A Changing Thermal Target: Discussing Haswell

Pat Gelsinger once taught me that a single microprocessor architecture can efficiently target an order of magnitude of TDPs. It's not that you can't scale above or below that range, but at that point it becomes more efficient to use a different microarchitecture.

Let's take Sandy Bridge for example. Current desktop variants of the chip exist at 65W and 95W TDPs and later this year we will see Sandy Bridge E with a 130W TDP. If we pick 130W as the upper bound for the architecture, it should be able to efficiently scale down as low as 13W - or one order of magnitude. Looking at mobile SNB processors, it does.

Intel's ultra low voltage SNB carries a 17W TDP, while mainstream mobile SNB chips are in the 35W (dual core) to 45W (quad core) range. These TDPs all include processor graphics. With Sandy Bridge, Intel has an architecture that spans from 17W all the way up to 130W. I wouldn't be too surprised if we eventually saw some ~13W SNB parts for really low power applications in the future.

At the end of Q1 of next year Intel will introduce Ivy Bridge, its first 22nm microprocessor. I fully expect Ivy Bridge to target relatively similar TDPs as Sandy Bridge, however the initial launch will be confined to TDPs less than or equal to 95W (much like SNB was).

With Lynnfield Intel made it very clear that it's possible to get high-end desktop class performance out of a 95W part. While the 130W chips were still faster, the majority of enthusiast users would get by just fine with Lynnfield. The move to Sandy Bridge highlighted Intel's move away from 130W TDPs for high-end desktop processors and down towards 95W. As I noticed in my transition to a mobile quad-core SNB notebook as my primary workstation, I believe this generation of 130W CPUs will target a smaller portion of enthusiast users than the previous generation. The trend is definitely downward, towards lower TDPs.

Haswell is where Intel's architectures take a dramatic turn. Ivy Bridge is a derivative of the Sandy Bridge architecture, which of course was designed for that 13 - 130W range. Haswell however is a brand new architecture. It'll likely look similar to Sandy and Ivy but its target TDPs will be shifted down. In mobile, Haswell designs will be set at 10 - 20W. That's not the lower bound of the design, just the target for mobile. What does that do to the rest of the scale? Intel presented this slide at its analyst day earlier this month:

In Sandy Bridge, mobile occupied the 35 - 45W range - roughly the bottom third of the architecture's target. Around Haswell two things happen: the mobile design drops and the Atom design target moves upward.

Atom will service a new expanded range from ~800mW to 8W, leaving Haswell to address the ~10W and above market. Multiply that number by 10 and we have our upper bound of 100W - which isn't much different from the 95W we see today for high-end SNB SKUs. That being said, I do believe we'll see a lot more focus around 65W in the desktop.

Meet the Ultrabook Where Does This Leave Atom?
Comments Locked

33 Comments

View All Comments

  • soydeedo - Tuesday, May 31, 2011 - link

    Haha, you super sleuth, you. Thanks for the help. =)

    I actually thought that might be it because of the giant hinge connection/protrusion, but when I went on Google's image search looking for profile pics, nothing really caught my eye. It's a shame, though - I was hoping there was a fully fledged notebook that looked like that. It looks great from that angle anyhow.
  • duploxxx - Tuesday, May 31, 2011 - link

    The final requirement is all Ultrabooks must be sold at mainstream price points, which Intel calls sub-$1000.
    Intel estimates that by the end of next year 40% of consumer notebooks will be Ultrabooks. Given the desirable set of features and reasonable price point, I can see that happening.

    With that pricepoint? your joking right? the share will be about 1-2% some geeks that want to show off on an airport or tradeshow. For that price i buy a decent tablet and a notebook. They will only be able to meat there perf goals in 2013, competition isn't idle in that area.

    7x GPU perf means HD3000 like if they calculate from atom based right now, that ain't magic either :), that will just be brazos perf in early 2012 :).
  • A5 - Tuesday, May 31, 2011 - link

    Apple has reached 5+% of the notebook market selling almost exclusively $1000+ products.

    If I were spending the same amount of money either way, I'd rather have a super-thin notebook than a tablet + a "normal" laptop.
  • Dribble - Tuesday, May 31, 2011 - link

    Got to agree. It's just a macbook air for windows with a different name. The macbook air for mac hasn't sold that well, and that's got all the apple fanboys that should love this sort of thing. The windows market is going to be substantially smaller - haven't other companies already tried this sort of thing?

    Now if it was $400 they'd sell plenty but it's not so I don't see how it'll ever be more then a tiny niche market.
  • name99 - Tuesday, May 31, 2011 - link

    "The macbook air for mac hasn't sold that well, "

    And you know this based on the scientific technique of pulling a number out of your ass?

    COMMON SENSE would tell us that the product is doing well, given that Apple broadened the line from one to two different models. But we can do more than that, we can look at the numbers.
    Q4 2010 Apple sold 1.1 million macbook airs. How does that compare?
    In the same period, Apple sold a total of 2.9 million notebooks.
    I'd say that's doing pretty well.
  • jjj - Tuesday, May 31, 2011 - link

    " I fully expect Ivy Bridge to target relatively similar TDPs as Sandy Bridge"
    Is that just an assumption or any solid info? The combination of smaller process and FinFET might allow them to play a bit with perf,power consumption and dies size.
  • TEAMSWITCHER - Tuesday, May 31, 2011 - link

    I am suspicious about this statement well. The big TDP drop should happen with Ivy Bridge and its process change. Haswell, using the same process, will also have a much lower TDP. If not, I fear that the ultra book will get to market too late to counter the growing onslaught of ARM tablet.
  • Khato - Tuesday, May 31, 2011 - link

    Here's to hoping that they can actually bring it to fruition. There might be a decent chance of it happening, since I'd bet that Intel is basically taking their learnings from trying to get into the phone and tablet market and applying them to a notebook platform. After all, why can't the same basic tricks used in an atom smartphone be applied to a sandy/ivy bridge laptop? Especially since they could now bring 4G mobile connectivity into the platform the same way they brought wireless into the mainstream with centrino. There's some pretty awesome potential, that's for sure.
  • Calabros - Tuesday, May 31, 2011 - link

    at the middle of 2012 ULV IvyBridge (22nm) will be far better in performance/watt ratio than high end Saltwell Atom (yet 32nm).. and its exactly the gap AMD will fill
  • A5 - Tuesday, May 31, 2011 - link

    With what? They haven't said anything about updates to Llano for that timeframe.

Log in

Don't have an account? Sign up now