Overclock.net › Forums › Overclockers Care › Overclock.net BOINC Team › Hardware Optimization › Collatz Conjecture project support thread
New Posts  All Forums:Forum Nav:

Collatz Conjecture project support thread - Page 17

post #161 of 521

Collatz may have been down temporarily, but everything seems to be good now.  I have 4 GPUs on Collatz, and they are all crunching away.

 

For everyone that is running NVIDIA cards, all of the Collatz Sieve tasks are now OpenCL.  Might be a good idea to check your preferences and make sure that you have the opencl_nvidia_gpu tasks selected to run.

post #162 of 521
It's the way Jon has changed the DB updates. Collatz puts a lot of weight on the servers and we would usually get crashes that could last upwards of a week. He has a setup that flushes the DB every day so it doesn't load up and crash the whole system so occasionally it appears to "Go Down" for a few hours.

But it's not it is simply clearing the database to prevent a larger crash later.

And note that Collatz is limited on the server side to a 200 WU cache so it doesn't matter what you set your client to. The max you will ever get is 200 WU.

Also cause of DB crashing issues. when the project would go down, everyone would still be crunchin away and when it came back up the servers would be inundated with new WU's being returned.

It became a serious issue.

It has been much much more stable now, with the occasional period of down which usually last around 40 to 120 minutes.

Much better than it used to be.

And YES enable Open CL on nvidea, you do not want to cut your own throat...
post #163 of 521
I have been running them prior to last night. The PC is down for some reason now so I can't even check remotely. I started the FAH client on the GPU before I left or work. It must not have like having the CPU at 100% or something.

in FAH, if a unit fails to send for whatever reason, your end or server end, the timer until next attempt to send increases. This way not everyone is sending back units at once when a down server on Stanford's end doesn't go down right away due to the influx.
post #164 of 521
Quote:
Originally Posted by mmonnin View Post


in FAH, if a unit fails to send for whatever reason, your end or server end, the timer until next attempt to send increases. This way not everyone is sending back units at once when a down server on Stanford's end doesn't go down right away due to the influx.

It works that way in collatz also, the standard wait is 5 minutes between connects, if the server is down flushing the DB the next period is 40 minutes, and increments upwards after every connection attempt.

You can manually override it, but it re-index's the wait period if no transfer takes place.

I've gotten to the point of letting it ride for a day, it is usually back up and running in less that 24 and the two time it took longer I didn't have time to fully switch to another project (6 machines) by the time it came back up.

Much better than it going down for a week at a time. So far in the three months since this system has been implemented there hasn't been a major crash. When in the past it was a once a month scheduled outtage. (you just didn't know when it was scheduled)

When that happens give it a day. so far it has worked well.

AS far as a machine quitting, your either overclocked, under powered, or temp protected. Those are the only things that will cause a collatz cruncher to quit.

Overclocking causes the driver to bail, Underpowered causes the card to shut down machine to lock up, over temp causes the whole machine to shut down.

Running 21 tahitis on air, you learn a few things.
post #165 of 521
Quote:
Originally Posted by DarkRyder View Post

This thread is specifically for Collatz Conjecture project support.
Collatz Conjecture Website


Project info:

Collatz Conjecture is based in Wood Dale, Illinois, USA and continues the work of the previous 3x+1@home BOINC project which ended in 2008. It can run on an nVidia GPU, ATI GPU, Intel GPU, or Intel or AMD CPU.

Collatz Conjecture is a research project that uses Internet-connected computers to do research in mathematics, specifically testing the Collatz Conjecture also known as 3x+1 or HOTPO (half or triple plus one). You can participate by downloading and running a free program on your computer.

Optimizing Collatz:

When you download the Collatz app you will  have some empty gpu.config files in your Collatz folder.  These files can be populated with values to optimize how Collatz Conjecture runs on your GPU.

These configs are generally configs that will maximize the output of your GPU.  With increased GPU usage you may see a rise in GPU core and VRM temps, so it is a good idea to monitor temps for a few tasks.
More info on Optimizing Collatz can be found on the Collatz forum: Optimizing Collatz 

Below is an example gpu.config from @fragamemnon


gpu.config settings:
gpu.config for R9-290(x) This gpu.config also increased production on an NVIDIA GTX 980

Code:
Code:
[code]verbose=1
kernels_per_reduction=128
threads=8
lut_size=17
sieve_size=30
sleep=1
reduce_cpu=0
[/code]





**This will be updated by the BOINC Editors from time-to-time.

Tagging @4thKor. You should try this with your army of 980s. I used the options above and it cut down collatz by almost half. 3:10 to 1:40 but it does create some serious input lag. OK for dedicated systems.
post #166 of 521
Collatz is down for maint..
post #167 of 521
Well, unfortunately, tomorrow we will lose a spot to GC in Collatz.

.60 days before overtake.

But we put up on hell of a fight. But we also see why GC didn't fare so well in the Pentathlon cross country. They left most of their GPU's at home....

But we get it back in a week when we pass SUSA......
post #168 of 521

I just threw a 290 at Collatz, so maybe we can hold them off for a little bit longer.  They dipped a little and we are getting a little boost.  Either way, not too bad for 9 vs 90. :D

post #169 of 521
Quote:
Originally Posted by tictoc View Post

I just threw a 290 at Collatz, so maybe we can hold them off for a little bit longer.  They dipped a little and we are getting a little boost.  Either way, not too bad for 9 vs 90. biggrin.gif

Very true, we are only 27 million ahead of them right now and they posted 8 million more than we did. So the distance is actually a couple of days if current production holds, any and all help would be appreciated.. Mine are finally back to speed but I'm down a GPU. (converting a 7990 to water) Replaced it with a 7970 temporarily.

Wanted to take a break from heavy crunchin after the Pentathlon, but it looks like that isn't possible....

Oh well, on to 11 billion.
post #170 of 521
Quote:
Originally Posted by Egilman View Post

Very true, we are only 27 million ahead of them right now and they posted 8 million more than we did. So the distance is actually a couple of days if current production holds, any and all help would be appreciated.. Mine are finally back to speed but I'm down a GPU. (converting a 7990 to water) Replaced it with a 7970 temporarily.

Wanted to take a break from heavy crunchin after the Pentathlon, but it looks like that isn't possible....

Oh well, on to 11 billion.

I can throw a couple of 980's in the ring.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware Optimization
Overclock.net › Forums › Overclockers Care › Overclock.net BOINC Team › Hardware Optimization › Collatz Conjecture project support thread