Skip & Limit Calculator

[evilmog]

If you are like me and hate middleware I've developed a script, because I'm bloody lazy,  I've made the default the DES keyspace but you can make it whatever you want.  To use the script modify the array to list the number of GPU's in each node,  then execute, if you specify an argument it will override with a new keyspace.  It will output the -s and -l values for complete distribution taking into account the number of GPU's.  Now if your GPU's are say twice as powerful on one node as say another you can adjust to get better skip and limit values.  I use all GTX 980's so I coded with that in mind.


***BEGIN OUTPUT***


Keyspace: 34359738368
Chunk size: 858993459
Node count: 8

Node 1: -s 0 -l 6871947672
Node 2: -s 6871947672 -l 6871947672
Node 3: -s 13743895344 -l 3435973836
Node 4: -s 17179869180 -l 3435973836
Node 5: -s 20615843016 -l 3435973836
Node 6: -s 24051816852 -l 3435973836
Node 7: -s 27487790688 -l 3435973836
Node 8: -s 30923764524 -l 3435973844


***END OUTPUT***

Let me know if you find this useful, I fixed a bug by adding a remainder calculation to deal with a rounding error and put it on the final node.

*** EDIT ***
One final update, fixed more bugs and made it cleaner

[atom]

should work on other algorithms as well, just need to adjust keyspace with result from --keyspace

[d2]

Thanks evilmog. One question

Code:

# Enter in the GPU count per node seperated by spaces
gpunodes=(8 8 4 4 4 4 4 4);

As you said, you use gtx980 exclusively, but if I have 5x9tx970 and 5xgtx1080 in 2 computing nodes the array would look like:

Code:

gpunodes=(5 12);

?

I treat whole system as gtx970 power, and looking at benchmarks gtx1080 is about ~2.5 times faster than 970. So my second node would act like it was built with 12 gpus, right?

[atom]

Sounds about right