.

[QUOTE=dragonbud20;416374]Don't know enough to know what it means but I'm getting the same thing with www. is down for all and without is just me. I also tried isitdownrightnow.com and that reports both variations as being down for everyone.[/QUOTE]

[QUOTE=LaurV;416368]This is what I am getting from here, and the "only me?" says "no".

edit: interesting, if I only type gpu72.com (without www) the isup.me says it is only me. But with www it says not. DNS problem?[/QUOTE]

When I visit [url]http://gpu72.com[/url], I'm redirected via a 301 Moved Permanently to [url]http://www.gpu72.com[/url], which gives a 500 error. It appears that isitdownrightnow.com follows the redirect before reporting that it does or doesn't work, while isup.me does not: it sees the 301 and says "yep, it's good".

Edit: It's working at the moment. Visited the home page, and got assignments thru misfit. We'll see if it stays up.

[QUOTE=LaurV;416386]Checking the fresh satellite images, Barbados is still on the map, no meteorite strike, no tsunami*... then it must be something from the server....
[/QUOTE]

Did you see Chris? What's he doing?

[QUOTE=Bdot;416408]Did you see Chris? What's he doing?[/QUOTE]

Hmmm...

Sorry guys. I took the rest of the day off yesterday (AFK, cellphone off, etc), just before the GPU72 server "went south".

Looking at the logs I see some unusual activity...

Time for a security audit, and then a rebuild (just to be sure). I'll let everyone know when this is scheduled.

Is this updating?
 

[url]http://www.gpu72.com/reports/current_level/[/url]

DC 70 Column has not changed all day ..21,300

[QUOTE=petrw1;417198][url]http://www.gpu72.com/reports/current_level/[/url]

DC 70 Column has not changed all day ..21,300[/QUOTE]

Well it seems DC 71 is changing so I guess whomever has all the rest of the DC 70 assignments is just tardy.

[QUOTE=petrw1;417269]Well it seems DC 71 is changing so I guess whomever has all the rest of the DC 70 assignments is just tardy.[/QUOTE]

Yes.

Call out to Squirrels... Where are you? What are you doing? :smile:

Soon. Someone moved his supercomputer stack to a new building where the connecting of fiber has been delayed :| Construction - never on time.

Portions of the systems are running, but headless and without network connections. I didn't feel particularly inclined to head over there with a monitor and thumb drive.I figured no one would mind since I'm just hoarding well ahead DCTF, but if I'm holding anyone up let me know and I'll figure out what assignments I can purge and release.

[QUOTE=airsquirrels;417315]I figured no one would mind since I'm just hoarding well ahead DCTF, but if I'm holding anyone up let me know and I'll figure out what assignments I can purge and release.[/QUOTE]

Nope, you're not holding anyone up. And your assignments will not expire.

Look forward to seeing a serious dump sometime in the future.... :smile:

[QUOTE=airsquirrels;417315]Soon. Someone moved his supercomputer stack to a new building where the connecting of fiber has been delayed :| Construction - never on time.

Portions of the systems are running, but headless and without network connections. I didn't feel particularly inclined to head over there with a monitor and thumb drive.I figured no one would mind since I'm just hoarding well ahead DCTF, but if I'm holding anyone up let me know and I'll figure out what assignments I can purge and release.[/QUOTE]

I'm glad you're still around. :)

[QUOTE=chalsall;417316]...Look forward to seeing a serious dump sometime in the future....[/QUOTE]

You realize how that sounds out of context, right? :smile:

[QUOTE=Madpoo;417340]You realize how that sounds out of context, right? :smile:[/QUOTE]

Yes.

[QUOTE=Madpoo;417340]You realize how that sounds out of context, right? :smile:[/QUOTE]

True Story:
Some years back I was taking an Oracle DBA course,
2 co-workers and my boss were in the same course (he wanted to know enough to be dangerous).
The topic of the day was creating and deleting database objects: Tables, Indexes, Tablespaces, Change Logs
In Oracle Syntax (as it is for most SQL Bases databases) the high level commands are CREATE and DROP.
Our class exercise was to practice some of these commands.
A few minutes in my boss turned to me and in a completely straight face said:
"Do you know how to drop a log?"
A few seconds of silence was followed by laughter from my co-workers.
And this was quickly followed by my boss turning red and commenting something like.
"Oops,I could have worded that better"

[QUOTE=chalsall;417343]Yes.[/QUOTE]

But you're still going to share your dump with the rest of us right?

[QUOTE=0PolarBearsHere;417403]But you're still going to share your dump with the rest of us right?[/QUOTE]

Actually, other's dumps... In [URL="https://www.gpu72.com/reports/overall/graph/month/"]graphic detail[/URL].... :wink:

[QUOTE=chalsall;417408]Actually, other's dumps... In [URL="https://www.gpu72.com/reports/overall/graph/month/"]graphic detail[/URL].... :wink:[/QUOTE]

I had to hover over that link and see where it went. For a moment I thought you might have linked to some German website that specializes in a certain form of ... umm... whatever.

[QUOTE=Madpoo;417419]For a moment I thought you might have linked to some German website that specializes in a certain form of ... umm... whatever.[/QUOTE]

Just because German people are known for optimization doesn't mean necessarily they are bad.

Niccolò Machiavelli was Italian.

George W. Bush was American....

[QUOTE=chalsall;417425]Just because German people are known for optimization doesn't mean necessarily they are bad.

Niccolò Machiavelli was Italian.

[B]George W. Bush was American....[/B][/QUOTE]

Last I heard, W still is American, worse luck, as is Darth Cheney, and a host of other evil ones.

[QUOTE=kladner;417430]Last I heard, W still is American, worse luck, as is Darth Cheney, and a host of other evil ones.[/QUOTE]

Touché.

