News for hackers. Stuff that matters.
Re:Hard Tech is Back by Retric on Friday March 11, 2016 @11:48AM (#11267691)
You can go pretty far with AI using just a few GPU's.

10's of Teraflops is cheap.

Re:Hard Tech is Back by ethbro on Friday March 11, 2016 @12:53PM (#11268226)
Curious what YC could do from their perspective to leverage their semi-unique position here as well.

The big cloud vendors are probably more geared towards the infrastructure side, but given problems (a) that "hard tech" requires non-trivial capex, (b) that many startups fail (nature of the space), & (c) that YC sees a lot of startups, then maybe there are opportunities to more efficiently handle "hard tech" needs by time-sharing common resources.

There are always going to be bespoke needs, but things like shared vehicle fleets with standardized interfaces, etc seem plausible when you've got a stable of 5 startups, with each being perfectly fine only testing their autonomous UAV / vehicle platform Tuesdays and Thursdays (or once a month) and letting someone else use it the rest of the time.

Re:Hard Tech is Back by hobofan on Friday March 11, 2016 @12:43PM (#11268141)
Purchasing a Titan X for 1k is still quite painful if you do a startup with virtually no funding.

Source: I just did that.

Re:Hard Tech is Back by jkyle on Friday March 11, 2016 @01:18PM (#11268432)
You don't have to get a Titan X to get to work. People were doing good GPU work prior to 2015 and Titan's didn't even exist then.

You can get the 980 Ti w/ 6GB of ram for $600 and some change. And if that's too much, the 970 w/ 4GB can be found for under $400.

And if even that is too high, you can troll gamer forums for used cards and cut that cost in 1/2.

Re:Hard Tech is Back by hobofan on Friday March 11, 2016 @01:34PM (#11268557)
See my other comment.

You need to have a Titan X to produce comparable benchmarks, otherwise a lot of people won't take them seriously. Also the 6GB of a 980Ti already rule out quite a few popular networks we needed to run.

I was not really in a position to wait much when buying the card, and I didn't want to risk waiting a few weeks to get a used card I bought on a forum, which then comes out being just as expensive in living cost. You also as a startup don't generally have the luxury to risk waiting.

I don't regret the purchase, I more wanted to show the reality of a "hard tech" startup.

Re:Hard Tech is Back by mappu on Friday March 11, 2016 @07:07PM (#11270534)
>the 970 w/ 4GB

3.5GB usefully. The last 512MB performs significantly slower.

Re:Hard Tech is Back by rorykoehler on Friday March 11, 2016 @12:53PM (#11268228)
Why buy when you can rent?
Re:Hard Tech is Back by hobofan on Friday March 11, 2016 @01:07PM (#11268326)
You mean on AWS/Azure? Because I have to develop a framework (https://github.com/autumnai/leaf) with it, and there is no way I am going to introduce any more complexity than necessary that early.
Re:Hard Tech is Back by rorykoehler on Friday March 11, 2016 @01:17PM (#11268422)
Understood. Leaf looks interesting. Will be keeping an eye on it. Having just recently focused on machine learning I still haven't settled on a stack. I find your rational behind using Rust quite helpful. What kind of timeframe are we looking at for when this is production ready?
Re:Hard Tech is Back by hobofan on Friday March 11, 2016 @01:42PM (#11268613)
Really depends on what use case. There should be a nice production-ish demo out towards end of the month, but I am afraid I can't say anything less vague at the moment.
Re:Hard Tech is Back by tavert on Saturday March 12, 2016 @06:58AM (#11272496)
Does the rust core team have any plans towards being able to write kernels directly in Rust using the NVPTX LLVM backend? Otherwise if you're all just calling into CuDNN anyway, pretty narrow differentiation in performance (and big difference in development resources) between you, Torch, and TensorFlow.
Re:Hard Tech is Back by hyperbovine on Friday March 11, 2016 @12:13PM (#11267889)
Take it away, Koons Crooks.