GIMPS is starting to release exponents in the 80-81M range TF'd to 2^71-72. For Research is grabbing the loin's share of those assignments the rest seem to be for P-1 testing.

Was just surprised that that range is starting to be operated on.

[url]http://www.mersenne.org/M80007341[/url]

[QUOTE=flagrantflowers;385431]GIMPS is starting to release exponents in the 80-81M range TF'd to 2^71-72. For Research is grabbing the loin's share of those assignments the rest seem to be for P-1 testing.
[/QUOTE]

Was it completely random that ~2000 assignments were taken up and dropped over the course of four hours in the 80-81M range?

[QUOTE=flagrantflowers;385443]Was it completely random that ~2000 assignments were taken up and dropped over the course of four hours in the 80-81M range?[/QUOTE]

Not quite sure if it was completely random or not.

According to my logs, Primenet was having some serious issues in communcations today. Probably at least correlative, if not causal.

Definitely not intentional on my part, but "when the cat's away, the spiders may play....

This coming weekend the company is moving offices. I may no longer be able to use two of the borged machines for TF. On the weekend my progress will be basically stopped (down to a single GT 430; my GTX 760 is out of commission at the moment). So that means about 70 fewer exponents TF 73->74 over those two days. More released at 73 bits I guess.

chris,
can you enable 38M and 39M range for DCTF
or will misfit start fetching there automatically
thanks
[url]www.gpu72.com/reports/available[/url]

[QUOTE=srow7;387691]chris,
can you enable 38M and 39M range for DCTF
or will misfit start fetching there automatically
thanks
[url]www.gpu72.com/reports/available[/url][/QUOTE]

But that would sabotage my nefarious plan to finish all the DCTF at GPU72! ;)

[QUOTE=srow7;387691]can you enable 38M and 39M range for DCTF
[/QUOTE]
Have you considered the following....
[QUOTE=chalsall;387624]So everyone knows, there is just short of a year's worth of DC Cat 4 candidates already appropriately TF'ed, while we're still releasing some LL Cat 4 at only 73 bits.

Or, in other words, for those doing TF'ing, please don't feel the need to do any additional DC TF'ing. If anything, anyone doing DC TF'ing should move back to LL TF'ing (if so inclined).[/QUOTE]

[QUOTE=srow7;387691]chris, can you enable 38M and 39M range for DCTF or will misfit start fetching there automatically[/QUOTE]

As Uncwilly pointed out, I mentioned on another thread that we currently have quite a bit of DC candidates appropriately TF'ed. And, also, HBendtz is independently taking the 38M range from 69 to 70.

But, since I want to facilitate and cooridinate the work that people want to do, I'll bring in a few thousand candidates in the 39M range only at 69 for processing.

[QUOTE=Prime95;387732]This wasn't as minor as I thought! Nearly 1000 cat4 DCs have been assigned in the last day. At that rate, we'll chew through 60000 cat4 DC assignments in the next 60 days! I don't see that as much of a problem.

I wonder why our retention rate is so low or why so many stress testers can't read the initial startup screen and not get Primenet assignments.[/QUOTE]


looks like we will be releasing no more Cat4 LLs below 74 anymore for quite some time, but the buffer for the Cat4DC (which seems to be about 7000 assignments) will run out not in close to a year, but rather in ... a week?

[QUOTE=manfred4;387797]looks like we will be releasing no more Cat4 LLs below 74 anymore for quite some time, but the buffer for the Cat4DC (which seems to be about 7000 assignments) will run out not in close to a year, but rather in ... a week?[/QUOTE]

Indeed.

I also didn't appreciate just how many of the LL TF Cat 4 assignments go to heuristically likely "churners" (read: first time users && stress testers who can't read).

Although this probably means that, like George said, ~60,000 DC TF Cat 4 assignments will be needed; GPU72 has only appropriately TF'ed ~20,000 candidates in that range (and averaging ~250 a day).

But the good news is, if LL Cat 4 behaviour is any indication, ~93% of them will simply timeout in 60 days and renter the pool; we'll process most of them "at the end of the day".

I would argue that it is better to save a LL than a DC, and that George's "tweak" is a very good idea.

That's mostley true, but I think, that some of the LLTF work should be headed back to DCTF again.
A single saved LL test is worth more than a DC test, but to save a DC test much less work is needed. So the same amount of work will be saved, no matter if you do LLTF to 74 or DCTF to 71 all the time.

I for myself will go back to DCTF for my PCs again for some time, I liked that much more anyway, because you can see the results much quicker and feel like having done more (although I know its even either way)

Hmm, so we are now probably well ahead in cat4 range to 74... Does it make sense to bring cat2/3 to 75?

[QUOTE=manfred4;387800]That's mostley true, but I think, that some of the LLTF work should be headed back to DCTF again.

...

I for myself will go back to DCTF for my PCs again for some time, I liked that much more anyway, because you can see the results much quicker and feel like having done more (although I know its even either way)[/QUOTE]

You beat me to it! :smile:

I had intended to follow up and mention that with the heavy LL requests in the Cat 4 zones transfered from LL to DC, it wouldn't be a bad idea for some of the LL TF firepower to move down to DC TF until the demand / recycling algorithm stablies the zone (read: in about 60 days).