Finally had a factor in my TF. (and credited as such, so it wasn't an exponent that I added manually just to check the graphics card)
Happy that my card is still functional.
Back to crunching then.

Temporary focus on LLTF Cat 4...
 

Just a quick update...

Based on the /possibility/ that a new Mersenne Prime has been found (and, to be clear, I have no "insider" knowledge of this) GPU72 is currently directing some of its workers' firepower to the LL Category 4 which has already had a P-1 run done.

My thinking is that while new LL'ing workers will generally initially be assigned DC'ing work in the DC Cat 4 range until they've demonstrated they're serious and stable (where we're _years_ ahead), new workers who explicitly ask for them will get LL Cat 4.

We do currently have a reasonable buffer in all the LL categories, but I just wanted to ensure we can handle large amounts of assignment requests from new workers which are appropriately TF'ed and P-1'ed.

Iff (if and only if) this does turn out to be a real new Mersenne Prime we can adjust our targeting as needed and appropriate. If this doesn't turn out to be a new Mersenne Prime then we'll go back to "just in time" for Cat 1, 2 and 3.

Fine with me, you can feed all my workers with LLTF to 74, until I will have time/access to change the preferences in Misift. The DCTF is dying anyhow.

Now I hate you guys, if you kill all the blue, how can I make [URL="http://www.gpu72.com/reports/worker/2423ae6e8f696d5e7d1447de91ca35a6/"]my cityscape[/URL] any longer? It will all be green and ugly... :razz:

Most of my effort is now on What Makes Sense LL. DCTF has been granted a (brief) reprieve. I suspect Anonymous will keep running us towards the finish line though.

Laurv, don't worry after this we will tune and optimize the GPU TF programs until it is worth taking everything up one more bit level and the cycle will repeat!

50M 60M 70M
 

Thanks in a big part to GPU72 all 3 of these range now have less than 200,000 Exponents unfactored (i.e. requiring LL/DC).

This is very roughly about 10% better than the number of exponents factored via the standard PrimeNet factoring limits.

[QUOTE=petrw1;421944]Thanks in a big part to GPU72 all 3 of these range now have less than 200,000 Exponents unfactored (i.e. requiring LL/DC).[/QUOTE]

GPU72 is just a very small part of this.

The true credit needs to go to Oliver ("TheJudger") and Bertram ("Bdot") for writing the GPGPU code which completely changed the Trial Factoring landscape, and "Mr. P-1" who used to coordinate this effort via e-mails and PMs which inspired the whole idea of GPU72.

Also, Mike ("XYZZY") for creating, hosting and managing this forum, and of course, George ("Prime95") for creating the GIMPS project in the first place.

And that is just a small list of those involved. It's nice being associated with a group of really smart people with different talent sets! :tu:

[QUOTE=chalsall;421950] :tu:[/QUOTE]
+1 from me.
:tu:

Seeing a mini-surge!
 

[QUOTE=chalsall;421840]My thinking is that while new LL'ing workers will generally initially be assigned DC'ing work in the DC Cat 4 range until they've demonstrated they're serious and stable (where we're _years_ ahead), new workers who explicitly ask for them will get LL Cat 4.[/QUOTE]

Just an update... The Skylake bug hitting the press has generated a "mini-surge" of assignment requests in the DC Cat 4 range, and to a lesser degree in the LL Cat 4 range (as expected).

We're looking good, even in the LL Cat 4 range, but it's a good thing we refocused some of our firepower to be ready for the "big surge" expected when the new MP is announced!

Thanks for all the cycles everyone!

[QUOTE=chalsall;422043]assignment requests in the DC Cat 4 range, and to a lesser degree in the LL Cat 4 range (as expected).
[/QUOTE]

I'm curious about what the Cat ranges are. I see them talked about semi-frequently on the forums and I would really appreciate if someone could explain the concept to me.

[QUOTE=dragonbud20;422524]I'm curious about what the Cat ranges are. I see them talked about semi-frequently on the forums and I would really appreciate if someone could explain the concept to me.[/QUOTE]

I'm not sure if GPUto72 uses the same one as primenet, but the latter are here [url]http://www.mersenne.org/thresholds/[/url]

I think GPU72 follows the same rules, since it is pulling assignments from PrimeNet.

[QUOTE=kladner;422528]I think GPU72 follows the same rules, since it is pulling assignments from PrimeNet.[/QUOTE]

Yes. In fact, it's a little more complicated than that...

GPU72 gets assignments from Primenet through a reservation spider, and then "loans" them out to GPGPU TF'ers and some (GPGPU or CPU) P-1'ers.

We carefully balance the available fire-power to ensure all the categories are appropriately "fed", trying to avoid any LL assignment being issued by Primenet not already TF'ed optimally nor not already P-1'ed. We're not always 100% successful, but we do our best.

Please note that this is why we're currently focusing about half our cycles on the "LL Cat 4" range, in preparation for the new Mersenne Prime announcement on Tuesday. We currently have a comfortable buffer for LL Cat 1 and 2, and are "just in time" for low Cat 3, but are a little tight for P-1.

[QUOTE=chalsall;422547]
Please note that this is why we're currently focusing about half our cycles on the "LL Cat 4" range, in preparation for the new Mersenne Prime announcement on Tuesday. We currently have a comfortable buffer for LL Cat 1 and 2, and are "just in time" for low Cat 3, but are a little tight for P-1.[/QUOTE]
I would be happy to work in that range for TF, if you feel the need. I have been taking WMS assignments for some time now. Just say, or change my internal GPU72 status, or whatever helps.

[QUOTE=kladner;422550]I would be happy to work in that range for TF, if you feel the need. I have been taking WMS assignments for some time now. Just say, or change my internal GPU72 status, or whatever helps.[/QUOTE]

Thanks. So you (and everyone) knows, this happens automatically. The (rather primitive) AI assigns work as needed at the time of assignment, taking into consideration the heuristic latency of each user's assignment to completion temporal delta.

