Live blog of Intel Datacenter Summit

https://www.anandtech.com/show/13192/intel-datacenter-day-20… Lots of implications for our many of our businesses.

they see 3 growth drivers.
cloud, network and AI
autonomous cars combine all those into one product.
cars will need 10x intelligence of today
4 Terabytes of data per hour generated
car can only have 30-40km of data on it right now

volume of Cloud processors increased from 18% to 50% 2013-2017

Ai 2.5 billion TAM, 30% CAGR expect 10billion by 2022
highlights need to move data with AI (think pure) and networking (arista) …obviously intel would like you to think of them. They have optane which they are pushing hard. My understanding is nobody has really found a use case for it. Maybe AI in the future. They are shipping Optane to google.
New partnership with google for cloud.
starting to tailor processors to AI with int8 etc instructions
brought out jim keller (semiconductor design superstar) he is responsible for AMDs original athlon which demolished intel at the time. Then went to apple and got them started on processor design for their incredible mobile SOCs. Then went back to AMD and designed Zen which has made AMD competitive again. Then went to Telsa and designed their AI chip, now at intel.

calling their new strategy a data centric approach
processing - cpus tailored to AI
transport - networking 5g etc
storage - optane next gen, faster than SSD

announced intel nervana NNP L-1000. processor for AI for 2019 (
developing customer processors for google and amazon
focusing on the software stack (like nvida)

they address their “dark” times 2013-2017. They are making the point they have invested in cloud and ai and are making a pretty strong statement that Intel is back.

comment TAM CAGR is accelerating 30+%

my comments
So far intel has totally ceded the AI, data market to nvidia. I think it is pretty clear that intel is going to be competition in the future. They have been putting the people and resources in place of the last couple of years. They have some advantages over nvidia in that they can control much more of the technology. I wouldn’t count intel out but I’m firmly in the camp that they need to show me before I worry much. As far as I could see they are only comparing benchmarks against themselves which makes me think they aren’t competitive with what nvidia has to offer. Of course they can do a 11x speed up over their intel processors. Nvidia can show 100x over cpu. Anywho, the link is a good read for anyone invested in these companies.

9 Likes

Nice write up.

Jim Keller is certainly a nice feather to have in your cap. This bodes well for Intel. Personally I am expecting someone to up-end the GPU as the engine of choice for AI, just not sure who yet. Intel is on my list of possibilities here.

As per cars, Wayve just showed a video of using reinforcement learning to teach a car to lane follow in under 20 minutes, with a single GPU. https://www.cnet.com/roadshow/news/an-ai-learnt-to-drive-an-… This just further bolsters my thought that either new hardware or new approaches to learning will lead to less dependance on lots of GPUs. (This will be a good thing, by the way).

Bill Jurasz

1 Like