Balance the load's and work between the LL TF and DC TF though (read: not every LL TF'er should jump over to DC'ing; we're still going to need additional LL TF'ing).

(And, as always, anyone is free to do any kind of work they enjoy doing.)

[QUOTE=NickOfTime;387801]Hmm, so we are now probably well ahead in cat4 range to 74... Does it make sense to bring cat2/3 to 75?[/QUOTE]

Interesting question... :smile:

I would like to suggest we wait at least 60 days until the Cat 4 demand type shift completes. Then we can better understand exactly where we stand with regards to demand and available FirePower[SUP](TM)[/SUP], and make the decision then.

But, yes, we should be going to 75 for most of the LL TF zone already; we may perhaps soon be able to! :yay:

for my graphics card going to 75 Bit should be done from 70M up according to the charts at mersenne.ca. But I think in 60 days when the recycling is back to normal the need in the LL range will be raised again, because there won't be churners left to take away the assignments.

But one more thing for this now: You said there is one person doing 38M 69-70 by himself? Looks like the assignments there will be given out for DC in a week so you should grab at least the assignments there at 70bits and maybe soon the other ones too. Just nothing slips through at 69.

[QUOTE=manfred4;387805]But I think in 60 days when the recycling is back to normal the need in the LL range will be raised again, because there won't be churners left to take away the assignments.[/QUOTE]

Respectfully, nope. It's a function of production and demand.

Over the last 30 days, the average amount of LL's actually completed per day was ~250.

Over the last 30 days, the everage amount of LLTF'ing to 74 or above was ~445. As Primenet is currently raising the LL Cat 4 range by 120 a day, up to 370 candidates need to be approriately TF'ed (although I think this redrection of Cat 4 is going result in the 120 a day being much less impactful.

[QUOTE=manfred4;387805]You said there is one person doing 38M 69-70 by himself? Looks like the assignments there will be given out for DC in a week so you should grab at least the assignments there at 70bits and maybe soon the other ones too. Just nothing slips through at 69.[/QUOTE]

I've been thinking of how to handle this...

I try to not "step on toes", so will try to get in contact with the worker to figure out how to handle this (I think he's on this forum).

If this doesn't happen before any in DC Cat 4 are about to be issued at below 70 bits, I'll reactivate one of my spiders with a "rip courd" to monitor the DC range, and release at 70 bits below 38M if anything is about to be released at only 69 bits.

It's going to take me approximately 20 days to DCTF the 2800 assignments I've reserved in 36M from 70->71. Would it be best if I unreserved a bunch to leave me with only a week's worth?

[QUOTE=Mark Rose;387811]It's going to take me approximately 20 days to DCTF the 2800 assignments I've reserved in 36M from 70->71. Would it be best if I unreserved a bunch to leave me with only a week's worth?[/QUOTE]

If you wanted to squeeze every last possible cycle out of the overall system, then yes, it would be best (as higher candidates are more expensive to LL than TF).

But, the increased throughput is probably so tiny as I think it's not worth your time and effort. And, we _are_ going to be releasing DC Cat 4 at 69 within two weeks or so.

One thought though -- consider going to 70 for a while when available....

[QUOTE=chalsall;387812]If you wanted to squeeze every last possible cycle out of the overall system, then yes, it would be best (as higher candidates are more expensive to LL than TF).

But, the increased throughput is probably so tiny as I think it's not worth your time and effort. And, we _are_ going to be releasing DC Cat 4 at 69 within two weeks or so.

One thought though -- consider going to 70 for a while when available....[/QUOTE]

There's not much -> 70 work at GPU72 right now, only 18 days worth at my current throughput. Can we pull more work in?

[QUOTE=Mark Rose;387813]There's not much -> 70 work at GPU72 right now, only 18 days worth at my current throughput. Can we pull more work in?[/QUOTE]

Yes, but it's only valuable if everything below is also assigned to someone to hold for TF'ing. I'll do my be to get in touch with HBendtz to explain what's going on, and that we need to reserve all of 38M and 39M to allow GPU72 to temporarily TF to 70.

But still, this isn't going to completely mitigate the situation; some DC Cat 4 *IS* going to be released at 69 for at least the next 60 days -- we simply don't have the firepower to prevent this.

Okay.

I've released everything at 70 back to GPU72, except for my assignments in progress. I've configured all my machines to fetch ->70 going forward, with a 12 hour queue.

[QUOTE=chalsall;387799]Although this probably means that, like George said, ~60,000 DC TF Cat 4 assignments will be needed; GPU72 has only appropriately TF'ed ~20,000 candidates in that range (and averaging ~250 a day).[/QUOTE]

So we need about 40,000 DCTF in the next 60 days. If we only DCTF ->70, that's a total of 2 THz-d/day. I can do half that alone. We need about 1 THz-d/day for the next two months to avoid releasing below 70 if anyone is interested in helping.

Chris, can "what makes sense" and "let gpu72 decide" issue assignments to only 70?

[QUOTE=chalsall;387815]Yes, but it's only valuable if everything below is also assigned to someone to hold for TF'ing. I'll do my be to get in touch with HBendtz to explain what's going on, and that we need to reserve all of 38M and 39M to allow GPU72 to temporarily TF to 70.