Those who choose "What Makes Sense" (WMS) or "Let GPU72 Decide" (LG72D) will be assigned what (range) is needed most. The difference between the two is that WMS honours the pledge level given (for example, only going to 72), while LG72D will set the pledge level needed (for example, 75).

All other options (for example, "Lowest TF Level", "Lowest Exponent", et al) honours the request and the pledge level.

Cool! It is good to know that I have been doing what is needed. I have kind of given up on factor chasing by getting lower bit level assignments. My understanding is that it is advantageous to let AMD GPU people have these. :smile:

Would you say P-1 is in need of assistance? I can move my Cat I DC box to P-1 temporarily if needed.

[QUOTE=Dubslow;422596]Would you say P-1 is in need of assistance? I can move my Cat I DC box to P-1 temporarily if needed.[/QUOTE]

We're good.

Wow!!! What a week!
 

An exciting week around here! :smile:

So everyone in "TF Land" knows...

The unprecedented surge in LL/DC assignment requests from Primenet is going well. The new Primenet policy of giving new users DC Cat 4 work (unless they explicitly ask for LL work, in which case they're initially given LL Cat 4) has worked out perfectly.

But thank goodness the "RIP DCTF" team had made so much progress! For a while there were over 600 DC requests an hour, and even now we're still seeing somewhere around 250 to 300 requests an hour. We're still years ahead, but it was good that such a large buffer existed!

What was somewhat unexpected was the number of new users explicitly asking for LL work. While we had (and still have) a reasonable buffer for LL Cat 4, additional firepower was redirected from low LL Cat 3 to low Cat 4 to ensure we remained "good".

As far as I can tell, not a single assignment was issued by Primenet for DC'ing nor LL'ing (nor PM-1'ing) not optimally TF'ed (although not always at the lowest candidate for each category).

Please know that this balancing act will be continuing for a few days longer, so those who choose "What Makes Sense" or "Let GPU72 Decide" may see their assignments bounce around the different ranges. Don't be alarmed by this.

Questions and/or comments welcome.

And, as always, thanks for all the cycles everyone! :tu:

M49 may be a good opportunity to split this thread (from the last post(s)) into a "new" one with less pages? Easier to handle and find the info? (it already contains lots of subjects together, like a "thematic ho-pouri thread" :wink:

(just saying it, but if it is a stupid idea, please ignore it and/or delete the current post)

[QUOTE=LaurV;423668]M49 may be a good opportunity to split this thread (from the last post(s)) into a "new" one with less pages? Easier to handle and find the info? (it already contains lots of subjects together, like a "thematic ho-pouri thread" :wink:

(just saying it, but if it is a stupid idea, please ignore it and/or delete the current post)[/QUOTE]
That could have advantages. I don't know what the general effect of creating a " [URL="http://www.mersenneforum.org/showthread.php?p=423668#post423668"] [/URL]GPU to 72 status...02" would be. It might prove more confusing even than 357+ posts in a thread. :confused2:

RIP DCTF...
 

Just so all those working in the DCTF range know...

The last of the DCTF candidates have now been assigned. So, I've added a "rewrite rule" such that DCTF requests instead get LLTF.

The same "Option" rules apply. As in, if "What Makes Sense" is selected the pledge level (for example, 72 or 73) is honoured. If "Let GPU72 Decide" is selected, then the pledge level is set (currently to 75).

Those who use a fetching spider such as MISFIT or mfloop may want to adjust their settings; the above was done to ensure no "fired and forgotten" systems ran dry.

[QUOTE=chalsall;423914]
The last of the DCTF candidates have now been assigned. [/QUOTE]

That is excellent news.

Chris, are you going to bring in the 62M to 74 bits for DCTF?

tl;dr:

I mean, it would make some sense, and they are only ~2000 of them. For the other marked in [URL="http://www.gpu72.com/reports/current_level/"]your table[/URL] (60M and 61M to 74) it doesn't make any sense anymore, neither for AMD cards, now that we have a better clLucas which almost can compete with cudaLucas, so you can mark them as yellow. But the 62M are not many, so there is a little bit of sense to tickle them, especially for a gtx 560, 570, or for AMD cards (good Nvidia cards like 580s or Titans are already better doing DC than DCTF, at these bitlevels).

[QUOTE=LaurV;423948]Chris, are you going to bring in the 62M to 74 bits for DCTF?[/QUOTE]

I could certainly do that, if people wanted to do that work.

But, let's wait a little while, and let the last of the 50M to 57M finish to 73 (or, at least, a good part of it).

My thinking is while we're looking good for the LLTF'ing in the various ranges, we're still "just in time" for low LL Cat 3 and Cat 4. Further, as new workers prove their commitment and stability, more and more LL Cat 3 assignments are going to start being given out. It would be good to build up a bit more of a buffer at the low end of the ranges.

Secondly, this would give those who have spiders configured to automatically get DCTF work time to adjust their settings. Some have "Let GPU72 Decide" set (which sets the pledge level), but might not want to TF to 74 (e.g. those running AMD GPGPUs).

Lastly, I would have to write a bit of code to issue appropriate LLTF assignments for anyone who has the "What Makes Sense" et al Option set. What I did yesterday was a quick and dirty "rewrite rule" hack.

tl; dr: Are there those out there who would like to take DCTF 62M to 74? 61M? 60M?

If so, please say so.

If not, now would be a good time to adjust your spider's settings to ask for LLTF.

