice that recently curtisc is starting to do some DC.
And "some" to curtisc is no small potatoes.

I have no way of knowing if he chose to allocate some CPUs to DC or if it is via this CPU Setting:
[CODE]DC instead of LL percentage* [/CODE]

In the last 90 days he still has 27905 LL and "only" 525 DC but in the last
[url]http://www.mersenne.org/report_recent_cleared/[/url]
report over 40% are DC.

SO........... if curtisc is even temporarily doing "some" DC its going to be more than "a bit of a surge".

I'll leave it you to handle as you so capably do so....

[QUOTE=petrw1;403497]I'll leave it you to handle as you so capably do so....[/QUOTE]

Thanks for the vote of confidence (sincerely :smile:).

At the moment curtisc has 3,230 LL and 249 DC assignments. The latter are taking ~8 days each; in other words he's doing ~31 DCs a day.

I also don't know why this is happening, but we're still "good". It is possible that these are new machines which haven't yet "proven" themselves to Primenet. If this is the case, we may soon need a "surge" in LLTF!

[QUOTE=chalsall;403504]Thanks for the vote of confidence (sincerely :smile:).

At the moment curtisc has 3,230 LL and 249 DC assignments. The latter are taking ~8 days each; in other words he's doing ~31 DCs a day.

I also don't know why this is happening, but we're still "good". It is possible that these are new machines which haven't yet "proven" themselves to Primenet. If this is the case, we may soon need a "surge" in LLTF![/QUOTE]

I think you need a new GIMPS Job Title:
"The Surge Sarge"

[QUOTE=chalsall;403411]I misspoke. s/stay ahead of the LL'ing Cat 4 wave front/stay ahead of the DC[LL]'ing Cat 4 wave front/ should be more correct.[/QUOTE]

How is the balance, currently, between LLTF and DCTF in the above categories?

[QUOTE=kladner;403506]How is the balance, currently, between LLTF and DCTF in the above categories?[/QUOTE]

We're currently a little heavy on the DCTF'ing (as requested).

For those who use automated systems to fetch work (such as [URL="http://www.mersenneforum.org/misfit/"]MISFIT[/URL]), please choose "LLTF; Let GPU72 Decide". My Spiders and AIs will then make decisions based on the real-time heuristics.

[QUOTE=chalsall;403504] It is possible that these are new machines which haven't yet "proven" themselves to Primenet. If this is the case, we may soon need a "surge" in LLTF![/QUOTE]

Makes sense since they are in the upper DC range.

[QUOTE=chalsall;403507]We're currently a little heavy on the DCTF'ing (as requested).

For those who use automated systems to fetch work (such as [URL="http://www.mersenneforum.org/misfit/"]MISFIT[/URL]), please choose "LLTF; Let GPU72 Decide". My Spiders and AIs will then make decisions based on the real-time heuristics.[/QUOTE]

Tanks! :smile:

[QUOTE=kladner;403513]Tanks! :smile:[/QUOTE]

Tank you. :wink:

[QUOTE=chalsall;403504]taking ~8 days each
<snip>
It is possible that these are new machines[/QUOTE]
These two things don't really match. Even if they don't work 24/7, but only few hours per day, still is a lot slow for new computers. I don't think the university is buying second-hand computers. I think he just reassigned old/not so efficient machines, or they fell under the "DC percentage" told above. Which (in either case) is very good for us.

Re curtisc, thinking more about this, we are just talking here like children, but the real reason is that he [U]knows[/U] where the primes are... As said in the past, he just let us fool around, but now he got pissed off because we are keeping missing the primes in that range, and went for them by himself. So, if he switched few computers to DC, than we have to pay more attention to that range...
:smile:

A minor bug: Looking at the [url=https://www.gpu72.com/reports/workers/dctf/week/]Workers' DC Trial Factoring Progress[/url] page, it doesn't show the assignments I currently have.

Since all my assignments are currently in 53M, my guess is that the query ranged limited and doesn't catch them.

[QUOTE=Mark Rose;403821]Since all my assignments are currently in 53M, my guess is that the query ranged limited and doesn't catch them.[/QUOTE]

The Daily / Weekly / Monthly reports show not only what work as been done, but also what was assigned during that time period.

You took the 4,000 assignments in 53M more than a week ago.

Ahhhh, makes sense. That's useful, now that I know exactly what that column represents! :)

Back to LLTF'ing, please...
 