But still, this isn't going to completely mitigate the situation; some DC Cat 4 *IS* going to be released at 69 for at least the next 60 days -- we simply don't have the firepower to prevent this.[/QUOTE]

I will happily pick up some DCTF if that is desirable. The GTX 580 is currently doing just under 400 GHzD/D in the LL range.

[QUOTE=chalsall;387799]
Although this probably means that, like George said, ~60,000 DC TF Cat 4 assignments will be needed; [/QUOTE]

There are other options:

1) I could change assignments to new computers to a blend of cat4 DC and cat4 LL. Although I'd rather not.
2) I could investigate (next week) changing the expiration rules to 30 or 45 days after no contact from a computer (or user?) that has never returned a result. It would remain at 60 days for a computer (or user?) that has returned a result.

[QUOTE=kladner;387850]I will happily pick up some DCTF if that is desirable. The GTX 580 is currently doing just under 400 GHzD/D in the LL range.[/QUOTE]

That would help a lot! We have no chance of factoring to 71 bits (we need another 3 THz-d/d), but if you pledge to 70, we'll only need another 600 GHz-d/d to not release at 69.

I get ~425 in the current LL range with my GTX 580's with these settings:

GPUSievePrimes=70000
GPUSieveSize=128
GPUSieveProcessSize=16

It's only a 5% boost, but worth trying if you haven't done any tuning.

I've been running with similar numbers, though my impression was that GPUSieveProcessSize=8 might give a bit more. It's been a while since I really messed with testing that. ATM mfaktc 0.2 says it's doing 478 GHz-D/D. This is with it clocked at 861 MHz, where it seems quite happy, now that there's cooler air down by the floor.

The 400 GHz-D/D came from an estimate of where the 30 day average line is in my GPU72 Overall Statistics. :smile:

EDIT: Another hour of LLTF, and it will start on 39M DC to 70.

EDIT2: 9:15 to do one low 39M from 69-70.

Grrr... take your clothes off, I wanna [redacted] you... put your clothes back, I don't wanna [redacted] you...
Few days ago I moved my last card doing DCTF into doing LLTF because because we didn't have enough LLTF power...
(still finishing the DCTF queue, luckily, but now the queue ia already filled with LLTF assignments)
Now I have to go back home and move it back to DCTF?? :shock:
Ok, if desired, I will (happily, due to the lower bitlevel!) move all my current available power to DCTF for a while. I will do that tonight, you can count on ~1800GHzD/D of DC.

[QUOTE=Prime95;387851]2) I could investigate (next week) changing the expiration rules to 30 or 45 days after no contact from a computer (or user?) that has never returned a result. It would remain at 60 days for a computer (or user?) that has returned a result.[/QUOTE]

I think this would be a good (temporary) option. Setting it to 30 days would reduce the demand from ~60,000 to ~30,000 candidates, and based on the firepower being brought to bear we probably won't need to release TOO many at only 70.

BTW George, with your new Cat 4 assignment strategy, I think it's safe for you to set the LL Cat 4 offset to 100,000 now; no more need for the 120 step per day.

I've moved my GTX580 to DC-TF too. I think George doesn't need to resort to either option. We should be fine with LaurV and me coming to join the effort.

@George: You could look at that expiry SQL that says 60 days *and* overdue by 10 days as madpoo indicated in this post: [url]http://www.mersenneforum.org/showpost.php?p=387883&postcount=1520[/url]

[QUOTE=chalsall;387877]
BTW George, with your new Cat 4 assignment strategy, I think it's safe for you to set the LL Cat 4 offset to 100,000 now; no more need for the 120 step per day.[/QUOTE]

I'd leave it on. There is no harm. There will be plenty of Cat3 to go around.

[QUOTE=LaurV;387857]Grrr... take your clothes off, I wanna [redacted] you... put your clothes back, I don't wanna [redacted] you...
Few days ago I moved my last card doing DCTF into doing LLTF because because we didn't have enough LLTF power...
(still finishing the DCTF queue, luckily, but now the queue ia already filled with LLTF assignments)
Now I have to go back home and move it back to DCTF?? :shock:
[/quote]

:sorry:

[quote]
Ok, if desired, I will (happily, due to the lower bitlevel!) move all my current available power to DCTF for a while. I will do that tonight, you can count on ~1800GHzD/D of DC.[/QUOTE]

Please do! It looks like we're getting close enough in pledged power to factor everything needed to 71 bits.

Hello Chris,

There is something funky with the gpu72 website with regards to getting DCTF assignments. I'm setting options to "Let GPU72 decide" and it states it will factor to 70 bits (which is where I understand the immediate need is). The preview states that it will factor from 69 to 70 bits. However when I get the assignments, I get ones to factor from 70 to 71 bits. I'll finish the ones I've obtained, but I keep an eye for a fix to this before I request more.

-- Rich

[QUOTE=monst;387931]Hello Chris,

There is something funky with the gpu72 website with regards to getting DCTF assignments. I'm setting options to "Let GPU72 decide" and it states it will factor to 70 bits (which is where I understand the immediate need is). The preview states that it will factor from 69 to 70 bits. However when I get the assignments, I get ones to factor from 70 to 71 bits. I'll finish the ones I've obtained, but I keep an eye for a fix to this before I request more.