Nothing to do with the current topic, but I notice on my individual statistics page the bottom graph is getting ugly with overlapping X-axis labels. Maybe you could limit it to no more than 8 labels so there's room?
[url]http://www.gpu72.com/graphs/worker/56f1b7572536a14513b08c88b2ba9578/[/url]

Slight tweaks to the "View Assignments" report.
 

Hey all.

Just so everyone knows, I've made some slight tweaks to the [URL="https://www.gpu72.com/account/assignments/bydate/"]View Assignments[/URL] report.

Nothing drastic. The Age and Estimated Completion columns are now rendered to a 0.01 days resolution. The Percent column now correctly takes into account the two stages of P-1'ing; previously the count would climb up to 100% during stage 1, and then drop back down and count up again during stage 2.

As always, please tell me if I've made a SPE anywhere.

What would the impact be to GPU-to-72 if the LL assignment rules changed to encourage more machines working on cat 2 LL assignments? According to the work distribution map it looks like about 60,000 cat 2 exponents are reserved for TF.

[QUOTE=Prime95;429842]What would the impact be to GPU-to-72 if the LL assignment rules changed to encourage more machines working on cat 2 LL assignments?[/QUOTE]

It would be a Good Thing [sup](TM)[/sup]!

And I think you're mis-reading the distribution map, confusing Cat 2 with Cat 3. At the moment there are only 1,030 Cat 2s reserved for TF'ing, with 2,683 Cat 2 candidates ready for LL assignment (and another 3,452 Cat 1s).

Assigning more Cat 2 (and Cat 1) had been discussed on the Milestone thread a while ago, but that conversation seemed to have petered out. I think it would be really good if this was re-opened. The pace of Cat 1 and Cat 2 completions is very slow from a number-of-assignments perspective.

[QUOTE=chalsall;429856]
And I think you're mis-reading the distribution map, confusing Cat 2 with Cat 3.

I think it would be really good if this was re-opened.[/QUOTE]

Yes, I was confused.

I intend to re-open the discussion. Prior to doing so, I wanted to gauge the impact on GPU-to-72.

[QUOTE=chalsall;429856]The pace of Cat 1 and Cat 2 completions is very slow from a number-of-assignments perspective.[/QUOTE]

If anyone has the data, I would like to see how many assignments are handed out /completed per day by category. If there is a steady pattern, it is probably as simple as readjusting the cat 1 & 2 pool size.

[QUOTE=axn;429871]If anyone has the data, I would like to see how many assignments are handed out /completed per day by category. If there is a steady pattern, it is probably as simple as readjusting the cat 1 & 2 pool size.[/QUOTE]

I don't have hard data, but from [URL="http://www.mersenne.ca/status/ll/0/30/3/6000"]James' report[/URL] only about 27 Cat 1s are completed a day. Thus, it's going to take about 150 days just to complete what's already in the Cat 1 range. Note that almost no Cat 2s are being assigned.

With regards to your second point, I disagree that changing the pool size(s) would help. What would be better (IMHO) is assigning more Cat 1s and Cat 2s to machines which heuristically will complete them in a timely matter but haven't "promised", and thus are currently given Cat 3s.

But let's continue this discussion over on the Milestone thread where a lot of this discussion has already taken place (or, perhaps, a new thread specifically to discuss the strategy).

[QUOTE=chalsall;429872]I don't have hard data, but from [URL="http://www.mersenne.ca/status/ll/0/30/3/6000"]James' report[/URL] only about 27 Cat 1s are completed a day. Thus, it's going to take about 150 days just to complete what's already in the Cat 1 range. Note that almost no Cat 2s are being assigned.
[/QUOTE]

Do I interpret this correctly as meaning very few people are choosing to ask for preferred assignments here:
[url]http://www.mersenne.org/thresholds/[/url]

If so I am surprised with so much chatter in the past regarding people holding up milestones by taking "for-ever" to complete assignments at the trailing edge.

I may be MORE surprised at how many of these preferred assignments expire...considering the "you must prove worthy" rules to get them in the first place.

[QUOTE=petrw1;429879]Do I interpret this correctly as meaning very few people are choosing to ask for preferred assignments here:
[url]http://www.mersenne.org/thresholds/[/url][/QUOTE]

That is correct. Just take a look at the empirical: only [URL="http://www.mersenne.org/primenet/"]906 Cat 1s[/URL] are currently assigned.

Unscheduled downage...
 

Just so everyone knows, GPU72 is currently offline. It should be back in about two hours.

Edit: Scratch that. It's back now.

Not that I'm a huge TF contributor, but a heads-up I'm retiring my [url=http://www.ebay.ca/itm/191848267263]GTX 570[/url] and [url=http://www.ebay.ca/itm/191848711671]GTX 580[/url] (both currently going for a dollar, hope that improves :smile:) and I'm going to be running my GTX 670 on [url=http://www.mersenne.ca/tf1G.php?available_assignments=1]TF >1000M[/url] for a while. I'll continue to do P-1 for GPU72, for what that's worth.

[QUOTE=James Heinrich;431687]I'll continue to do P-1 for GPU72, for what that's worth.[/QUOTE]

Thanks for the heads up, and the cycles (past and future).

GPU72 saved a year of LL ... kind of
 

Total Saved by GPU72 has now reached 18.5M GhzDays.
[url]http://www.gpu72.com/reports/workers/[/url]

Total GhzDays of LL work done this year totals about 18.2M.
[url]http://www.mersenne.org/report_top_500_custom/?team_flag=0&type=1003&rank_lo=1&rank_hi=9500&start_date=&end_date=[/url]

Now I realize NOT all the work saved is LL (some is DC too) but the majority of it is LL so close enough for me to make this post.

[QUOTE=petrw1;433554]Now I realize NOT all the work saved is LL (some is DC too) but the majority of it is LL so close enough for me to make this post.[/QUOTE]

Yeah. We've done some really great work here.