So everyone knows...

The "surge" request for DCTF was VERY well answered. Thanks!

We're flying a little blind at the moment because of the death of the spidering machine here in Barbados which drove Mersenne.info. We know how much TF'ing work is being done (within a percent or so), but I can't currently tell exactly how much DC'ing and LL'ing is being done in the different ranges / categories. (Rarely a good thing in a "Just in Time" system; this is being worked.)

What I can tell you, with some certainty, is we need to bring some firepower back to LLTF'ing.

Soon.

Interestingly, this is largely needed in order to feed the P-1'ers at at least 74 bits.

So, for those who brought resources to DCTF from LLTF for the surge request, I would ask that you please choose "Let GPU72 Decide" (either worktype (DCTF or LLTF), pledge level doesn't matter). Or choose "What Makes Sense", and pledge to at least 72 (again, either worktype).

(As my sister and I used to ask our parents when we were very young: "Are we having fun yet?" (It was an unusual, but valuable, upbringing. :wink: :smile:))

[QUOTE=chalsall;404025]So everyone knows...

The "surge" request for DCTF was VERY well answered. Thanks!

[B]What I can tell you, with some certainty, is we need to bring some firepower back to LLTF'ing.

Soon.[/B]

Interestingly, this is largely needed in order to feed the P-1'ers at at least 74 bits.

So, for those who brought resources to DCTF from LLTF for the surge request, I would ask that you please choose "Let GPU72 Decide" (either worktype (DCTF or LLTF), pledge level doesn't matter). Or choose "What Makes Sense", and pledge to at least 72 (again, either worktype).

(As my sister and I used to ask our parents when we were very young: "Are we having fun yet?" (It was an unusual, but valuable, upbringing. :wink: :smile:))[/QUOTE]

Soon, as in yesterday? Well, I can't quite give you that, but my own urges led me to stop taking DCTFs in the last day or so. The last such 41M assignment was noted to have finished at 2015-06-13 16:15:49. Since then, (while I was at work) I have been doing LLTF on both GPUs.

It's nice to have my hunches line up with the needs. :smile:

I'm going on a vacation / family reunion thing and I will return July 1.

~1.2 THz will be offline for the next five days.

SSL Cert updated...
 

Just to let everyone know, I updated the SSL Certificate for [url]www.GPU72.com[/url] tonight.

Shouldn't cause anyone any issues, but if you see anything strange which might be related to this, please let me know.

[QUOTE=chalsall;406691]Just to let everyone know, I updated the SSL Certificate for [url]www.GPU72.com[/url] tonight.

Shouldn't cause anyone any issues, but if you see anything strange which might be related to this, please let me know.[/QUOTE]

The intermediate SSL certificate, StartCom, is signed with SHA1 and not SHA256. My browser is giving me a warning. Apparently StartSSL's certificate wizard offers the option to use SHA2/256, but if not, the SHA2 intermediate cert can be found [url=https://www.startssl.com/certs/class1/sha2/pem/sub.class1.server.sha2.ca.pem]here[/url].

Just so everyone knows, my productivity has been down lately and I am fully aware.

My X99 system has been sitting as an open bench system in my room for a few months now and summer has been a bit taxing on the less than optimal cooling system. I am working on getting a proper case for this hardware (which is going to become my primary computer anyway) which will hopefully restore it to 24/7 usage.


I also need to redo my other system's liquid cooling. The use of distilled water without any additives has caused problematic amounts of algae growth. For normal usage, the system is fine but under full load, it can't keep up on a hot day.

Likewise my i7-920 system has been... quirky. At first it stopped wanting to run mfaktc on one GPU, then on both GPUs, then finally Prime95 itself started spewing errors. After downclocking the CPU to stock and under-clocking the RAM it's not spitting errors anymore, I might even try restarting mfkatc if underclocking the CPU/RAM fixed the instability. Doesn't seem to be running hot, but it is 6 years old... :paul:

Have you tried replacing the power supply?

It's not impossible that's the cause, but I don't have a spare one lying around and I'm not ready to buy a new one on a whim that it might be the cause. Maybe when I need to replace the machine anyways I'll give it a shot, but it'll be even older by then :smile:

OTOH, me is back... :flex:

[QUOTE=LaurV;407170]OTOH, me is back... :flex:[/QUOTE]

Welcome back! Where you been? :hello:

Holiday, back to Ro for a month. Now back to Th. Unfortunately only half of the capacity is up, because I can't find the time to refill the water pipes for the biggest machine. At job, there were few hundred mails waiting for me... ye know how it is after holiday... :smile: and the days are a bit longer and busier right now ...