-- Rich[/QUOTE]

It looks like we're going to have enough power working on DCTF that we can factor all the way to 71 bits. If you really want to factor to just 70 bits, set it to 70 bits and pick "lowest exponent". Everything for Cat 1-3 DCTF is already fully factored, so what's left is all the same to GPU72.

[QUOTE=Mark Rose;387942]It looks like we're going to have enough power working on DCTF that we can factor all the way to 71 bits. If you really want to factor to just 70 bits, set it to 70 bits and pick "lowest exponent". Everything for Cat 1-3 DCTF is already fully factored, so what's left is all the same to GPU72.[/QUOTE]

Does this mean that I should upgrade my current assignments to 71? It is easy enough to do: dump unstarted assignments and grab a new bunch.

I usually go with mostly 2-3 days of queue, so if you are in the same case, better let it finish the 70s.
(actually it doesn't really matter, all of them would need to be done eventually...:smile:)

[QUOTE=LaurV;387945]I usually go with mostly 2-3 days of queue, so if you are in the same case, better let it finish the 70s.[/QUOTE]

I'm good with that. I've got a little less than 2 days queued.

It's not necessarily bad to have a lead of exponents sitting at 70 though, just in case some need to be released (at 70 vs lower).

I did a rough estimate of how much DCTF throughput we should be getting in a day or two based on who's already doing DCTF plus those who've said they will.

Result? Approximately 7 THz-d/d!

We have approximately 147 THz-days of work available or assigned at GPU72, with an approximately average of 7.5 GHz-d per assignment. That works out to about 933 exponents to 71 per day. At 7 THz-d/d we'll chew through all the available and assigned work in approximately 3 weeks. We'll be short about 50 exponents a day for this work (1000 total).

We'll still need another 35,000 exponents by mid January when the effort ends. Should we pull in two more ranges now? Since 38M, 40M, and 41M are mostly at 69 bits, they'll need about 8.9 GHz-d each. That will drop our throughput to about 800 exponents a day. We'll be short about 150 exponents a day for this work (short 5700).

We're short almost exactly the buffer of 6700 assignments, so if anyone else would like to jump on board to give us some margin, please do!

:chalsall:

An even balance of bit-first vs exponent-first is probably a good thing. Too many going from 69 to 71 means more exponents being released at 69. Too many going from 69 to 70 means too many released at 70. The latter is a bit unfortunate if we have the firepower to hit 71.

There's no sense getting tens of thousands of exponents to 70 without going to 71. That's a big buffer we don't really need.

[QUOTE=Mark Rose;387951]
We're short almost exactly the buffer of 6700 assignments, so if anyone else would like to jump on board to give us some margin, please do![/QUOTE]I don't have much but I'll throw a GTX650 (~70GHz-Days/D) at some to 70.

[QUOTE=legendarymudkip;387961]I don't have much but I'll throw a GTX650 (~70GHz-Days/D) at some to 70.[/QUOTE]

Thanks for all the cycles everyone; looks like we'll be able to squeak by without releasing anything at less than 71.

I've also brought a EC2 instance online for a while; 2 * 2050 Titans at ~520 GHz-d/d (aggregate) for ~13.5 cents an hour! :smile:

[QUOTE=legendarymudkip;387961]I don't have much but I'll throw a GTX650 (~70GHz-Days/D) at some to 70.[/QUOTE]

Every bit helps! I'm currently running 4 cards weaker than that one, 2 x GT 520 (29 GHz-d/d), and 2 x GT 430 (50 GHz-d/d). It adds up!

LOL, I didn't know people were running those kind of cards, I'm almost wasting as much GHzD/D. Not that it matters, every bit counts.
I'm running my HD7950 at 265GHzD/D during the day and 400GHzD/D at night, because I can't stand the screenlag/video stuttering.

Y'all making me feel guilty so I've picked up 1THz-day of work.
I also noticed that the assignment-preview showed me 38M,69-70 but the assignments I got were 36M,70-71 (as [i]monst[/i] [url=http://www.mersenneforum.org/showpost.php?p=387931&postcount=3136]said a few posts back[/url]).

LOL, I guess I can do some too... I have almost 700 GHz/days of power I have at the moment, you want the default, to 71 bits on what makes sense?

[QUOTE=monst;387931]There is something funky with the gpu72 website with regards to getting DCTF assignments...[/QUOTE]

Thanks for pointing that out -- one constant in the javascript that does the "LG72D" thing. Should be fixed now (you might need to clear your cache to pick up the new version).

And, so everyone knows, with the firepower which has suddenly come forward what makes the most sense (in my opinion, and currently codified on GPU72) is simply lowest exponent, regardless of it's current TF level. This will mean initially 70 to 71.

But, of course, anyone who wants to only go to 70, that's cool too.

We'll already have about 20,000 candidates at 70 which we can return if needed, but it's looking very unlikely we won't immediately start to pull ahead of the "churners".

[QUOTE=VictordeHolland;387981]LOL, I didn't know people were running those kind of cards, I'm almost wasting as much GHzD/D. Not that it matters, every bit counts.
I'm running my HD7950 at 265GHzD/D during the day and 400GHzD/D at night, because I can't stand the screenlag/video stuttering.[/QUOTE]

I also stop mfaktc on my primary card (almost back from RMA, actually) when using my home desktop for the same reason. At work I use the on-board Intel graphics so the two GTX 580's in that box can stay flat out 24x7.