Interestingly, we're still only just keeping ahead of the P-1'ers.

[QUOTE=chalsall;433566]Interestingly, we're still only just keeping ahead of the P-1'ers.[/QUOTE]

Just wait until our famous Anonymous guy, who did most of the DCTF and now switched to LLTF starts getting Assignments in P-1 ranges, then we'll be set for some time again I guess

[QUOTE=manfred4;433628]Just wait until our famous Anonymous guy, who did most of the DCTF and now switched to LLTF starts getting Assignments in P-1 ranges, then we'll be set for some time again I guess[/QUOTE]

Yeah, but that's three months out or so. What is really annoying is there are some users who get hundreds of manual P-1 assignments from Primenet and then never work them. These are held up for six months before being released back into the pool.

Also, just so everyone knows... We are very comfortably ahead of the Cat 1, 2 and 4 wavefronts. We're currently balancing most of our fire power keeping ahead of the P-1'ers (in 77M) and the Cat 3 LL'ers (in 73M).

Interestingly, the assignments for Cat 3 and 4 are less than the daily offset jump. As in, many candidates are being "skipped" as the boundaries move upwards. Cat 2 assignments, on the other hand, are currently being fully assigned.

What this means is that in a few months Cat 1 will consist of candidates which were previously assigned but were then recycled, mostly because of abandonment. Exactly as the newest assignment rules were designed for.

I wonder if someone of those manual assignments were requested by bots. Perhaps it would be a good idea to require non-logged-in users to complete a CAPTCHA?

[QUOTE=ixfd64;433668]I wonder if someone of those manual assignments were requested by bots. Perhaps it would be a good idea to require non-logged-in users to complete a CAPTCHA?[/QUOTE]

Possibly a CAPTCHA would be a good idea. But for the most part the manual P-1 assignments are by logged in users.

See [URL="http://www.mersenne.org/assignments/?exp_lo=74000000&exp_hi=80000000&execm=1&exdchk=1&exfirst=1&extf=1"]this report[/URL] for example.

I had recently taken to running 2 P-1 instances, along with 3 DC's in P95. I did this mainly because 25 GiB of idle RAM was begging for some action. However, when I realized that feeding P-1 is the primary TF challenge, I decided to switch the CPU back to all DC.

This proved easier said than done. I have a well-worn routine for changing assignment types, but it did not work this time. Usually, [INDENT]1) I stop P95
2) Change the settings in Worker Windows
3) Shut down P95 if I am changing worktodo.txt manually
4) Go to My Account>CPU and make changes there
5) Start up P95 and communicate with the server
[/INDENT]I also unreserved P-1 assignments which had not started, and moved the remaining ones to the same worker.

When P95 called home, it promptly pulled some more P-1s. I ditched these, first in GPU72. When they came back, I also used P95 to unreserve, as well as GPU72. Like bad pennies, they kept coming back.

I finally took the ultimate step of changing the workers in question to 1st LL on Prime Net, communicating, and then changing them back to DC, and communicating again. This seems to have succeeded. :smile:

Could be done to the end of 80M by October ...
 

At what point do you go to 76 bits?

80M?
85M?
90M?

I know there's a graph somewhere in mersenne.ca

[QUOTE=petrw1;434216]At what point do you go to 76 bits?[/quote]Around 100M-110M

[QUOTE=petrw1;434216]I know there's a graph somewhere in mersenne.ca[/QUOTE]Presumably you mean [url=http://www.mersenne.ca/cudalucas.php?model=490]this one[/url]. It doesn't go that high, but if it did it would look like this (all the way up to 1000M):

I would guess earlier than that. Mostly the global bit level change goes in line with the turnover point which is highest for current GPUs.
Another kinda 'rule' I found is that can approximately take the last point where we changed the bit level (around 70M or 66M for the "enthusiasts") which would relate to 76 bits at around 90M (or around 85M for the enthusiasts)

[QUOTE=manfred4;434231]I would guess earlier than that.[/QUOTE]
Not really. Titans** are very efficient bitches for LL. You have to add about 0.8-0.9 bits (i.e. move the red/orange lines up) for a gtx580, for example, and in that case your feeling would be right.

--------------
** maybe you didn't see the graphic is for a Titan, but the graphic is correct

(sorry for double post, time limit, been busy all morning with job, lunch break here - and we didn't know that is so difficult to make an animated gif! we had to make a movie first and convert it online - we somehow lost our gif animator program, as we didn't use it for ages...:blush:)

This shows very well how the TF limit is card-dependent, we were talking here for ages about that, but an image worth a thousand words... :razz:

[ATTACH]14387[/ATTACH]

[edit: from this image it seems that the offset is not 0.9 bits, but is well over a full bit, more like 1.3 bits or so. I will have to check my... memories]

I use Manual Assignments to get TF candidates for my GPUs [around 1000GHz-day/day total] and they are all at least 128M, way ahead of the 75M-80M LL wavefront.

Does this mean that all the TF necessary around the LL wavefront is completed? Or just that all the LLTF is already assigned to GPU to 72 and awaiting completion?

Let's say I want to do more TF between 75M and 100M how can I get those candidates assigned without joining GPU to 72?

How is that graph applied IRL please?

Does it tell us based on how long a particular CPU or GPU takes at TF at increasing bit levels vs LL that there comes a point at which further TF is pointless and LL should be done instead?

If so, given that the recommendation is
1. use your whatever GPU for TF and not for LL
2. use your whatever CPU for LL and not for TF

How can we then project a definite crossover point for TF to LL for each CPU and GPU when we don't know what GPU(s) [sometimes multiple persons will be assigned for TF at different bit levels] will be used for TF and what CPU for LL by computer D?

The crossover point seems necessarily arbitrary for such a distributed project.

