Artificial Intelligence and Dispersed Computing – #2

How is the world of processors changing, and what does it mean for the future of AI? Daniel Bogdanoff and Brig Asay sit down to talk about it.

Video (YouTube):

Discussion overview:

Intro 00:00
Intel acquisition of Altera 00:45
What does it mean for Intel to buy an FPGA company?
What is “dispersed computing” 1:17
Microprocessors used to handle everything
Then, GPUs became integrated 1:45
Offloading computing from a microprocessor 2:02
One option is to use an FPGA to share computing 2:10
ASIC vs FPGA 2:15
ASICs aren’t flexible 2:45
FPGAs give more flexibility than an ASIC 3:03
We use both FPGAs and ASICs in our instruments 3:25

Parallel vs serial buses 3:35

PCIe is x16, other tech going well past 2 and 4 lanes 4:00
This is helpful, but it adds a lot of design complexity
We’re starting to see 5:00
PCIe, USB, SerDes used to dominate but now we’re seeing some other technologies
like Generation Z and CCIX (Cache Coherent Interconnect for Accelerators) 6:00
Makes designs faster to market and easier to debug
Generation Z (Gen-Z) 6:25
Generation Z and CCIX build on PCIe technology
Why are these technologies coming out? 7:00
PCIe takes a lot of work to implement 7:35
So these technologies are less stringent 8:00
and are more open 8:15
We see a lot of PCIe Gen 2 that will start to be replaced by Gen-Z or CCIX type buses internally 8:30
How does the microprocessor connect to other chips in the design? 9:05
That’s the biggest opportunity for speed increases
Thunderbolt has been around for a while 9:45
But, Thunderbolt is finally taking off 10:00
It used to be an internal bus, but now we’re starting to see it externally on consumer devices

What are the next major tasks that will be offloaded? 10:25
AI, machines learning from themselves 11:00
“If true artificial intelligence happens, there’s no way a microprocessor can do it all” 11:10

Big data is huge, and that requires a lot of processing and computing 11:32
A processor and a server won’t be able to do it alone 11:50
Is this because there’s too much data? (it’s two-fold) 12:10
1. There’s tons of data 12:45
2. We want to know the answer right away
FPGAs/ASICs are currently doing a “filtering” of data which then feeds into a central processor 13:15
Right now, FPGAs are handling very specific tasks 13:56
Intel acquires Altera, which is a good indicator of where the industry is going 14:15
The FPGA is going to get smarter and smarter 14:50
Are FPGAs too slow? 15:15
What do designers need their FPGAs to do? 16:10
Companies creating FPGAs know that they have to have higher performance at lower cost 16:30
NVIDIA, Google, Facebook are all releasing their own chips
FPGA part costs will likely drop in the next 5 years as a result 17:20
Is there a blend of FPGAs and ASICs? 18:00
We’re seeing FPGAs starting to be implemented on data centers and servers 18:15
Using FPGAs instead of ASICs there for their flexibility

Servers lead the PC/consumer market in technology 18:45

Server loads are an order of magnitude greater than PC loads
Hyperscaling 19:40
Historically, you had storage, servers, and routers all separate. Now, they’re getting smarter with resource allocation

Localized vs remote dispersed computing 21:20

All the data has to go somewhere, there’s not a lot of point to point
Latency is becoming more of an issue 22:00
Is processor technology plateauing? 22:30
Consumers generally don’t need a lot more processing power as of today, but servers do
Are multiple core processors a harbinger of FPGAs taking on more tasks? 23:55

AI is becoming more and more important 25:05
There’s nothing more debated than artificial intelligence 25:40
We’re using it in a minimalist way 26:00
A “large tech company” had an AI go on Twitter and it didn’t work out very well 26:35
What is it going to take to make AI something that is integral to our daily life? 27:05
For data centers, AI is going to play a role in adjusting to the flux of data 27:50
What’s the difference between artificial intelligence and analytics 28:15

AI makes the decisions, analytics is just a flow of information
The 2016 USA presidential election is an good example of analytics vs AI 28:53
AI has been in Science Fiction for a long time 29:55
AI brings a lot of ethical discussions, but we don’t have time to talk about them 30:00
Predictions (Luddites, elections, and “the common man”) 30:45
AI and self driving cars 31:30

Leave a Reply

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

You are commenting using your 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 )

Google+ photo

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

Connecting to %s