[QUOTE=chalsall;387977]Thanks for all the cycles everyone; looks like we'll be able to squeak by without releasing anything at less than 71.

I've also brought a EC2 instance online for a while; 2 * 2050 Titans at ~520 GHz-d/d (aggregate) for ~13.5 cents an hour! :smile:[/QUOTE]

How do you do this???

[QUOTE=Primeinator;387992]How do you do this???[/QUOTE]

Spot instance pricing on Amazon EC2. One availability zone in us-east-1 has cg1.4xlarge for about $0.135/hour, or about $3.25/day.

[QUOTE=Mark Rose;387995]Spot instance pricing on Amazon EC2. One availability zone in us-east-1 has cg1.4xlarge for about $0.135/hour, or about $3.25/day.[/QUOTE]

Yup. Just sign up to EC2, spin up an instance which can support NVideo CUDA (there are a bunch free ones; choose your faviorate distro) on a cg1.4xlarge instance type, install CUDA (and apply security updates, etc).

Next, create a new 1GB volume, and amoung other customizations change /etc/fstab to include a line which reads "/dev/sdf1 /home ext4 noauto,noatime 1 0", and then attach this to your instance (by default it will be "/dev/sdf"). Use "fdisk /dev/sdf" to create a single linux partition, and then "mkfs.ext4 /dev/sdf1".

Now, copy (with the archive flag) the existing "/home/ec2-user" directory to "/tmp/" Lastly "mount /home" should bring the newly created partition online. Move (or copy) the "/tmp/ec2-user" directory to "/home".

"shutdown -r now", log back into the instance, and execute "sudo mount /home". Make sure the new home partition mounts correctly.

If it does, then install all your desired mfaktc software in two seperate directories (including any submission / fetching spiders desired, and create the related cronjobs) and give it a go. Launch two mfaktc instances (the second with "-d 1"). You may also want to install an instance of mprime as well.

If everything's working the way you want then "Create Image" from the "Instances" tab. This will create a custom AMI which you'll boot from in the future (thus avoiding the need to install CUDA and the software each time.

Note that, importantly, the AMI file system is ***NOT*** preserved between instance creations; this is the reason for the extra volume created for "/home" -- mounted volume's are preserved.

Oh, and know that spot instances can go away at any time without notice. But, they're dirt cheap; I usually bid at $0.15 per hour, and often have one (or more) instances up for many days paying on average $0.135 an hour.

It's actually relatively easy to do.

:goodposting:

That is more competitive with hardware @home than I thought!

The Amazon EC2 "g2.2xlarge" consists of:
Intel Xeon E5-2670 (Sandy Bridge)
15GB RAM
NVIDIA GPU with 1,536 CUDA cores

At $0.135/h = ~ $100 a month.

Buying+running hardware @home in Netherlands
Quadcore (Core i5) €160
16GB DDR3-1333 RAM €135
GTX770 (1536 CUDA cores) €230
Motherboard €65
PSU €50
--------------------------------------------
Initial investment: €640 or $800 (assuming you have a spare HDD and case)
running costs (very crude way)
CPU TDP 84W + GPU TDP 230W = 314W
314W * €0.20/KWh = ~€0.06/h or $0.075/h

Breakeven is after $800/($0.135/h-$0.075/h) = 13,000+ hours, that is 1.5 years!

Some remarks:
The throughput of the home system is higher. (Xeons have lower clockspeed than the i5s and Tesla cards also have lower clockspeeds than their consumer counterparts. And the home system has some residual value.
On the other hand a g2.2xlarge Spot Instance in EU (Ireland) is only $0.0801/h at the moment, which is very close to the $0.075/h running @home!


Question: Can EU residents start Spot Instances in all regions? In that case you could just pick the cheapest region.

[QUOTE=VictordeHolland;388019]Question: Can EU residents start Spot Instances in all regions? In that case you could just pick the cheapest region.[/QUOTE]

Yes.

For me, actually cheaper than the electricity to run a G580.

[QUOTE=VictordeHolland;388019]
Breakeven is after $800/($0.135/h-$0.075/h) = 13,000+ hours, that is 1.5 years!
[/QUOTE]

Yep! My electricity cost makes the break even point almost as much with the higher US dollar at the moment. But that's if buying new hardware. If you get "free" electricity, and buy old GTX580's at $100/each and pop them into machines, the recovery time is one month. I need to find more such machines lol

[quote]
On the other hand a g2.2xlarge Spot Instance in EU (Ireland) is only $0.0801/h at the moment, which is very close to the $0.075/h running @home!
[/quote]The g2.2xlarge is not such a good value. You only get a single GPU at 240 GHz-d/d instead of two GPUs at 280 GHz-d/d. For mfaktc, the g2.2xlarge's fair value in comparison is $0.05/hr.

[quote]
Question: Can EU residents start Spot Instances in all regions? In that case you could just pick the cheapest region.[/quote]The cg1.4xlarge is only available in us-east-1 and eu-west-1. Of those, only in one available zone in us-east-1 has the spot price been hovering around $0.135.

Just sayin': two factors since I got into the DC catch-up project. :smile:

Is the procedure to send out DC work when needed to people selecting "Let GPU72 Decide" in effect?