[QUOTE=vsuite;434740]Or just that all the LLTF is already assigned to GPU to 72 and awaiting completion?
[/QUOTE]

That is the case.
Also, related to manual assignments (and the rest of your first post), do you have something against Misfit, too? You can run Misfit and configure it to get the assignments for you, either from GIMPS or GPU72. There is no drawback joining GPU72. You still can get the exponents you want, by setting Misfit convenient, and when you finish with them, Misfit will report the results for you, [U]to PrimeNet directly[/U]. Then GPU72 will see they are done, and delete them from its assignment list. You still get all your credit at GIMPS/PrimeNet, and you get additional credit at GPU72.

But if you don't want that, you still can get (low) assignments directly from PrimeNet, with Misfit, although the chances to get them so low as 7xM are small because GPU72 will grab them.

In either case, Misfit will do [U]all the manual work[/U] for you. Bye bye manual work, writing scripts, etc...

[QUOTE=vsuite;434741]How is that graph applied IRL please?

<snip>[/QUOTE]

You got it right. The main goal of the project is to find primes, and texnically* you can find them faster if you can eliminate faster the exponents which will not give a prime. So, [U]you have to tune your system for every range you want to participate.[/U] Every system is different. I get much better results towards LL with my water-cooled cards than with the air-cooled, even if everything else is the same, for example. That is because LL uses the memory of the card a lot, when TF does not.

If you can eliminate more than one exponent by doing TF at a particular bitlevel and size of the exponent, that is, if you can find [U]new factors[/U] for more than one exponent, faster then your same hardware would do two LL tests, then you do TF. Otherwise you better do LL.

Same for DC, but only one test, that is why the DC line is lower with one bit, as the TF required time doubles with every bit.

--------
* this is not exactly true, there are many discussions, which generally agree that we will find prime faster if for example will not do any DC, etc. But that is behind the point, as the goal of GIMPS is not only to find primes, but also clearing the ranges behind, factoring small mersennes, etc. By "the goal of gimps" I mean the common goal of all participants, and not the officially declared goal. People do whatever type of job they think it better suits their particular goals and vanities.

[QUOTE=vsuite;434740]I use Manual Assignments to get TF candidates for my GPUs [around 1000GHz-day/day total] and they are all at least 128M, way ahead of the 75M-80M LL wavefront.

Does this mean that all the TF necessary around the LL wavefront is completed? Or just that all the LLTF is already assigned to GPU to 72 and awaiting completion?

Let's say I want to do more TF between 75M and 100M how can I get those candidates assigned without joining GPU to 72?[/QUOTE]

GPU72 is holding on to everything below 82M that isn't trial factored to what GPU72 considers high enough. You should be able to get exponents higher than that.

Also keep in mind that PrimeNet has a much lower threshold for TF being completed, as it's considering the case of CPUs. PrimeNet considers an 85M exponent to be fully factored at 71 bits. In fact, every exponent up about 125M or so is already TF'ed to 71 bits. If you want something between 82M and 100M, set the pledge level to 72 or 73 and you should get exponents.

[QUOTE=Mark Rose;434764]GPU72 is holding on to everything below 82M that isn't trial factored to what GPU72 considers high enough. You should be able to get exponents higher than that.[/QUOTE]

Further to this, GPU72 is currently managing on average 50,000 GHz Days per Day of GPU TF compute.

We carefully balance our resources to ensure all the "wave fronts" are appropriately "fed", including the very hungry P-1'ers.

In the past people have accused us of "hoarding" TF assignments. This is not the intention; instead we hold onto candidates until appropriately TF'ed so no candidates are assigned sub-appropriately worked.

If you don't want to work through GPU72, that's fine. There should be a way for you to get candidates to TF with your GPU through Primenet in the 82M range.

Thanks everyone.

On average is it still 1 in bit level factors found at the 82M level and the 128M level? eg 1 in 75 exponents tested at 75 bits will give a factor?

[QUOTE=LaurV;434743]That is the case.
Also, related to manual assignments (and the rest of your first post), do you have something against Misfit, too? You can run Misfit and configure it to get the assignments for you, either from GIMPS or GPU72. There is no drawback joining GPU72. You still can get the exponents you want, by setting Misfit convenient, and when you finish with them, Misfit will report the results for you, [U]to PrimeNet directly[/U]. Then GPU72 will see they are done, and delete them from its assignment list. You still get all your credit at GIMPS/PrimeNet, and you get additional credit at GPU72.

But if you don't want that, you still can get (low) assignments directly from PrimeNet, with Misfit, although the chances to get them so low as 7xM are small because GPU72 will grab them.

In either case, Misfit will do [U]all the manual work[/U] for you. Bye bye manual work, writing scripts, etc...[/QUOTE]
Sounds good.

This Factoring Effort Report page can give a list of exponents not carried to 75 bits for example, but I assume even if I added these to mfaktc, there could be a risk someone else legitimately gets them using MISFIT especially if I take a couple of days to return the results.

[URL]http://www.mersenne.org/report_factoring_effort/?exp_lo=82000000&exp_hi=82050000&bits_lo=1&bits_hi=99&txt=1&exassigned=1&tfonly=1&worktodo=1&tftobits=75[/URL]

I don't have something against MISFIT per se. I've just downloaded it. Just not set it up yet.

[QUOTE=Mark Rose;434764]{snip...}Also keep in mind that PrimeNet has a much lower threshold for TF being completed, as it's considering the case of CPUs. PrimeNet considers an 85M exponent to be fully factored at 71 bits. In fact, every exponent up about 125M or so is already TF'ed to 71 bits. If you want something between 82M and 100M, set the pledge level to 72 or 73 and you should get exponents.[/QUOTE]
I guess this means 'pledge level' using MISFIT.