Still didn't get the mosfet for the Titan, only the inductor was waiting on my desk, so the motivation to start the big mill is also not very high.. I could run it only with the 580's but welllll... Waiting for better days right now...

Getting stupid.. replying to myself... (again!)

Who is the Anon user with 8000GHzD/D output? He is now on the second position after Squirrels in the weekly report, but this is only because he started 5 days ago (so not yer a full week completed!).

And offtopic, continuing the former discussion (well, maybe ontopic too, it is GPU-related, and the GPU will do TF when it will be installed, and free :smile:)

Well, I got the mosfet, I replaced it, but the Titan still won't work properly, it looks like it is clocked very low (50MHz!?) and it can not go higher. The problem is that I can't really measure it, voltages, ticks, etc, because I need to install the radiator and put it in the socket (it gets hot without it) and in that position I can not access any measuring points. I will have to solder wires and spend a lot of time with it, and I am not sure I am able to fix it.

Therefore, I was in big doubt if I should "give away" the ~2500GHzDays of credit and send the file for somebody to continue the work. Who didn't know the story, I was DC-ing M332219809 in parallel with Manfred, at his request, we were checking/matching residues every 5M, and I was waiting for him, because my Titan was faster than his CPU, but then my Titan crashed and he continued by himself - and reported the result during my holiday, it was his assignment anyhow, hopefully is correct - we had a match up to 145M iterations, without errors on any side, and I have completed 149M.xxx iterations (very close to the 150M checkpoint, but not reached). My 580 cards are the one with smaller RAM (one G and half), so I could not continue the test. The 3G RAM 580 cards can run the larger FFT well, but I don't have such cards. The hope was to repair the Titan, but the holiday delayed it, right now the chances are very slim to fix it. The DC (if completed) worth about 5000GHzDays, and it is about half done, so I was quite reticent to "give them away" :razz: hehe, so guess what did I do? I looked for another Titan on ebay, and Mike (Xyzzy) is helping me to buy it. :fusion:

[QUOTE=LaurV;408738]The 3G RAM 580 cards can run the larger FFT well, but I don't have such cards.[/QUOTE]I have a 3GB 580, but little interest in LL :razz:
But, if you get desperate, let me know.

Well, the trade for a new Titan is "almost done" and it will be in my rig soon. And I still have (small small tiny tiny) hope to repair the old one. I am not getting desperate, and if I give it away, it would be to a guy (girl?) with a Titan, the time is shorter by half and much better use of the resources. I am sure few guys here have Titans, and even Teslas.

Oh, so you are really desperate about having your assignment get completed? That is good, because if you finally check in your result that would be a final proof, that all of the 4 cores that have been working on that are pretty reliable. :smile:

And that anonymous user is again really interesting - if he continiues to work at that pace it looks like our combined throughput on GPU72 has greatly increased with those two new top score users in the recent past.

[QUOTE=manfred4;408742]And that anonymous user is again really interesting - if he continiues to work at that pace it looks like our combined throughput on GPU72 has greatly increased with those two new top score users in the recent past.[/QUOTE]

Indeed! I don't know who it is either (and if I did know I couldn't tell anyone). But between them they're almost completing work faster than I can bring it in! :smile:

I remember only two weeks when my 1300 GHzD/D felt significant lol

[QUOTE=chalsall;408744]Indeed! I don't know who it is either (and if I did know I couldn't tell anyone). But between them they're almost completing work faster than I can bring it in! :smile:[/QUOTE]

Progress last 2 days (4:00 PM UTC): ... ROUNDED
DISCLAIMER: I know it depends on how often/when each of you check in results:

Mark Rose: +2,700
Air Squirrels: +28,700
ANONYMOUS: +22,300

To compare this shows just over 304,000 total DCTF in the last 30 days:
[url]http://www.gpu72.com/reports/workers/dctf/month/[/url]
47% of it is the Big-2 (Sorry Mark)

If the big-2 above continue at this rate they alone would complete 765,000 in 30 days.

