Deep Learning & CUDA Benchmarks On The GeForce GTX 1080 Under Linux

[F1esDgSdUTYpm0iy]

1000$ to over 4800$ of unjustified expense right there.

Unjustified to you, maybe. Likewise, others might feel the same way about cars, vacations, hobby expenses or... computer hardware.

[starshipeleven]

Unjustified to you, maybe. Likewise, others might feel the same way about cars, vacations, hobby expenses or... computer hardware.

unjustified = has no practical difference apart price.
It's not a matter of liking but of price vs looks/performance/whatever.

a 5000$ bike isn't going to be terribly better than a 500$ one, unless we are talking of serious sports hardware used by athletes.

[Michael]

Is there any benchmarks for OpenCL vs. CUDA on NVIDIA hardware?

http://www.phoronix.com/scan.php?page=news_item&px=GTX-1080-OpenCL-vs-CUDA

[F1esDgSdUTYpm0iy]

a 5000$ bike isn't going to be terribly better than a 500$ one, unless we are talking of serious sports hardware used by athletes.

Then you obviously know very little about cycling. I, on the other hand, do. At one point I was even considering becoming a pro, although life got in the way of that one in a horrible way. Trust me, even non-professionally, non-competitively, there is a vast difference between a 500 dollar bike and 5000 dollar bike.

There is less of a vast difference between, say, a 1500 dollar one and a 5000 dollar one but a difference nonetheless that, to even non-professional enthusiasts can be worth the expense. For one, the constituant components out of which the bike was constructed are typically (far) more robust in the more expensive bike; consequently able to handle a pounding a lot better. Sure, for getting around town it's pointless to go over a couple of hundred bucks tops but if you want to make it a point to pick up cycling as a means of staying fit, even non-professionally then you really need to consider spending a bit more. Otherwise you'd just end up having to replace the bike or having it serviced (at yet another expense) on a regular basis.

Same goes for basically any other expense; a GTX 1080 isn't exceedingly much better than a 1070 but to those that make use of the difference, the expense can be well worth it.

[vocoloco]

I posted it in the other article's thread, but here's a couple of links that could be used to produce other DL framework benchmarks without too much work: https://github.com/soumith/convnet-benchmarks and https://github.com/glample/rnn-benchmarks.

[Ardje]

unjustified = has no practical difference apart price.
It's not a matter of liking but of price vs looks/performance/whatever.

a 5000$ bike isn't going to be terribly better than a 500$ one, unless we are talking of serious sports hardware used by athletes.

You obviously haven't driven a quest... A quest allows an untrained slightly obese person to bike at speeds > 30km/h with ease, with an action radius > 50km.
Well, that even isn't $5000 (image search quest recumbent trike), but try an image search for velotilt. If that lives up to its promises, it's > 45km/h for the untrained.

[name99]

Unjustified to you, maybe. Likewise, others might feel the same way about cars, vacations, hobby expenses or... computer hardware.

It's the internet. Where people feel it's acceptable to tell perfect strangers exactly what hardware and software they should or should not want, in spite of knowing utterly nothing about that person's budget, life history, personal priorities, or anything else...

YOUR bike is an expensive frivolous luxury, but MY gaming set-up is an example of buying the best, and not settling for anything less, because I'm worth it.

[name99]

I posted it in the other article's thread, but here's a couple of links that could be used to produce other DL framework benchmarks without too much work: https://github.com/soumith/convnet-benchmarks and https://github.com/glample/rnn-benchmarks.

How much are these abstracted to use standard BLAS, FFT, and what Apple calls BNNS (basic neural network convolution) calls?
My point is --- how feasible would it be to modify these slightly to use the various Apple calls and run them on an iPhone?
I ask not because I expect an iPhone to beat a 250W nV card, but because the magnitude of the gap (and its consequences) would be very interesting to know. Eg Apple has demoed real-time facial expression recognition but, as far as I know, they don't yet claim real-time object recognition, and it would be interesting to know how far away we are from that possibility.

[starshipeleven]

You obviously haven't driven a quest...

I've been on trycicles when I was little, that's just a bigger specimen. I understand it handles better (being a trycicle it's expected).
Not a bike. Screams "my driver is too fat to use a bike".

YOUR bike is an expensive frivolous luxury, but MY gaming set-up is an example of buying the best, and not settling for anything less, because I'm worth it.

Bicycles have more limitations than gaming rigs, invalid example.

With a 1000 gaming rig you play at resolution X

With a 2000 gaming rig you play at resolution Y

and so on.

With a bike you hit hard limits on performance well before the price goes above 1000$, unless you are an athlete and that's your job so any inch counts.

So going beyond is like tuning your PC case with lights and stuff. Pointless for gaming.