[QUOTE=Chuck;388047]Is the procedure to send out DC work when needed to people selecting "Let GPU72 Decide" in effect?[/QUOTE]

It's probably not needed at the moment, given all the people manually working on DCTF. We still need work happening on TF.

[QUOTE=Chuck;388047]Is the procedure to send out DC work when needed to people selecting "Let GPU72 Decide" in effect?[/QUOTE]

It isn't, but as Mark said, it's not needed.

[QUOTE=Mark Rose;388023]The cg1.4xlarge is only available in us-east-1 and eu-west-1. Of those, only in one available zone in us-east-1 has the spot price been hovering around $0.135.[/QUOTE]

Mark: on a releative tangent... You have a lot more experience with EC2 than I; do you (or others) have any theory as to why there is such a huge discrepency in pricing between us-east-1d (currently ~$0.1342) and us-east-1c (~$2.10)? That's over 15 times!

I know there have been some studies on this; do you (or anyone) have any good references? (I'm just intriqued is all.)

[QUOTE=chalsall;388009]Yup. Just sign up to EC2, spin up an instance which can support NVideo CUDA (there are a bunch free ones; choose your faviorate distro) on a cg1.4xlarge instance type, install CUDA (and apply security updates, etc).

Next, create a new 1GB volume, and amoung other customizations change /etc/fstab to include a line which reads "/dev/sdf1 /home ext4 noauto,noatime 1 0", and then attach this to your instance (by default it will be "/dev/sdf"). Use "fdisk /dev/sdf" to create a single linux partition, and then "mkfs.ext4 /dev/sdf1".

Now, copy (with the archive flag) the existing "/home/ec2-user" directory to "/tmp/" Lastly "mount /home" should bring the newly created partition online. Move (or copy) the "/tmp/ec2-user" directory to "/home".

"shutdown -r now", log back into the instance, and execute "sudo mount /home". Make sure the new home partition mounts correctly.

If it does, then install all your desired mfaktc software in two seperate directories (including any submission / fetching spiders desired, and create the related cronjobs) and give it a go. Launch two mfaktc instances (the second with "-d 1"). You may also want to install an instance of mprime as well.

If everything's working the way you want then "Create Image" from the "Instances" tab. This will create a custom AMI which you'll boot from in the future (thus avoiding the need to install CUDA and the software each time.

Note that, importantly, the AMI file system is ***NOT*** preserved between instance creations; this is the reason for the extra volume created for "/home" -- mounted volume's are preserved.

Oh, and know that spot instances can go away at any time without notice. But, they're dirt cheap; I usually bid at $0.15 per hour, and often have one (or more) instances up for many days paying on average $0.135 an hour.

It's actually relatively easy to do.[/QUOTE]

Thanks! I am going to look into this.

[QUOTE=chalsall;388059]Mark: on a releative tangent... You have a lot more experience with EC2 than I; do you (or others) have any theory as to why there is such a huge discrepency in pricing between us-east-1d (currently ~$0.1342) and us-east-1c (~$2.10)? That's over 15 times!

I know there have been some studies on this; do you (or anyone) have any good references? (I'm just intriqued is all.)[/QUOTE]

Amazon has never said specifically the algorithm behind the spot pricing. People who have looked more into it than I have come up with some speculations that make sense though.

Spot pricing is basically for unused reserved instances. People buy reserved instances for two reasons: one, if running most of the time, they save money; and two, they get first dibs on capacity, ahead of on-demand and spot. However, they don't have to be fully using their reserved instances. These unused instances can be sold on the spot market.

Amazon sets a floor on spot capacity, my guess is to cover the electrical cost of having the machine run. If there are sufficient spot requests to run all the spot capacity for an instance type, the highest bids win. Some people may set a higher than on-demand spot price to take advantage of low pricing most of the time, while also have a better chance to keep their instance running if capacity becomes constrained.

If all the reserved instances are in use, and the least-winning spot bid is at the on-demand price, on-demand capacity can be used to fulfill spot requests. The on-demand price for a cg1.4xlarge is currently $2.10 in us-east-1. In the one zone that has no spot capacity (probably no instances are reserved, because most people are using the g2.2xlarge instances), so the spot price is the on-demand price.

In the other two zones, one has a slightly higher bid price, likely because someone is willing to pay more to keep their spot instances in the same zone (with other spot instances or otherwise).

In eu-west-1 my guess is no one has unused reserved instances, so there is no spot capacity.

So what exactly happened to that one-year buffer we had? I never really understood how we suddenly had a larger demand for DC's...

[QUOTE=TheMawn;388089]So what exactly happened to that one-year buffer we had? I never really understood how we suddenly had a larger demand for DC's...[/QUOTE]

Basically, this thread: [url]http://www.mersenneforum.org/showthread.php?t=19823[/url]

George changed it so that new users get DCLL instead of LL by default for various reasons. Instead of 50ish DC being requested per day, it jumped to 1000 per day. 90% of those will be recycled, so the rate will drop to 150ish per day, after 60 days (mid January).

[QUOTE=chalsall;388009]Yup. Just sign up to EC2, spin up an instance which can support NVideo CUDA (there are a bunch free ones; choose your faviorate distro) on a cg1.4xlarge instance type, install CUDA (and apply security updates, etc).
[/QUOTE]
If I could understand this (all), I would likely spend about $30 a month toward the 100M digit range.

