DARPA investing $80 million to develop HIVE processors

[WMGroomAK]

As is probably common knowledge by now, the US Defense Advanced Research Projects Agency (DARPA) tends to spend quite a bit of money on advanced technology, some of which makes it into the world, some of which proves to be unworkable for the moment.  In their most recent $80 million US project, they are looking to create a different kind of computer processor that would be the 'World's First' graph analytic processor.  So of course, the question is, what does this mean?  According to an Engadget article:

https://www.engadget.com/2017/06/11/american-military-backs-an-entirely-new-kind-of-processor/

Quote

Virtually every processor you see is based on the same basic (Von Neumann) computing model: they're designed to access large chunks of sequential data and fill their caches as often as possible. This isn't the quickest way to accomplish every task, however, and the American military wants to explore an entirely different kind of chip. DARPA is spending $80 million to fund the development of the world's first graph analytic processor. The HIVE (Hierarchical Identify Verify Exploit) accesses random, 8-byte data points from the system's global memory, crunching each of those points individually. That's a much faster approach for handling large data, which frequently involves many relationships between info sets. It's also extremely scalable, so you can use as many HIVE chips as you need to accomplish your goals.

Basically, these processors would be able to access data stored on the system memory in a more non-linear fashion than how current processors have to take in data and should allow for these processors to be able to detect cyber-attacks and disease outbreaks at a faster pace than existing processor technology.  I would also think that these processors/computing models would provide a greater advancement towards better Artificial Intelligence as the computers operating with these processors would be able to do something that our brains currently do, which is connecting dots in incomplete data sets or seeing patterns emerge out of data as it is being processed. Currently DARPA has Intel, Qualcomm, Pacific Northwest National Laboratory, Georgia Tech and Northrop Grumman on board to work on the development of these chips so it would appear that there is some horsepower behind the initiative, although I don't expect there to be a public computer running this for at least a decade, if it ever is fully developed.

 

DARPA News: http://www.darpa.mil/news-events/2017-06-02

[MrMcMuffinJr]

I hope this will eventually be modified for the consumer market.

[GDRRiley]

8 minutes ago, huilun02 said:

$80m gone

they are darpa that is a drop in the water for them. 

[Vode]

56 minutes ago, GDRRiley said:

they are darpa that is a drop in the water for them. 

lol it's probably like you spending $15 on an amusement ride.

 

Damn those 15 bucks sting...

[Teddy07]

I assume it can't be done with this low budget because other companies would have done it already.

[mach]

I would imagine this 80 mil is just for the very inital research to see if the concept is viable

The actual development is going to cost billions

[dizmo]

Why is it the more time passes, the more things start sounding like Resident Evil...next thing we know they'll house the processors underground...in The Hive...

 

1 hour ago, huilun02 said:

$80m gone

At 2.5% of their on the books yearly budget, I'm sure they couldn't care less.

[VerticalDiscussions]

29 minutes ago, dizmo said:

-snip-

Haha, yeah it definetly sounds like something straight out of Resident Evil! Glad someone else noticed .

[tlink]

this honestly doesn't sound like a bad idea to me. big data is, pun not intended, really big. a quicker way to process all that data would be revolutionary for AI development, deep learning and prediction algorithms.

[Sniperfox47]

I'll be interested to see whether HIVE or TrueNorth lead to better performance for deep learning applications.

 

I'm assuming HIVE can be used for training? Afaik TrueNorth is only really useful for inferencing your model afterwards.

[Thony]

Am I the only one thinking 80M is not much for this kind of a project ?

[SpaceGhostC2C]

4 hours ago, WMGroomAK said:

he HIVE (Hierarchical Identify Verify Exploit) accesses random, 8-byte data points from the system's global memory, crunching each of those points individually. That's a much faster approach for handling large data, which frequently involves many relationships between info sets.

Unfortunately, that didn't really explain anything to me.

Although looking a bit around, it seems to be focused on graph theory mathematical applications. I just didn't get how do  you specialize a processor's architecture for that...

[Curufinwe_wins]

15 minutes ago, SpaceGhostC2C said:

Unfortunately, that didn't really explain anything to me.

Although looking a bit around, it seems to be focused on graph theory mathematical applications. I just didn't get how do  you specialize a processor's architecture for that...

DL half/quarter precision ASICS are already basically doing this.... don't know that is so special here.