[QUOTE=petrw1;408963]Progress last 2 days (4:00 PM UTC): ... ROUNDED
DISCLAIMER: I know it depends on how often/when each of you check in results:

Mark Rose: +2,700
Air Squirrels: +28,700
ANONYMOUS: +22,300

To compare this shows just over 304,000 total DCTF in the last 30 days:
[url]http://www.gpu72.com/reports/workers/dctf/month/[/url]
47% of it is the Big-2 (Sorry Mark)

If the big-2 above continue at this rate they alone would complete 765,000 in 30 days.[/QUOTE]

LaurV and I used to be the big two, but we've been usurped! It's all good though. I want to see the DCTF finished. The question is whether LaurV will pass me again before we both get passed by these hot shot new comers ;)

[QUOTE=Mark Rose;408972]LaurV and I used to be the big two, but we've been usurped! It's all good though. I want to see the DCTF finished. The question is whether LaurV will pass me again before we both get passed by these hot shot new comers ;)[/QUOTE]
Hehe, I can't catch you. Since I am back from holiday am a bit below your speed rate, i.e. you are about 100Gd/d faster then me, due to the fact that one of my rigs are down, and I will not start it until I get another Titan for it (which was just shipped out from US, but it may take a while). After that it will do LL for a while, to finish some remnant assignments. I used to be "the big 2", "the big 3", "4", "5" etc. (getting older, hehe), with jerry, pete, nucleon, mike, etc, etc. but they all switched priorities in time, and gave up part of their power. For me also, I never had a rig totally dedicated to TF, they all did some other work, some design, some bitcoin, etc. Then other guys big-iron joined the fray (i.e Oliver, chuck), and we lost the cherry for good...

OTOH, now I am very upset, because I had a [URL="http://www.gpu72.com/reports/worker/2423ae6e8f696d5e7d1447de91ca35a6/"]nice cityscape[/URL] there, :wink: looking like NY before nine-eleven, hehe, but only a corner of it remained, it went to the left left and it was lost during my holiday :sad:

I am working to make it again... I have to see how and where I add some green... and maybe a touch of red :razz:

I've not been here and I'm curious about just how much more power you guys have . So when you say he has you beat by 100Gd/d what are the numbers behind that?

[QUOTE=dragonbud20;409057]I've not been here and I'm curious about just how much more power you guys have . So when you say he has you beat by 100Gd/d what are the numbers behind that?[/QUOTE]

If you look at the link LaurV provided for him or anyone else you choose, the graph shows him and Mark in the 1300-1400 Ghz Days / Day of GPU. This requires 2 or 3 upper-end full-time GPUs (or a few more mid-range GPUs).

On the other hand Air Squirrels and ANONYMOUS are in the 9,000-10,000 range.

[QUOTE=dragonbud20;409057]I've not been here and I'm curious about just how much more power you guys have . So when you say he has you beat by 100Gd/d what are the numbers behind that?[/QUOTE]

Well, as a rough ballpark, a top end AMD graphics card will do about 1000 GHz-days/day. A midrange card will tend to perform in the 200-300 range. I run 3 GTX 580's basically 24/7, and each produces about 431 GHz-d/d. I also have an old GT 520, giving 30 GHz-d/d, because why not. I also sometimes use my GTX 760 at home, which gives about 260 GHz-d/d.

AirSquirrels has a large number of top end AMD graphics cards. We don't know what ANONYMOUS has.

The GHz-d/d figure relates to the work a P4 at a certain speed can do in one day, doing a Lucas-Lehmer primality test. GPU's (at least NVidia at the moment) can also run the LL test, but they are vastly more efficient at "simple" division by trying to divide Mersenne numbers by very large factors (trial factoring in the millions of digits). The number you see for TF GHz-d/d is a rough measurement of the effective rate the GPU has at eliminating candidate Mersenne numbers doing trial factoring versus a CPU doing the LL test.

Every additional bit in a trial factor takes double the execute the division though, so trial factoring eventually becomes slower than doing the LL operation on the GPU. This cutoff grows the larger the Mersenne number gets, and is in the range of 72 to 75 bits for the exponents the GPUs are currently factoring.