[QUOTE=Uncwilly;388097]If I could understand this (all), I would likely spend about $30 a month toward the 100M digit range.[/QUOTE]

It really is quite easy. I'd suggest you sign up for EC2, and experiment with their "Free Teir" offer; gives you two "t1.micro" instances (one Linux and one Winblows) to experiement with for a year for free. Not powerful enough to do much LL'ing (and no GPU'ing), but a good way to get comfortable with the UI, creating and mounting volumes, etc.

If you're interested, I could probably figure out how to export my AMI (already configured with CUDA) for you to use.

[QUOTE=chalsall;388118]It really is quite easy. I'd suggest you sign up for EC2, and experiment with their "Free Teir" offer; gives you two "t1.micro" instances (one Linux and one Winblows) to experiement with for a year for free. Not powerful enough to do much LL'ing (and no GPU'ing), but a good way to get comfortable with the UI, creating and mounting volumes, etc.

If you're interested, I could probably figure out how to export my AMI (already configured with CUDA) for you to use.[/QUOTE]

That reminds me, I should really get that AMI I talked about building, built.

I would include Prime95/mprime, but I'm not sure about the licensing issues.

I've been avoiding using EC2 for the project because I'd rather spent the month's worth of expense on a used GTX 580's. But I haven't bought any more either lol.

[QUOTE=chalsall;388118]It really is quite easy. I'd suggest you sign up for EC2, and experiment with their "Free Teir" offer; gives you two "t1.micro" instances (one Linux and one Winblows) to experiement with for a year for free.[/QUOTE]Maybe I'll do that over the holidays.

I know it's not much, but I picked up a batch of 100 DC-TF assignments a few days ago. They should be complete tomorrow.

[QUOTE=ixfd64;388180]I know it's not much, but I picked up a batch of 100 DC-TF assignments a few days ago. They should be complete tomorrow.[/QUOTE]

Everybody's work is incremental on some scale, (except for Curtis Cooper, of course.) :smile:

[QUOTE=kladner;388187]Everybody's work is incremental on some scale, (except for Curtis Cooper, of course.) :smile:[/QUOTE]

He doesn't factor. He's too good for us lowly pond scum.

My only remaining TF system has been flaky lately and it finally took a dump. I've been working to get it back up, but the CPU fan went bad now so I have to wait until it comes in the mail. With Thanksgiving, will probably be a week or so :glare:

[QUOTE=flashjh;388202]With Thanksgiving, will probably be a week or so :glare:[/QUOTE]

Sorry to hear that; thanks for the head's up.

[QUOTE=chalsall;388053]It isn't, but as Mark said, it's not needed.[/QUOTE]

Eeeeep! Spoke too soon... Having had some time to see the empirical rates, we're still behind!

Over the last week we've done on average ~926 per day to 71, while the DC Cat4 "churners" have requested on average almost exactly 1,400 assignments per day over the last nine days!

Thus, I'm going to start giving out DC TF to 71 work for those who ask for LL TF work with the "Let GPU72 Decide" option set.

Those who don't want to do any DC TF'ing choose any other option (including "What makes sense"). Every option other than LG72D will only ever give LL TF work.

Alternatively, those willing to help out with the DC TF "surge", please either set the option to be LG72D, or explicitely ask for DC TF work (preferably to 71, but to 70 would help to).

Edit: Oh, and so everyone knows, because of the "churn" being moved from LL to DC, we could go for several months without doing any additional LL TF'ing and still not hand out anything at less than 74. This means we should be able to start going to 75 in the appropriate LL ranges when this DC surge is finished.

Grabbing another 10kGHd

[QUOTE=chalsall;388330]Eeeeep! Spoke too soon... Having had some time to see the empirical rates, we're still behind!

Over the last week we've done on average ~926 per day to 71, while the DC Cat4 "churners" have requested on average almost exactly 1,400 assignments per day over the last nine days!
[/quote]

Some people were a bit late in jumping on board, finishing up LLTF work I imagine. I see now on mersenne.info that 1831 factors were moved 69 to 71 yesterday. The day before it was 1100, and the day before that 1000. If we can keep up yesterday's pace we'll be okay.

[QUOTE=Mark Rose;388336]If we can keep up yesterday's pace we'll be okay.[/QUOTE]

If. Such an important word... :wink: :smile:

Until we pull out of this (downward) glide in DC, it might be fun to "watch the numbers" and possibly react (or not). :smile:

[QUOTE=chalsall;388330]we could go for several months without doing any additional LL TF'ing and still not hand out anything at less than 74.[/QUOTE]

Alright. This is an important statement. I will finish my week or so worth of LLTF and switch to DCTF.

Just a logistical question about GPU72: If I completed some work and submitted the results to Primenet but unreserved the work from GPU72.com before it had a chance to "see" that I finished the work, what happens?

I don't know, but GPU72 checks at least every hour (or half hour?) so if you wait an hour before unreserving you're good.

I thought "Let GPU72 Decide" was going to start handing out DC work, but the last batch I just received (via MISFIT) was 73M 71 to 74 bits.