I assume it is not possible to get manual assignments from Primenet above 82M to carry to 75 bits since PrimeNet considers them all completed at 71 bits.

[QUOTE=vsuite;434857] eg 1 in 75 exponents tested at 75 bits will give a factor?[/QUOTE]
Yes. Please mind that this calculus is "coarse", i.e. it ignores some things, as explained in the [URL="http://www.mersenne.org/various/math.php#lucas-lehmer"]gimps math page[/URL], see the "chance to find a factor" at the end of "LL test" subsection. For our purpose, "chance of finding a factor between 2[SUP]X[/SUP] and 2[SUP]X+1[/SUP] is about 1/x". Now, that is (much) better if there was no P-1 done in the range (up to almost double, as you go higher with the exponents) and it is much worse (down to half of it) if there was lots of P-1 done for the range. You can look to Chris' [URL="http://www.gpu72.com/reports/factor_percentage/"]tables[/URL] to see that odds, keeping in mind that the DC range had a lot of P-1 (therefore, there was about 1 factor found for every 100 "bitruns" or so, at bitlevel 73 or so,), and the first-LL had not much of it (therefore as good as 1 factor was found for every 60 or even 50 runs, at the same bitlevels).

I too have been wondering if there was a way for non-GPU-to-72 people to get exponents in the 80-90M range, and haven't run across one yet. (I am using Misfit to fetch directly from GIMPS, and getting work in the 130M range - which is fine, but it'll be a good long while before any benefit from that factoring is seen.)

I am not necessarily averse to joining GPU to 72, either, though I havent as yet done anything more than wade through some of the documentation.

I DO sort of wish that the Primenet assignment list wasn't "messed up" by all the just-in-time trial factoring... maybe that will take care of itself in some few months, if GPU to 72 gets ahead of the wavefront, so that most of the LL range will not show as checked out for factoring anymore?

[QUOTE=Siegmund;434887]I am not necessarily averse to joining GPU to 72, either, though I havent as yet done anything more than wade through some of the documentation.[/QUOTE]

It really isn't that difficult, especially when using MISFIT. After the initial account creation (which is only needed in order to keep track of assignments and to give credit appropriately) it's pretty much "fire-and-forget". And, just to be clear, while GPU72 "owns" the candidates, all credit for work goes to the workers.

[QUOTE=Siegmund;434887]I DO sort of wish that the Primenet assignment list wasn't "messed up" by all the just-in-time trial factoring... maybe that will take care of itself in some few months, if GPU to 72 gets ahead of the wavefront, so that most of the LL range will not show as checked out for factoring anymore?[/QUOTE]

This is needed in order to ensure no candidates are assigned for LL'ing without being appropriately TF'ed and P-1'ed. And rather than "messed-up", consider the TF column on Primenet as reading "Not yet appropriately TF'ed".

Anything in the "To be LL assigned" (second to right-most) column has already been appropriately TF'ed and P-1'ed, and anything in the "To be P-1 assigned" (third from right-most) column has already been appropriately TF'ed, but not yet P-1'ed.

In about three months or so we should have pulled ahead of all four LL'ing wave-fronts, and the P-1'ers. At that time I plan to pass control back to Primenet, with GPU72 acting as a proxy so the "pretty graphs" and stats will continue for those interested.

[QUOTE=Siegmund;434887]I too have been wondering if there was a way for non-GPU-to-72 people to get exponents in the 80-90M range, and haven't run across one yet. (I am using Misfit to fetch directly from GIMPS, and getting work in the 130M range - which is fine, but it'll be a good long while before any benefit from that factoring is seen.)

I am not necessarily averse to joining GPU to 72, either, though I havent as yet done anything more than wade through some of the documentation.

I DO sort of wish that the Primenet assignment list wasn't "messed up" by all the just-in-time trial factoring... maybe that will take care of itself in some few months, if GPU to 72 gets ahead of the wavefront, so that most of the LL range will not show as checked out for factoring anymore?[/QUOTE]

I just tried [url]http://mersenne.org/manual_gpu_assignment[/url] and entered:

TF for first-time LL
80000000 to 85000000
Lowest exponents
75 (bit level)

and got an assignment just above 82000000

Aha. So one could download a block of those manually, and paste them in to MISFITworktodo.txt. I think I will do that, next time my supply of exponents is low. (Of course I just got 50 more in the M130s just this morning.) Thanks for pointing out how that works.

@Siegmund: what is your pledge level in Misfit? If you go to 75, there is no reason why you shouldn't get ~83M work to do. :surprised:

[QUOTE=Prime95;434898]I just tried [URL]http://mersenne.org/manual_gpu_assignment[/URL] and entered:

TF for first-time LL
80000000 to 85000000
Lowest exponents
75 (bit level)

and got an assignment just above 82000000[/QUOTE]

Hmm. Can this be put as a link in the regular mersennse.org menu please?
Also the experimental List of Assignments page that showed the Category of the LL and DC tests and all the other fields?

@LaurV: yes, my Misfit is set to 2^75 -- and every automatic assignment I've ever gotten has been in the nosebleed region as if it were a regular manual CPU request from primenet.

Is it time for 76 bits? It seems the crossover point is between 80M and 90M, depending on the card (80M for a 1080, 84M for my 580's, 90M for my 760, which gets occasional work).

It also appears we'll be finishing the work for everything that isn't category 4 within a couple of weeks.

[QUOTE=Mark Rose;439168]Is it time for 76 bits? It seems the crossover point is between 80M and 90M, depending on the card (80M for a 1080, 84M for my 580's, 90M for my 760, which gets occasional work).

It also appears we'll be finishing the work for everything that isn't category 4 within a couple of weeks.[/QUOTE]

The answer would seem to be - yup.