[QUOTE=Mark Rose;409066]The GHz-d/d figure relates to the work a P4 at a certain speed can do in one day, doing a Lucas-Lehmer primality test.[/QUOTE]
Please see: [URL="http://www.mersennewiki.org/index.php?title=Computing_power"]mersennewiki.org entry "Computing power"[/URL]

[QUOTE=Uncwilly;409078]Please see: [URL="http://www.mersennewiki.org/index.php?title=Computing_power"]mersennewiki.org entry "Computing power"[/URL][/QUOTE]

Ahh, thanks :)

[QUOTE=Mark Rose;409066]Well, as a rough ballpark, a top end AMD graphics card will do about 1000 GHz-days/day. A midrange card will tend to perform in the 200-300 range. I run 3 GTX 580's basically 24/7, and each produces about 431 GHz-d/d. I also have an old GT 520, giving 30 GHz-d/d, because why not. I also sometimes use my GTX 760 at home, which gives about 260 GHz-d/d..[/QUOTE]

huh I guess my 2 980TIs and a 970 aren't too shabby then; perspective is a funny thing. Although since it's summer in SoCal right now I can only run nights.

[QUOTE=dragonbud20;409097]huh I guess my 2 980TIs and a 970 aren't too shabby then; perspective is a funny thing. Although since it's summer in SoCal right now I can only run nights.[/QUOTE]
Ha! You stay at about 1600GHzDays/day with[URL="http://www.mersenne.ca/mfaktc.php?show=631&sort=ghdpd"] those 3 cards[/URL]. More than I am currently doing, by about a half! Move your butt and get an account on gpu72! :razz:


Edit: actually... I see you have an account there already, but you only produce ~280 GHzD/D, why so? Or is that not you? Put your cards to do something useful, and forget those games.. They only waste your time :razz:

More LLTF'ing please...
 

Just to put on the table...

We recently received, unexpectedly, a great deal of DCTF'ing firepower. This is great, of course.

But... We're still right at the edge of properly feeding the LL'ers and P-1'ers. Slowly falling behind, in fact.

So, those who currently aren't sure if they should DCTF or LLTF, please choose LLTF'ing.

We now return you to your regular programming.... :smile:

[QUOTE=dragonbud20;409097]huh I guess my 2 980TIs and a 970 aren't too shabby then; perspective is a funny thing. Although since it's summer in SoCal right now I can only run nights.[/QUOTE]

Those cards not only do a lot more work than my two 580s, but at a lazy guess, they use less power between them.

[QUOTE=kladner;409118]Those cards not only do a lot more work than my two 580s, but at a lazy guess, they use less power between them.[/QUOTE]
Your 580's, however, are fairly good at LL. Those 9X0's aren't.

[QUOTE=frmky;409125]Your 580's, however, are fairly good at LL. Those 9X0's aren't.[/QUOTE]

In the last day or so, I began testing CUDALucas on one of the cards. So far, it will complete the self test, but crashes the driver doing cufftbench or threadbench. I have it quite throttled back both the GPU and memory. I may try the same things on the other card and see if it might be hardware related.

You might try increasing tdrdelay. I had the same issue until I upped that from the default of (I think) 2 seconds to something more like 10 or 20. I still occasionally get a crash, but it greatly increased the reliability. You can also use a looping batch file to have CUDALucas restart automatically when it crashes and stop when there's nothing left in the worktodo file (or stop when a results file is generated).

It's currently summer in SoCal which means I mainly run at night to avoid coming home to a molten pool of computer on the floor :furious: Also admittedly my 980TIs are both heavily overclocked and can put out around 800-850 GHzD/D each sooo yeah once winter starts to roll in I'll really be kicking ass until then it will be slow going.

[QUOTE=wombatman;409131]You might try increasing tdrdelay. I had the same issue until I upped that from the default of (I think) 2 seconds to something more like 10 or 20. I still occasionally get a crash, but it greatly increased the reliability. You can also use a looping batch file to have CUDALucas restart automatically when it crashes and stop when there's nothing left in the worktodo file (or stop when a results file is generated).[/QUOTE]

I am a bit slow tonight. "tdrdelay"? Is this a CuLu setting? Does it apply to cufftbench and threadbench?

I think I have an example of a looping batch file around. I thought that was used to work around random crashes when doing LL work. Would the same apply to the bench functions?

Thanks! I appreciate the reply. I know there are considerable discussions of CuLu stopping while doing LL here on the forum. It has been a while since I read them.

I still haven't tried the same functions on the other card. The cards in question are a 797 MHz EVGA with a Zalman aftermarket cooler, and a 833 MHz MSI "Lightning". The latter is certainly the better of the two. It runs faster and cooler on less voltage than the EVGA. My previous experience with CuLu taught me that memory clocks on consumer GPUs are too high for DP calculations. I have mine set back from over 2 GHz to 1.6 GHz.

