Question: does MapD automatically shard tables across multiple GPUs?


#1

Hi,

I just wanted to check my reading of the docs. As I understand it, if I’m using a machine which has multiple GPUs I have to manually shard my tables across the GPUs. MapD won’t automagically shard my tables across the GPUs.

Is that correct?


#2

Actually, scratch that. I hadn’t realised it’s only useful in multidimensional scenarios, joins, etc. (Unfortunately, I can’t delete the question. Doh.)


#3

anyway i guess you haven’t to shard anything to make the system use all the gpus on most scenarios


#4

Hi,

MapD does automagically fragment your work across all GPU’s.

When you shard a table you allow MapD to make even better choices when querying the data and consolidating the results, allowing better performance, less inter process communication and less heavy result reductions.

Regards