B
B
beduin012020-10-12 12:18:49
Video cards
beduin01, 2020-10-12 12:18:49

Is GPU parsing possible?

There is a parser which processes millions of records. The drawdown is 80% for the processor. The rest I think so IO. Is it possible to process text information on the GPU. I would like to have several hundred or even thousands of cores.

Are similar tasks generally solved on the GPU?

Answer the question

In order to leave comments, you need to log in

6 answer(s)
D
Dmitry Roo, 2020-10-12
@xez

Definitely possible.
For example: https://github.com/antonmks/nvParse

X
xmoonlight, 2020-10-12
@xmoonlight

Start from here .

The drawdown goes to 80% for the processor
read data in large blocks and process in asynchronous multi-threaded mode.

S
Sergey Gornostaev, 2020-10-12
@sergey-gornostaev

First you need to find what exactly consumes the processor and find out if it lends itself to vectorization. GPU is not a magic wand, not everything can.

A
Armenian Radio, 2020-10-12
@gbg

Very task dependent, probably not, because parsing processes data one-dimensionally and backtracks often. In addition, this is a task with low data locality, which means that a lot of time will be spent on transfers.
The GPU is fast when you need to grind a lot of data with a minimum of branches. Parsing is the opposite of this in terms of the structure of the algorithm.

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question