Is it better at this stage to go straight from 72->76 for the next grouping? Or should we concentrate on lifting a large forward region up one layer at a time once we are ahead of demand?

[QUOTE=airsquirrels;439174]The answer would seem to be - yup.

Is it better at this stage to go straight from 72->76 for the next grouping? Or should we concentrate on lifting a large forward region up one layer at a time once we are ahead of demand?[/QUOTE]

From what I can tell, we're already of the LL'ers, except the 6.1K at 71M (which should be done shortly).

We're still behind the P-1'ers, which are working in the 81M range. If I understand correctly the P-1 rate can be somewhat controlled by how high we take the TF work, since the more TF done, the smoother any factor found by P-1 would have to be, reducing the range to be checked by P-1.

I guess it might make sense to wrap up 79M to 80M at 75 bits first, then start going to 76 bits.

[QUOTE=airsquirrels;439174]The answer would seem to be - yup.[/QUOTE]

Definitely. Really right now we're mostly continuing to feed the P-1'ers. The 71M range should be finished well before the Cat 2 wavefront gets there.

[QUOTE=airsquirrels;439174]Is it better at this stage to go straight from 72->76 for the next grouping? Or should we concentrate on lifting a large forward region up one layer at a time once we are ahead of demand?[/QUOTE]

I think that's largely a function of what people want to do, taking into consideration what type of card they have.

Also, as I said earlier, I plan to pass control of TF'ing back to Primenet once we're comfortably ahead of all the wavefronts (including the P-1'ing), when GPU72 no longer has to hold candidates to prevent them from being assigned sub-optimally TF'ed and P-1'ed.

GPU72 will continue to function (pretty graphs, people competing in the rankings, etc.), but will be more of a proxy to Primenet; much like the transition for the LL'ing / DC'ing. George, Aaron and I will have to do some work to complete this, but it should be mostly transparent to everyone else.

Also, a separate topic is the 332M range. People seemed to have lost interest in TF'ing there, so I released all the candidates back to Primenet to assign as it seemed best. There's still a great deal of TF'ing to be done up there, but it seemed unreasonable for GPU72 to hold candidates when work wasn't being done on them.

As always, feedback welcomed. And also, as always, thanks for all the cycles everyone! :tu:

[QUOTE=Mark Rose;439176]We're still behind the P-1'ers, which are working in the 81M range.[/QUOTE]

We cross-posted...

Actually, if you drill down on the 1882 P-1 assignments in 81M, they are all held by a user who likes to come into Primenet and take thousands of Manual Assignments at a time, and then never completes any of them.

GPU72 pulled its "rip-cord" within a few minutes of observing this, and released candidates appropriately TF'ed so real P-1 workers would get them. The said 1882 assignments should expire mid October for proper processing.

Potential downtime...
 

So, as the recent issue with HTTPS connections has demonstrated, the GPU72 server needs to be upgraded. The poor little thing is ancient! CentOS 5.11!

The underlying hardware is still good (it's a co-located server), but I might take the opportunity to migrate permanently over to an EC2 or GCE instance instead.

I will let everyone know on this thread when changes are taking place, while trying to minimize any disruption. But if anyone sees anything strange, please bring it forward.

Having run EC2 extensively, I suggest using GP2 for storage. It will perform more than well enough, even if not using an EBS optimized instance. If you can tune RAM, a c4.large would be perfect.

[QUOTE=chalsall;439276]...but I might take the opportunity to migrate permanently over to an EC2 or GCE instance instead.[/QUOTE]

So, I was in the process of spinning up an EC2 and a GCE instance to compare their performance vs. pricing compared to my current co-located server in preparation for the GPU72 upgrade.

To my shock, there is no way to set a reverse DNS lookup PTR record for an IP address under GCE, even with a reserved IP!

Further, I've read (but haven't empirically tested to confirm myself) that all outbound port 25 (SMTP) traffic is blocked under GCE. (Based on the amount of spam I receive from EC2 instances, clearly this isn't the case there.)

Hmmm... While reverse lookups won't be needed by most people, this lack of functionality is troubling. There have been trouble tickets open on this with Google since 2013; they seem to be just ignoring it.

I don't believe AWS has reverse PTR customization either. I've never seen it.

Also, if you are going to be sending email, I'd configure your MTA to relay through SES. Most of the EC2 IPs have a bad reputation.

[QUOTE=Mark Rose;439411]I don't believe AWS has reverse PTR customization either. I've never seen it.[/QUOTE]

Reverse DNS is possible under EC2. It involves filling out a form, and (possibly) interacting with a human.

[QUOTE=Mark Rose;439411]Also, if you are going to be sending email, I'd configure your MTA to relay through SES. Most of the EC2 IPs have a bad reputation.[/QUOTE]

Thanks Mark. Your advice about AWS is always valued and appreciated.

Is it safe to say the Leading Edge of LL is well behind TF?
 

I'm in no way suggesting people should stop doing GPU72-LLTF.
Of course anyone can do whatever turns their crank.

It seems with TF almost done in the 60M and 70M ranges that there are over 100,000 ready for LL with a good number in all 4 categories under 80M?

We're not quite there with P-1.

Also, it's advantageous to go to 76 bits above 80M for some cards. So that will slow down the progress a lot.

[QUOTE=Mark Rose;441214]We're not quite there with P-1.

Also, it's advantageous to go to 76 bits above 80M for some cards. So that will slow down the progress a lot.[/QUOTE]
I'm sure it has been discussed before, but are GTX 580s part of the group for 76 bits above 80M?

[QUOTE=kladner;441229]I'm sure it has been discussed before, but are GTX 580s part of the group for 76 bits above 80M?[/QUOTE]

84M and above, according to the chart at mersenne.ca, IIRC.

Thanks! It often doesn't occur to me to check mersenne.ca. I should learn. :gah: