Intel Unveils Strategy for State-Of-the-Art Artificial Intelligence

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by pjreddie on Thursday November 24, 2016 @08:27PM (#13033851)

If Intel is smart they will invest heavily into library development. Researchers use whatever is fast and right now cuda is fast, not just because nvidia has the best GPUs (it's about even) but because the primitives in cuda are so much faster. Matrix multiplication on the same hardware is like 3x faster in cuda than opencl or competing libraries and using the neutral networking primitives is even faster. Intel needs to invest in good, low level libraries so researchers can hack on their platform, build new things, etc. Ultimately I think researchers drive what platform gets widely used since training takes so much longer than inference.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by deepnotderp on Thursday November 24, 2016 @11:20PM (#13034389)

No, researchers use whatever is fastest for development and flexible. Otherwise we'd all be using neon.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by auvi on Friday November 25, 2016 @05:21AM (#13035647)

i agree. development speed is more important. is there any benchmark that compares neon to other systems?

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by deepnotderp on Friday November 25, 2016 @02:55PM (#13039393)

IIRC, because of Winograd kernels, 3x3 convolutions are amazingly fast and almost all the nets now have switched to 3x3 convolutions.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by rdlecler1 on Friday November 25, 2016 @03:39AM (#13035127)

I bought both NVIDIA and AMD. AMD has a big GPU patent portfolio as well. AMD trades at about 1/5th NVIDIA and could be a takeover target for a bigger player trying to move into AI.

In my opinion NVIDIA needs a moat. CUDA is a good start but they need proprietary data that's hard to create. I thought they should buy Yahoo which would give them a large and unique data set that NVIDIA users could tie into through an API.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by visarga on Friday November 25, 2016 @06:53PM (#13040798)

Won't get anywhere with a dataset. You need dozens or hundreds of them, and they are usually created using Amazon Turk, so anyone with some money can create one. And they need to be commissioned by the AI researchers, because they know best what they want to study.

NVIDIA needs to become more efficient and build deep learning hardware instead of graphics cards that do faster computation than CPU. If only they could create a small form AI processor for robotics that could do vision at 500+ frames per second. Take a look at this video from 7 years ago which does 1Khz vision: https://youtu.be/-KxjVlaLBmk?t=158

We need that.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by rsp1984 on Friday November 25, 2016 @06:17AM (#13035895)

This. Intel has a lot of catchup to do there. NVidia's drivers and libs are simply on a different level. For example I've seen the same OpenGL code execute faster on an ancient Tegra 4 (mobile) platform than on an Intel Haswell with HD Graphics (which is supposed to beat the Tegra 4 by miles according to raw specs).

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by mikepurvis on Thursday November 24, 2016 @08:42PM (#13033897)

See OpenCV and TBB for examples (from Intel) of this strategy at work.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by astrodust on Thursday November 24, 2016 @07:01PM (#13033588)

It really depends on how the market shapes up. You might see one software company rise to dominance where they interoperate with different platforms interchangeably, or you might see a single hardware vendor come to dominate a number of much smaller software vendors.

NVidia is the horse to bet on now, but it remains to be seen how they can can expand beyond their current model of leaning heavily on their GPU technology.

As AI becomes more pervasive you'll want to have it integrated into smaller, more power conscious devices, such as would be the case in robotics. Does NVidia have a solution here that scales down? Clearly they're focused on scaling up, as that's where the money is today.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by nl on Thursday November 24, 2016 @11:38PM (#13034436)

Does NVidia have a solution here that scales down? Clearly they're focused on scaling up, as that's where the money is today.

Yes, they have invested heavily in this. Everytime they talk about their "car/auto segment" (which is all the time if you listen to their investor calls) it's mostly about scale-down.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by astrodust on Friday November 25, 2016 @12:21AM (#13034555)

Cars, especially the sort that will need to do self-driving, will have massive batteries in them. I'm talking about Roomba-sized devices where they can't power a full GPU worth of gear.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by nl on Friday November 25, 2016 @04:30AM (#13035391)

They may have large batteries if they are electric cars, but it won't be for the GPUs.

The NVidia PX2 platform[1] - which is what the Telsa self-driving features uses[2] - is available in a 10W config. I presume this isn't the full self-driving mode, but the Jetson TK1 can do full image tracking and recognition in less than 30W (and that is a 2 year old platform).

[1] http://www.nvidia.com/object/drive-px.html

[2] https://blogs.nvidia.com/blog/2016/10/20/tesla-motors-self-d...

[3] http://developer.download.nvidia.com/embedded/jetson/TK1/doc...

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by imtringued on Friday November 25, 2016 @04:01AM (#13035244)

Where did you get the nonsensical idea that they are going to use desktop GPUs in cars? The board they specifically designed for self driving cars merely needs a few watts for 500 GFLOPS. Why would this not be suitable for a "Roomba-sized device"?

https://m.f.ix.de/scale/geometry/1280/q50/imgs/18/1/6/4/4/6/...

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by astrodust on Friday November 25, 2016 @03:10PM (#13039485)

Google packs their cars to the gills with hardware, so just going based on what fully autonomous systems use today.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by Teever on Thursday November 24, 2016 @07:46PM (#13033743)

They don't necessarily have to scale down to get the benefits of AI in side smaller devices such as robotics. There's no reason why the robotics devices can't or won't be equipped with a variety of radios to facilitate communication.

The AI agent can easily be somewhere centralized where power isn't a big deal.

Then the robot stops being analogous to an animal with it's own brain and starts being something more like an appendage.

Or maybe they'll be like those magical mops in Fantasia.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by dontreact on Thursday November 24, 2016 @09:21PM (#13034015)

At least two good reasons that you would have things running locally: Latency sensitive applications Privacy sensitive applications

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by astrodust on Thursday November 24, 2016 @11:16PM (#13034382)

There could come a day when you have to sign in to Facebook to use your car, like if they get to harass you with advertising the trip's free.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by AJ007 on Friday November 25, 2016 @09:07AM (#13036781)

Optimistically your car, perhaps instead their car.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by eloff on Thursday November 24, 2016 @10:03PM (#13034142)

Oh god, imagine being in a self driving car that does it's computing in the cloud - and then you enter a tunnel.

Obviously cars are not really power or space limited and they can't rely in connectivity. But it's funny to think about anyway.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by Teever on Friday November 25, 2016 @09:03AM (#13036761)

I'm thinking more like warehouse or factory that has a central rack-server and all the robots in the factory are wirelessly connected to it.

People look at a humanoid robot and they see an entity that is similar to a human but made of silicone and steel instead of meat and bone. I think it's more accurate to compare the robots in this factory to the hands and eyes of a large disembodied rackmount brain.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by tnecniv on Thursday November 24, 2016 @07:48PM (#13033748)

There's the Jetson boards, which are ARM + an NVidia GPU wrapped into one small package.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by astrodust on Friday November 25, 2016 @03:12PM (#13039498)

That's an interesting device: http://www.nvidia.ca/object/jetson-tk1-embedded-dev-kit.html

Also can't remember the last time I saw a USB port jammed on sideways.

[ Parent | on HN ]

Re:Intel Unveils Strategy for State-Of-the-Art Artificial In... by empath75 on Thursday November 24, 2016 @10:59PM (#13034341)

I think there are probably efficiencies to be gained by making chips specifically designed for tensor flow rather than repurposing gpus.

[ Parent | on HN ]