And you are here why?

Database technology for transaction processing and that for analytics have, kinda, moved in different directions. This has largely been from a “separation of concerns” perspective . Transactions are all about velocity, with people speeding up databases by making them in-memory (e.g. HANA), and/or adding columnar capabilities (e.g. Postgres, Oracle, and, well, everybody). Analytics, OTOH, are usually about volume — having   massively parallel   or distributed back-ends that can run through gobs of data very fast (e.g. Vertica, Redshift, Netezza, etc.) is pretty much table-stakes in this world. Mind you, when one talks about   massively parallel , GPUs should come to mind, and it clearly did for   Kinetica ,   SQream , and   MapD   who are all using GPUs to run the parallelized parts of SQL queries in their analytics databases. The result, as you might imagine, is a   huge   speed up in query processing for analytics. MapD, in partic...

RiseML benchmarked Google’s TPUv2 , and — surprise! — they’re fast. Also — surprise! — they’re cost efficient too. I know, shocking, right? After all, if you consider selection-bias,   of course   they wouldn’t release something that wasn’t fast, and   of course   they’d price it so that it would be cost efficient. Then again, this wouldn’t be the first time a company has done something stupid, so, hey, we should probably keep that in mind   . Anyhow, it turns out that the TPUv2 is around 7.3x faster than an Nvidia P100, and around 2.8x faster than an Nvidia V100. Yes, it’s more expensive, but you actually end up saving on a   $ per hour   basis — it’s that much faster. It’s important to note that the TPUv2 is   not   just a speeded up version of the TPUv1 — it’s quite a different beast. Unlike the TPUv1, which was focused on inference, the TPUv2 is actually focused on speeding up learning. Inside, it has systolic arrays (•) which do ...

Shiv Manne ran a whole bunch of tests against GPU instances from platform providers like AWS, GCE, Paperspace, Softlayer, etc., and   the results were…interesting . Not surprising mind you, just   interesting . The lede — high end GPUs get very expensive very fast. Yes, they train wicked fast, but the ROI, frankly, sux. In fact, the best deal seems to be to run lower-end single-GPU instances (more info on this   here ). Why single-GPU? Largely because,   right now , support for multi-GPU either doesn’t work, or when it does, seems to be surprisingly inefficient. Go figure! Paperspace   seems to be on the game here, both for costs, and for performance. In fact, if you’re just testing stuff, they’re pretty much your best bet, as seems to be   validated elsewhere   too. And, of course, the obvious remains — AWS is expensive, and GCE is not   . The results are below, but go take a look at   the whole post . Note : There is also  ...