Ungo (a.delete@this.b.c.d.e) on April 23, 2013 12:33 pm wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on April 23, 2013 8:14 am wrote:
> > As always, I encourage discussion, so let the fun begin!
>
> I disagree with this bit:
>
>

Realistically, retention is not a problem for Intel since Haswell is a client
> chip that is unlikely to reach junction temperatures much above 80C, whereas
> IBM’s server processors must operate at far higher temperatures.
>

>
> Client chips on the desktop seldom reach high temperatures, but mobile is a different story. I have
> temperature monitoring software on a 2011 MacBook Air (Sandy Bridge), and I've seen 90C or higher readings
> from CPU on-die sensors during heavy CPU load. (The i7-2677M CPU is rated at Tjmax=100C.)
>
> SNB/IVB ultrabooks and other slim/light designs depend on permitting the delta-T between
> silicon and ambient air to grow as high as it can without exceeding Tjmax. I expect to see
> Haswell systems doing exactly the same. Nobody would be very happy with Haswell portables
> needing a significantly larger and/or noisier thermal system to keep GT3e eDRAM cool.
>

That's true, but remember that you're measuring Tj for the CPU+GPU. That's physically a separate die from the eDRAM. I doubt the eDRAM has anything that's quite so hot as an FPU inside, so I rather doubt that you'd see high Tj.

David