Data science benchmarking

I just wanted to see if this forum would be interested in data science benchmarking thread but from more of a custom builder standpoint than enterprise solution benchmarks?

If so then I’d be interested in coming up with some consensus of what we use to benchmark and what we measure. Again, potentially with more of a focus on features or metrics that might be more aligned with custom rigs running at some ones house or desk than COTS solutions running in a datacenter.

Is there any interest?

5 Likes

if your willing to go through with the methodology and a suit of tests that are platform agnostic i am willing to bet that there would be an interest.

I will gladly draft up an initial methodology. Hopefully, some others will jump in with their thoughts as well.

Fairly new to ML and this Data science thing and as such also very interested in benchmarking information to help inform decision making. I know I recently completed some tutorials using my 2060 Max-Q and saved the some of the stats from the results. Since running the tutorial was pretty easy and all of the parameters fairly reproducible, something like that could be used as a user friendly benchmark?

I personally care more about scientific computing applications than gaming ones, so this would be nice to see.

yes, running out the door, otherwise I would engage this topic more!@

Scientific compute benchmarks would be handy for reference, there are cases where a consumer GPU makes more sense on the cost to performance ratio than a Pro GPU but there are some Pro cards which have better precision compute.