Tested MapD with mix of 1060 and P106 mining card


Hi there, just to let u know i tested with my rig using 3x MSI 1060 6GB and 1x MSI P106 mining card. on old Dell T5500 PCIe Gen2, CPU X5650.

It’s part of my school thesis, i did some benchmarks with queries on different sets of 70M, 300M, 600M, 800M and 1B records, on CPU mode and gpu mode.

That’s awesome the mining card can do SQL processing well. Big thanks to MapD.


nice info, but for CPU testing with a small number of records you should change the fragment size of your tables to make sure MAPD is able to use all 6 cores of your CPU


as you can see the from your graph the query on a dataset of 70-150 are slower than the one on 300 because the software is able to use more or less 2 core on the first one and 4 on second one, leading at a subpar result.

changing the fragment size to 10M finstead the default 32M would help


Thanks, i think you’re right, checking back the screen record i found it does not utilize all the cores. i’ll try and let u know the result.


yap in mapd each fragment is assigned to a core, so to get a more accurate (and faster) response time in CPU mode you should organize your table with a number of fragments multiple of number of cores the cpu you have in the system /in this case 6)


Many thanks Aznable, you saved me a shoot in the head.
i could not find the alter table to change the fragment, i guess i have to recreate the table and put back data.


@pta84vn thank you so much for the performance report, and thanks @aznable for helping get even more performance!


A shot? Is you teacher so bad tempered? (kidding)