New Collatz Conjecture application

[GLeeM]

Today I noticed Collatz taking a very long time to finish WU. After looking at the web site I see that there is a new app exe. 🙄

Anyone running Collatz Conjecture will need to renew their "config" file. Getting the new executable deleted the old config file. The WU will take forever to finish if you don't redo the config file.

I don't remember what I had in each config file so I used what I had in the computer that I have not recently run Collatz.

Here it is, does it look right?

verbose=0
items_per_kernel=19
kernels_per_reduction=9
threads=8
sleep=1

 

[petrusbroder]

Thanks for the head-up, GLeem!
I plan on returning to Collatz rather sooner than later ...

 

[kowalabearhugs]

I would add that only the OpenCL and CUDA apps were updated. Those of us still running the ATI Brook+/CAL apps (ati14 v6.04) need not take action.

 

[waffleironhead]

I didnt even realize this was a thing that needed to be done.
Hmm, after reading thru it seems there is a new value as well to be edited. Lut size?
I have a cape verde m275 that i run some wu's on occasionaly so I guess I need to figure out optimal settings for that. Also, does this need to be done for the ati14 apps as well?

 

[waffleironhead]

Oh my. I just grabbed the sample that slicker posted:
verbose=1
items_per_kernel=20
kernels_per_reduction=9
threads=8
sleep=1
lut_size=12
build_options=-Werror
Though after reading about the lut size, I edited the lut_size up to 14 and boy did the times drop on mt m275. I was doing units in 15 mins...dropped them to ~6 mins. Also, Ive sorely overtaxed this laptops cooling system. The cpu is throttling down from 1.8ghz to 1.2 ghz to make up for the increased load on the shared heatsink. Id rather take 15 mins and not kill the system, so Im going to work with the settings to try and decease the load a bit.

 

[GLeeM]

I edited the lut_size up to 14 and boy did the times drop

Thanks for the information, waffleironhead 🙂

I upped my lut_size now, so we'll see what happens.
I am trying items_per_kernel=22 on my 280X. It worked better at 21 than 20 so we'll see about that too!

verbose=0
items_per_kernel=22
kernels_per_reduction=9
threads=8
lut_size=14
sleep=1

Did they have to delete the old file????!!!

 

[waffleironhead]

Looking at your newer wu's it seems that the lut size didnt make a whole lot of difference if any to your setup. Guess mine was memory bound? so optimizing the amount held in the l2 cache allowed the gpu core to crunch more. I think that your 280x has more l2 than mine so you may be able to bump up the lut size to 15 or 16.

 

[GLeeM]

so you may be able to bump up the lut size to 15 or 16

Thanks waffleironhead 🙂 I bumped lut_size up to 16