bit-tech.net

Intel to ship first FPGA Xeons early 2016

Intel to ship first FPGA Xeons early 2016

Intel has confirmed that the first Xeon to feature an embedded Altera FPGA will be hitting its largest customers in the first quarter of 2016.

Intel has announced that the first of its chips to feature field-programmable gate array (FPGA) hardware will be hitting servers in the first quarter of 2016, in limited quantities.

An FPGA is a middle-ground between a general-purpose processor which may not be terribly efficient at any given task and an application-specific integrated circuit (ASIC) which is extremely efficient at one particular task but cannot do anything else. Frequently used to test designs prior to building an ASIC and popular for accelerating particular workloads, FPGAs are typically dedicated chips - though versions embedded into system-on-chip (SoC) designs, typically using ARM CPU cores, are popular in embedded development. Unlike an ASIC, which is fixed at the point of manufacturing, an FPGA can be reprogrammed at any time: today it's accelerating SHA-512 hashing, tomorrow it's acting as an Ethernet MAC.

It's this flexibility which convinced Intel to integrate FPGAs with selected processors, starting with its Xeon family, and it's a project the company is putting serious money behind: back in June it acquired FPGA specialist Altera in a $16.7 billion deal. Now, that acquisition is bearing fruit with the announcement that the first FPGA-equipped Xeon parts are to roll off the production line early next year.

According to Intel's Diane Bryant, speaking at the Structure conference attended by IDG, the first FPGA Xeon will be shipped to selected customers - the largest cloud service providers, Bryant explained - in the first quarter of 2016. This, however, does not represent general availability: the units shipped will be for the company's biggest customers to 'begin tuning their algorithms' ahead of a proper roll-out at an undetermined future date, Bryant claimed.

Typically, features found in the company's Xeon chips eventually trickle down to its mainstream Core processor family. When this might happen with Altera's FPGA technology, though, remains to be seen: while having FPGA hardware could be of benefit to desktop users, the overwhelming majority will lack the knowledge to make use of it directly - meaning it would be up to either hardware companies or software developers to find a way to make use of the FPGA outside the data centre.

Pricing and specifications for the FPGA Xeon range has not yet been confirmed.

8 Comments

Discuss in the forums Reply
Alecto 19th November 2015, 14:07 Quote
'being tuning their algorithms'

Spell-checker much?
Gareth Halfacree 19th November 2015, 14:14 Quote
Quote:
Originally Posted by Alecto
'being tuning their algorithms'
Spell-checker much?
Fun fact: not a single one of those words is spelt incorrectly, and thus the error ("being" instead of "begin") would not be picked up by a spell-checker. (Also, you should have written "spell-check much," as "spell-checker much" makes no sense.)
Pookie 19th November 2015, 15:10 Quote
Quote:
Originally Posted by Alecto
'being tuning their algorithms'

Spell-checker much?

Comments like that don't impress a single person on here
wuyanxu 19th November 2015, 15:55 Quote
How does this differ from an existing Altera or Xilinx FPGA plugged into PCI express?

Sounds like it's up to the end users to write their own stuff, which is no different to above setup.
Gareth Halfacree 19th November 2015, 16:00 Quote
Quote:
Originally Posted by wuyanxu
How does this differ from an existing Altera or Xilinx FPGA plugged into PCI express? Sounds like it's up to the end users to write their own stuff, which is no different to above setup.
Much, much, *much* faster communication with the CPU - and that's assuming Intel doesn't allow the FPGA to share cache with the CPU, which makes things even quicker. Massive difference betwix that and an add-in card.
chriscase 19th November 2015, 17:41 Quote
Quote:
Originally Posted by Pookie
Comments like that don't impress a single person on here

I beg to differ! It was impressive, although perhaps not in the way the poster intended.
wuyanxu 20th November 2015, 07:38 Quote
Quote:
Originally Posted by Gareth Halfacree
Much, much, *much* faster communication with the CPU - and that's assuming Intel doesn't allow the FPGA to share cache with the CPU, which makes things even quicker. Massive difference betwix that and an add-in card.
ah, okay. So Intel is selling Intel CPU with a re-programmable co-processor. OR selling a FPGA with an Intel core.

May be further down the line, in addition to Nios and ARM cores, we can get a barebone Intel IP core.
Bindibadgi 21st November 2015, 07:14 Quote
Quote:
Originally Posted by wuyanxu
How does this differ from an existing Altera or Xilinx FPGA plugged into PCI express?

Sounds like it's up to the end users to write their own stuff, which is no different to above setup.

PCI-Express latency overhead is actually pretty high and has limited bandwidth compared to alternatives like NVLink/Omnipath etc.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums