On the relative merits of GPUS, or, CUDA crashed Steam

My laptop is not an elite machine. Nevertheless, it has a decent processor (Core I5, dual core, running at 2.8gHz) and a very decent (for mobile) GPU (GeForce 860M).

My media server has an older CPU, a quad-core AMD Phenom II running at 2.6gHz, and a brand new GeForce 1060.

So, how do they compare?

First, it turns out that you MUST close Steam out before running a heavy workload. About two minutes in I was assaulted with the loudest fan noise (though, it should be noted, no alarms) I’d ever heard, and I had to reboot the system to get it back to a stable state.

Now, then, my largest Neural Network is five layers deep and 5,000 nodes wide. On my laptop, it would take around 56 seconds to train one epoch on a dataset of 100,000 items.

The GTX1060?

26 seconds.

I just cut my run time in half.

True, I could have gotten around an extra 40% performance by going for the GTX1070, but that would cost literally twice as much (and the GTX1080 even more on top of that). Also, as it has more memory, I should be able to load larger models in the future.

I’m quite pleased with it. Oh, and I can play The Witcher 3 with all the settings on maximum. Take that, PS4.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s