Sat 17 May 2008

RSS Feed

Edited by Paul Hales

Published by Incisive Media Investments Ltd.

Terms and Conditions of use.

To advertise in Europe e-mail here

To advertise in Asia email here.

To advertise in North America email here.

Join the INQbot Mail List for a weekly guide to our news stories:

Subscribe

Intel, Cray flesh out HPC plans

Chip shop analysis Shining Quickpath

CRAY'S decision to develop its future HPC products around Intel silicon could be the beginning of the end for AMD's performance ambitions.

Although Cray says its existing XT3, XT4 and XT5 ranges, which use a mix of AMD and home-grown silicon, are unaffected by the deal, the emergence of Cray's first Cascade products, billed for 2011, must put the fear of God into AMD. Its desktop offerings are increasingly being relegated to the low end of big manufacturers' ranges and to be consigned to the second division of HPC must hurt a great deal.

To date, Cray's development efforts have centred around building custom interconnects supporting multiple AMD chips through Hypertransport. But Intel's upcoming Quickpath, which arrives with Nehalem later this year, replaces the frontside bus with a new fast, wide pipe connecting multiple CPUs together.

Cascade will enable one machine to handle several HPC tasks concurrently, rather than throwing all its power at a single application.

The combination of Quickpath and Cray's own high-bandwidth memory subsystems should result in a significant leap forward in high performance computing. The companies reckon the new machines should be able to deliver an exaflop, or one quadrillion operations a second.

The official line from Cray and Intel is that systems with AMD technology will continue to be available, but describe their new partnership to be 'comprehensive'. That has an ominous sounding ring for AMD, as does Intel saying it intends to 'meld innovations' around Cray’s interconnects and Intel’s future processor technology.

Raise the Itanic
Intel added that while most of its current volume in HPC is Xeon based, the Itanium remains important for a number of applications such as where a very large shared memory subsystem is required for applications such as weather and climate modelling.

The companies also say that they will be making 'substantial' investments in terms of both money and collaboration between engineers on 'a variety of technologies', although, as usual, refused to be drawn on details of financial arrangements or specific future products.

One detail that did emerge was that Cray is seeking US Government approval to incorporate Intel technology into its HPC initiative, which receives significant Government funding.

The benefits for both companies are clear: Cray gets both improved technology and removes its exposure to using a single CPU vendor that's looking increasingly-shaky. Intel moves onward and upward in the highly-lucrative world of HPC, which IDC estimates should be worth in excess of $18 billion by 2012. µ

Comments

Peta,not exa

I guess you rather meant petaflop,not exaflop.Petaflop amounts to 10^15,or quadrillion flops,while exaflop is equal to 10^18-quintillion flops,aka 3 entire orders of magnitude greater.
posted by : Sharpedon, 29 April 2008

Glad, You Mentioned weather.

Some of neat stuff mentioned today n' theINQ. 256 bit width. Why? idea is to prepare as long of bit string as possible so most reliable changes can be effected within chip enviorment in each pass.(Then bringing Pass Rate to Full).
Not much faster at 256 bits due to less end & start bits, just smoother & more stable. Basicly somewhere machine must be 2X faster to prepare 2x wider bit string, if same number of passes is made Unit/Time. Less stop & go confusions.

RISC was memtioned in another article as improvement. RISC is Reduced Instruction Set Code. Basicly you have todays most advanced instruction set, up from yesterdays. Somehow in that interaction there might be sliming or quality increasing. Also adding alpha instructions means reducing what is needed to make newbie instructions feel comfortable, More able to document effect of change.
Every Time CRAY Changes Hardware NEW RISC Is In Order. Then whole mess can be repasted back togerher with more advanced instruction set final outcome. RISC is NOT Better in itsrlf, RISC is Hard.
Once Brains have invented fundumental improvement, those pathways are stuck, So Whole NeW Fresh Team is in Order, To take that Next Leap. Good For CRAY.
See Moore, Get mor.
Drashek
posted by : Symoure_Ultie, 29 April 2008
IThound
Search for solutions, reports & analysis

Newsletter signup