Intel unveils new family of AI chips to take on Nvidia’s GPUs - Previously known as Lakes Crest

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Nothingness

Platinum Member
Jul 3, 2013
2,410
745
136
Of course not, AI is rapidly growing and they've committed to a long term roadmap.
By killing I meant failing at making it a success. Intel has destroyed many companies it has bought in the past. I know many Infineon/Intel ex-employees across Europe and what they have to say about Intel isn't pretty.

Of course that doesn't imply they will fail with Nervana. It's just a hint of the issues they have with their culture.
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
974
146
Intel went from "Let's try IA* for everything" to "Let's try everything for AI".


*Intel Architecture (x86), which Intel tried to stick into GPU's, phones, etc.
I'm glad they removed the IA and X86 part and realised that won't do.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Intel went from "Let's try IA* for everything" to "Let's try everything for AI".

*Intel Architecture (x86), which Intel tried to stick into GPU's, phones, etc.

Let's see that again.

TI's cable modem division: Used to use ARM cores, now use Atom-based cores

ME engine for the Lewisburg chipset on the Purley Xeon platform: x86 Quark cores

Nervana IP: Lake Crest, moving to Knights Crest. The latter is described as a Bootable Intel Xeon CPU. Sounds like Xeon Phi, but instead of general purpose AVX-512, it adds NN specific accelerators

Altera FPGA: Later versions of Xeon Scalable "Purley" will come with FPGA's on package

I wouldn't be surprised if the FPGA portion moves to Atom cores. The thing is, it takes time to port to your own technology. There was an article that said Intel happens to be the largest licensee by volume for ARM cores. That's because every Ethernet product from Intel uses tiny ARM cores, and DSP needed by their chipsets likely use tiny ARM cores too. The leaks for 300-series CNL PCH say it'll use "Quad Core Audio DSP". The generic naming suggests a tiny ARM core. Their SSD controllers probably use an ARM core.

If the compute requirements rise, its likely they'll move to their own IP.
 
Last edited:

Dayman1225

Golden Member
Aug 14, 2017
1,152
974
146
DQYjx8TUQAALM-A.jpg
DQYjx8zVQAAmjn5.jpg


Now intel is showing off it’s Nervana chip publicly at their booth. Last night was the reveal at their private party. Naveen himself is here fielding questions.

Flexpoint: numerical innovation underlying the Intel® Nervana™ Neural Network Processor

Intel Nervana™ Neural Network Processor: Architecture Update
 

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
No idea when that update is coming though, I imagine sometime in H2 2019 - Early 2020.
I mean, the jump from 28HKMG to whatever Intel calls it's 10nm node will be huge, but this will be one hell of a long journey.
Well unless current 28nm product surprises everyone with it's outstanding performance (sometimes that happens).
 
  • Like
Reactions: Dayman1225
Mar 10, 2006
11,715
2,012
126
I mean, the jump from 28HKMG to whatever Intel calls it's 10nm node will be huge, but this will be one hell of a long journey.
Well unless current 28nm product surprises everyone with it's outstanding performance (sometimes that happens).

The current 28nm product isn't meant to sell, it's meant basically as a software development/ecosystem building tool as well as proof-of-concept.

Nobody is going to adopt these chips in serious volumes for another few generations, and that's assuming that the products live up to all they're cracked up to be.
 

NTMBK

Lifer
Nov 14, 2011
10,237
5,019
136
The current 28nm product isn't meant to sell, it's meant basically as a software development/ecosystem building tool as well as proof-of-concept.

Nobody is going to adopt these chips in serious volumes for another few generations, and that's assuming that the products live up to all they're cracked up to be.

Ah yes, the same strategy that went so well for Xeon Phi.