Giving her all she can take, Captain! :max:

EDIT: :wink:

[QUOTE=chalsall;388348]If. Such an important word... :wink: :smile:

Until we pull out of this (downward) glide in DC, it might be fun to "watch the numbers" and possibly react (or not). :smile:[/QUOTE]

Down to 1,068 yesterday :(

[QUOTE=TheMawn;388357]Just a logistical question about GPU72: If I completed some work and submitted the results to Primenet but unreserved the work from GPU72.com before it had a chance to "see" that I finished the work, what happens?[/QUOTE]

The Internet crashes... :wink:

Seriously though, GPU72 will observe (within an hour) that the work completed, but won't credit you. There will be up to a one hour window where GPU72 might assign the work to another worker, thus wasting cycles.

Thus, if you're going to unreserve work, it would be best to submit everything done, clear everything else out of your worktodo.txt file(s), and reserve new work. Then, wait two hours (just to be sure), and then unreserve the remaining work.

[QUOTE=Chuck;388361]I thought "Let GPU72 Decide" was going to start handing out DC work, but the last batch I just received (via MISFIT) was 73M 71 to 74 bits.[/QUOTE]

I didn't implement the code change yesterday; I have today (including the preview).

[QUOTE=chalsall;388390]I didn't implement the code change yesterday; I have today (including the preview).[/QUOTE]

Thanks.

I changed the recycling SQL to expire DCs after 40 days of no contact if the computer has never returned an LL result.

Honey I'm Home!!!!!!!!!!!!!!!!
 

Have GPU will Travel....or in this case DC.

Just got my first 100 x 37M DC to 71 bits.
Screen suggests 18 minutes per assignment. Is this good?
My rough Math tells me it will be about 500GhzDays / Day.

GTX 970 mfaktc 0.20

No fiddling with the ini file yet...I suspect the defaults are not best suited for my card?

Have to get the Grandkid to bed first.

[QUOTE=petrw1;388422]Have GPU will Travel....or in this case DC.

Just got my first 100 x 37M DC to 71 bits.
Screen suggests 18 minutes per assignment. Is this good?
My rough Math tells me it will be about 500GhzDays / Day.
[/quote]

That's great!

[quote]
GTX 970 mfaktc 0.20

No fiddling with the ini file yet...I suspect the defaults are not best suited for my card?
[/QUOTE]

The defaults will be okay, but there's usually room for improvement.

The three that matter are:

GPUSieveProcessSize
GPUSieveSize
GPUSievePrimes

I would adjust them in that order. As soon as you see a GHz-d/day number you've run it long enough. CRTL-C and adjust. There are only a few values for the first two, and I'd adjust GPUSievePrimes by jumps of 5000 or so.

It's telling me 502 - 507 per day right now.

Should complete about 80 per day of the 37M to 71 Bits category.

Makes me think the limit of getting 100 at a time could be bumped?

I know I should look myself but does anyone want to offer up whether [SPOILER]Misfit[/SPOILER] has been altered to run on Win-64?

I might be out or order here on the name and confused re related posts.

I mean the "Submission Spider"

Have you tried running the existing misfit?

[url]http://mersenneforum.org/misfit/downloads/MISFIT/MISFIT-2.9.2.zip[/url]

[QUOTE=Mark Rose;388430]Have you tried running the existing misfit?

[url]http://mersenneforum.org/misfit/downloads/MISFIT/MISFIT-2.9.2.zip[/url][/QUOTE]

Guilty .... I will give it a try this weekend.

Thx

Also, James is probably going to want benchmarks. The GTX 970 is listed as only doing 388 GHz-d/day, not over 500!

[QUOTE=petrw1;388422].....
Screen suggests 18 minutes per assignment. Is this good?
My rough Math tells me it will be about 500GhzDays / Day.

GTX 970 mfaktc 0.20

.......[/QUOTE]

Is that 18 minutes from 70 to 71?

[QUOTE=petrw1;388431]Guilty .... I will give it a try this weekend.

Thx[/QUOTE]

You will need up to a certain level of .NET. I don't remember what is required, but there's probably a sticky in that sub-forum to explain such things. MISFIT runs just fine for me in Win 7 Pro-64bit.

[QUOTE=kladner;388436]Is that 18 minutes from 70 to 71?[/QUOTE]

no factor for M37767361 from 2^70 to 2^71 [mfaktc 0.20 barrett76_mul32_gs]

18 Min 8 Secs.
Screen says 503 GhzDays/Day

Benchmark submitted

Which GTX 970 did you get?

I'm trying to come up with an excuse to get one. Maybe during the December sales.

First 7 GPU Results submitted....
 

...1 factor.

2 questions
 

1. Can I get it to output intermediate results less often...currently every second.

2. If the Userid AND Computer Name parameters are set in the .ini file and show up in the results file why is the computer name not on the results page instead of "Manual Testing"?

Thanks

3. I think this should be obvious but I assume whenever I submit the results I need to remove them from my results file so they aren't sent again?
I don't see any timestamps on the file that would tell Prime95 what has already been processed

[QUOTE=Mark Rose;388447]Which GTX 970 did you get?

I'm trying to come up with an excuse to get one. Maybe during the December sales.[/QUOTE]

Gigabyte NVIDIA GeForce GTX970.

In case it matters to the thruput I am running mfaktc 0.20 64 BIT Windows on an Intel i5-750 processor.