Intel adds 22nm octo-core 'Haswell' to CPU design roadmap
tock follows tick follows tock etc.
IDF Intel's 'Tick, Tock' model of chip development is now well established and mapped out through the transition from 45nm to 32nm. Now we know what's happening at the 22nm.
The 22nm era, for Intel, runs through 2011 and 2012. The first of those years will see the arrival of 'Ivy Bridge', the 22nm die-shrink of the brand new, 32nm microarchitecture the chip giant will introduce in 2010, 'Sandy Bridge', the chip formerly codenamed 'Gesher' - the codename changed when the chip company discovered that Gesher is an Israeli political organisation. Whoops.
At some point in 2012 - probably late in the year, if we follow the example of Intel's upcoming tock, the 45nm 'Nehalem' architecture - the chip giant will release 'Haswell', its first native 22nm microarchitecture.
Expect Intel to say more at its bi-annual Developers Forum, due to kick off in San Francisco later today.
As a recap, Intel has already revealed that Nehalem will be followed in 2009 by its 32nm die-shrink, 'Westmere'. Westmere, codename fans, was originally called 'Nehalem-C'.
Back to Haswell, which will apparently be a native octo-core design with a new approach to on-die data and instruction caching. It will also allow for in-package co-processor cores. Programmers, it will add at least one new instruction: FMA, Fused Multiply Add, which allows three values to be added and multiplied in a single operation, for example A = ±B * C ± D.
FMA is part of IEEE 754r, a revised version of the industry standard algorithm for floating-point maths, currently expected to be formalised - it's still in draft form - by the end of the year and long before Haswell ships.
64 bit Media Center OK.
Just got MSN tip to download msn olympics player for ultimate media center. works ok & has special 64 bit edition. that with 64 bit silverlight should make 64 bit practical, beyond FLASH problems.maybe adobe gets better pitch from microsoft? anyway low gb memory & few remaining IE player probs seem likely to quickly be fading from ultimate 64, so bring on CPU & other good stuff. aHSO, Hyper Visor has been describe as E-Z & straightforward, so it should all click together.
Signed:PHYSICIAN THOMAS STEWART VON DRASHEK M.D.
The end is not *quite* nigh
If a nano meter is 1x10^-9 meters and the physical limits of quantum mechanics is 1x10^-43 meters, I'd say we've got a way to go. However these dimensions are at least 1x10^20 times smaller than electrons. (If you didn't know, at 1x10^-43 matter becomes everywhere at once. Ask a quantum physicist.)
Still... There are limits to how small we can get and still maintain structural integrity with those high-speed electrons slamming into the walls of the processor pathways.
Don't know when we'll "hit the limit", and be forced to measure processor performance in a completely new way. We're obviously already moving past clock speed and adopting die-size. What's the next point? Running the "number-of-simultaneous-threads-in-a-single-instruction-set" root of the FSB multiplied by interconnect length divided by pipeline width?
AMD is going next to 6 core discrete and 12core dual 6 core
HI, as usual Intel is behind is the innovative department...
The end is nigh!
22nm in three years. That's crazy!
You guys all know that the END comes at 16nm, right? No molecular computer could be any smaller...
FMA? In 2012?
It's pretty pathetic that they've waited this long to add FMA to the architecture. The PowerPC 604e in my old PowerMac 8500 had that back in the mid 90's. I wrote some bitching matrix multiplication code in assembler that used it.