Algorithm for the spread sheet calculator ?

Hello,

The spreadsheet calculator… I think it was called “occupancy calculator” or something like that… what is it’s algorithm ? Does it use a solver ?

I would definetly be interested in it’s algorithm… perhaps then I and others could code it into their applications for automatic calculations of optimal kernel execution configurations…

Bye,
Skybuck.

Hello,

The spreadsheet calculator… I think it was called “occupancy calculator” or something like that… what is it’s algorithm ? Does it use a solver ?

I would definetly be interested in it’s algorithm… perhaps then I and others could code it into their applications for automatic calculations of optimal kernel execution configurations…

Bye,
Skybuck.

Therez no specific algorithm. It is a collection of facts about NVIDIA hardware. You can examine the logic of the spreadsheet if its not hidden or protected. That should give you the clue…
I think there are some XL to C++ converters available. Check out.

Therez no specific algorithm. It is a collection of facts about NVIDIA hardware. You can examine the logic of the spreadsheet if its not hidden or protected. That should give you the clue…
I think there are some XL to C++ converters available. Check out.

The Occupancy calculator does not solve for the optimal number of threads per block. As Sarnath stated, it’s just a collection of facts about hardware of different CUDA capabilities that you can also find in the Programming Guide, and calculates the occupancy for each of them. Check it out yourself, the formulas in it are not hidden or protected in any way.

As occupancy is not directly related to performance, there would also be no point in optimizing for it. The usual use case is to use the Occupancy calculator to roughly see where you stand, and then try a few sensible configurations (i.e., multiples of 32 threads) to find the one that is performing best.

The Occupancy calculator does not solve for the optimal number of threads per block. As Sarnath stated, it’s just a collection of facts about hardware of different CUDA capabilities that you can also find in the Programming Guide, and calculates the occupancy for each of them. Check it out yourself, the formulas in it are not hidden or protected in any way.

As occupancy is not directly related to performance, there would also be no point in optimizing for it. The usual use case is to use the Occupancy calculator to roughly see where you stand, and then try a few sensible configurations (i.e., multiples of 32 threads) to find the one that is performing best.

Microsoft has file viewers available for example:

Word viewer.

They also appear to have an excell viewer:

http://www.microsoft.com/downloads/en/details.aspx?familyid=1cd6acf9-ce06-4e1c-8dcf-f33f669dbc3a

However it’s pretty useless since it doesn’t allow me to change the fields.

Hmm so much for that ;)

Microsoft has file viewers available for example:

Word viewer.

They also appear to have an excell viewer:

http://www.microsoft.com/downloads/en/details.aspx?familyid=1cd6acf9-ce06-4e1c-8dcf-f33f669dbc3a

However it’s pretty useless since it doesn’t allow me to change the fields.

Hmm so much for that ;)

Ok,

I found something that seems to work:

Google Docs.

Simply sign into google, upload the spreadsheet, set it to convert automatically when uploading.

And then simply edit it…

It’s a bit slow but it seems to work ! ;) =D

Pretty cool ! ;) =D

Ok,

I found something that seems to work:

Google Docs.

Simply sign into google, upload the spreadsheet, set it to convert automatically when uploading.

And then simply edit it…

It’s a bit slow but it seems to work ! ;) =D

Pretty cool ! ;) =D

OpenOffice works as well - I don’t use any Microsoft software either (apart from CBMbasic V2 :smile: ).

OpenOffice works as well - I don’t use any Microsoft software either (apart from CBMbasic V2 :smile: ).