The looping file won't work for the benchmarks. You might have to do them in pieces instead.

Here's a link about TdrDelay. The long and short is that Windows will kill the video driver after 2 seconds of non-responsiveness or something like that. But CuLu sometimes does that without actually locking everything up, and Windows restarts the driver as though it crashed.

[url]http://http.developer.nvidia.com/NsightVisualStudio/2.2/Documentation/UserGuide/HTML/Content/Timeout_Detection_Recovery.htm[/url]

And here's a link that shows how to directly edit it in the registry: [url]https://www.reddit.com/r/battlefield_4/comments/1xzzn4/tdrdelay_10_fixed_my_crashes_since_last_patch/[/url]

I believe mine's set to 20 now. Like I said, you will still get an occasional crash, so I would use a looping batch file, but I definitely saw much much fewer crashes and restarts with the increase TdrDelay.

[QUOTE=wombatman;409154]The looping file won't work for the benchmarks. You might have to do them in pieces instead.

Here's a link about TdrDelay. The long and short is that Windows will kill the video driver after 2 seconds of non-responsiveness or something like that. But CuLu sometimes does that without actually locking everything up, and Windows restarts the driver as though it crashed.

[URL]http://http.developer.nvidia.com/NsightVisualStudio/2.2/Documentation/UserGuide/HTML/Content/Timeout_Detection_Recovery.htm[/URL]

And here's a link that shows how to directly edit it in the registry: [URL]https://www.reddit.com/r/battlefield_4/comments/1xzzn4/tdrdelay_10_fixed_my_crashes_since_last_patch/[/URL]

I believe mine's set to 20 now. Like I said, you will still get an occasional crash, so I would use a looping batch file, but I definitely saw much much fewer crashes and restarts with the increase TdrDelay.[/QUOTE]

Thanks very much! I will try this at once. I see Timeout Detection Recovery, above now, which sort of rings a bell. I really appreciate the quick and detailed response.

EDIT: Both FF and IE are balking at the Reddit link, citing certificate problems.

[QUOTE=chalsall;409116]Just to put on the table...

We recently received, unexpectedly, a great deal of DCTF'ing firepower. This is great, of course.

But... We're still right at the edge of properly feeding the LL'ers and P-1'ers. Slowly falling behind, in fact.

So, those who currently aren't sure if they should DCTF or LLTF, please choose LLTF'ing.

We now return you to your regular programming.... :smile:[/QUOTE]


I had to take the system down for 12 hours yesterday for a chiller upgrade, but I can transition some more power to LL. Is there an easy way to see our current pace vs. demand?

[QUOTE=kladner;409155]Thanks very much! I will try this at once. I see Timeout Detection Recovery, above now, which sort of rings a bell. I really appreciate the quick and detailed response.

EDIT: Both FF and IE are balking at the Reddit link, citing certificate problems.[/QUOTE]

Weird...Chrome had no issue with it. Here's the relevant bit:
[QUOTE]Simply go to your registry: Start --> Run --> Type in "Regedit"
Go to "HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers" and create a key of type "DWORD (32-bit)" and name "TdrDelay" with a value of 10 as Decimal value.[/QUOTE]

[QUOTE=airsquirrels;409173]I had to take the system down for 12 hours yesterday for a chiller upgrade, but I can transition some more power to LL.[/QUOTE]

That would be very much appreciated. Perhaps 50/50, at least for a week or so?

[QUOTE=airsquirrels;409173]Is there an easy way to see our current pace vs. demand?[/QUOTE]

It's a bit tricky (many "waves", etc), but if you look at [URL="https://www.gpu72.com/reports/available/"]this report[/URL] and don't see at least a few hundred candidates available in the "75" column, then we're short feeding the P-1'ers.

[QUOTE=wombatman;409174]Weird...Chrome had no issue with it. Here's the relevant bit:[/QUOTE]

Again, many thanks! I did spend a while this morning running chunks of cufftbench.bat on the MSI card. It runs for a while, stops, I start it again, it zips through what it's already done, and then does some more. No error messages. The EVGA card always quit with an 'unknown error."

DCTF to 70 now fully assigned...
 

Just so everyone knows, all DCTF candidates to 70 are now assigned.

I have adjusted the DCTF assignment page such that if MISFIT or mfloop is the requester and the "pledge" is 70, it is automatically bumped up to 71. This is so no workers are left without work.

