The Inquirer-Home

Nehalem bits bubble up

Some new ones too!
Thu Mar 29 2007, 02:49
INTEL IS FINALLY starting to talk about Nehalem, oh happy day. In the talk today about Penryn, Pat Gelsinger followed it up with some bits about Nehalem.

While it is nothing as detailed as this, this, this, or this, it is still a good first step. For the basics, Nehalem isn't the massively new ground up architecture they want you to think it is, but it definitely is a heavily tweaked core. The biggest advance comes in FP capability, so expect big gains there.

The parts that will change heavily is the IO, they will bring CSI and IMCs to the table. We told you about the IMC situation earlier, and not much has changed there. The only noteworthy part is that Intel is copying AMD's micro-buffer strategy on 2S.

The CSI component is the interesting part, with the lower end Gainestown having 2 connections and Becton having 4. All run at 6.4GT/S unless they go for the longer length option, then it will drop speeds to 4.8GT/S. The electricals will be similar to PCIe2, giving less off-chip voltages to deal with, avoiding one of the major scaling headaches AMD ran into.

For memory, the situation is well known, 4S gets FBD2, 2S gets DDR3 with the micro-buffers, and normal parts get plain old DDR3 with external controllers. The more things change, the more, well, we tell you about.

HT makes a comeback with two threads per core, no shock there. The core count itself goes up quite a bit, with Beckton being listed as having four, six and eight core with TDPs ranging from 90W to 130W. This means the current TDP envelopes will stay the same with double the core count and an IMC. No easy feat there.

The biggest new thing they dropped on us was integrated graphics as an option. Since this is not set to be pulled on die till Gesher, I can only take a semi-educated guess that they mean on package graphics, not on die. The heavy use of 'optional' only reinforces this.

Last up, we get to performance. More FP, no FSB bottlenecks, and an IMC, that does not sound slow. It isn't. We would guesstimate that Thurley will be better than it's Penryn predecessor by a tad more than 1.5x and 2x on Int and FP respectively. Stoutland will be closer to 2.5x the performance of Dunnington, the platform advances will pay off here, especially on the FP side.

It is on track for a late 2008 launch, and with Intel executing well, we have no reason to doubt that. Had a lot of this tech come out in 2006 with Whitefield it would have been a monster. Three years later, it is just a big step in the right direction. µ


Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

INQ Poll

Heartbleed bug discovered in OpenSSL

Have you reacted to Heartbleed?