[QUOTE=chalsall;409175]It's a bit tricky (many "waves", etc), but if you look at [URL="https://www.gpu72.com/reports/available/"]this report[/URL] and don't see at least a few hundred candidates available in the "75" column, then we're short feeding the P-1'ers.[/QUOTE]

when you say the 75 column do you mean the 67-79M table or the 332.2-332.9M table?

[QUOTE=kladner;409186]Again, many thanks! I did spend a while this morning running chunks of cufftbench.bat on the MSI card. It runs for a while, stops, I start it again, it zips through what it's already done, and then does some more. No error messages. The EVGA card always quit with an 'unknown error."[/QUOTE]

Please excuse the forgoing ignorant drivel. I completely misunderstood what was happening and what was being said.

After the registry change, the MSI card completed:[INDENT]CUDALucas -cufftbench 1024 8192 5
CUDALucas -threadbench 1024 8192 5
CUDALucas -r 0
[/INDENT]I'll run -r 1 overnight. It takes quite a while, doesn't it?

[QUOTE=dragonbud20;409228]when you say the 75 column do you mean the 67-79M table or the 332.2-332.9M table?[/QUOTE]

The former.

[QUOTE=kladner;409249]Please excuse the forgoing ignorant drivel. I completely misunderstood what was happening and what was being said.

After the registry change, the MSI card completed:[INDENT]CUDALucas -cufftbench 1024 8192 5
CUDALucas -threadbench 1024 8192 5
CUDALucas -r 0
[/INDENT]I'll run -r 1 overnight. It takes quite a while, doesn't it?[/QUOTE]

-r 1 completed. Yay!

Woohoo! Glad it worked! I don't think my EVGA 570 ever made it through -r 1.

[QUOTE=wombatman;409277]Woohoo! Glad it worked! I don't think my EVGA 570 ever made it through -r 1.[/QUOTE]

Have you tried reducing the RAM clock?

Hmmm, perhaps not. I'll have to try that out at some point. I was generally content to only have one or two crashes and restarts per LL test, but it's worthwhile to try and do better than that.

I can't say if it is universal, but many consumer Nvidia cards (as opposed to Teslas) have to have the memory slowed by a couple to a few hundred MHz to get through the CuLu self tests. It certainly has been true of all except (possibly) one of the cards I have tested with CuLu.

More "Depth First" LLTF'ing please...
 

Just so everyone knows, we're about a day away from starting to have to release candidates for P-1'ing at 74 bits again instead of the more appropriate 75.

Not the end of the world, of course. But sub-optimal.

I would ask that those who are doing LLTF'ing, please "pledge" to at least 74, or choose the "Let GPU72 Decide" option (which will take candidates to 75 currently). There are many "big guns" who are currently doing "breadth first" (read: TF'ing from 71 to 72); yes, that work has to be done in the future, but we have hungry P-1'ers and LL'ers to feed right now.

Also, those who are currently doing DCTF'ing might consider doing some LLTF'ing for a little while. We're currently about a year or so ahead of even the DC'ing Cat 4 wave so we have lots of time to divert some resources for a while.

But, as always, everyone's free to do whatever they enjoy most; it's your hardware / time / money after all. :smile:

[QUOTE=chalsall;409354]Just so everyone knows, we're about a day away from starting to have to release candidates for P-1'ing at 74 bits again instead of the more appropriate 75.

Not the end of the world, of course. But sub-optimal.

I would ask that those who are doing LLTF'ing, please "pledge" to at least 74, or choose the "Let GPU72 Decide" option (which will take candidates to 75 currently). There are many "big guns" who are currently doing "breadth first" (read: TF'ing from 71 to 72); yes, that work has to be done in the future, but we have hungry P-1'ers and LL'ers to feed right now.[/QUOTE]

I just grabbed 1300 GHzd of LG72D LLTF and it gave me back 34 72->74 assignments in the 72.9M range. Was that intended?

[QUOTE=Mark Rose;409355]I just grabbed 1300 GHzd of LG72D LLTF and it gave me back 34 72->74 assignments in the 72.9M range. Was that intended?[/QUOTE]

Yes. Sorry, I should have been clearer...

LG72D will balance between going to 75 or 74, depending on our current "real-time" situation. Right now we're really short of candidates without P-1 done at even 74, so the "AI" decided we needed some "to 74" work done.

Thanks for pitching in! :smile: