Hans, out of posts on the mod thread for another few hours.
You wrote:
A Quad Core at 2.9 GHz with 125W seems OK since 31W per core
is less damaging as the 40W per core for CWM. (The 40W/core
at 65 nm corresponds to a ~160W dual core processor at 90 nm
in terms of energy density, high indeed)
At 45nm the maximum allowable power per core goes down to ~20W
Intel stated 20% higher transistor speeds at 45nm compared to
65 nm, part of that goes lost in the interconnect. Now, I'm
rather pessimistic here and I expect the higher electric
stresses and half the power budget will limit higher speeds
for Intel DC 45 nm parts.
Up to now there hasn't been any statement from Intel about
the clock frequencies of its 45 nm DC parts. I think we should
not base our judgments here on hobbyist guesswork on the
internet.
Last point first: All we have for any of this is internet rumor, plus anything we can reason out.
---
As to AMD and a 2.9GHz QC K8L @ 65nm. I don't think so. Here's why:
- AMD is currently at 65W with a 2.6GHz K8 DC part. It must be very near the actual limit, otherwise, they would've gone to 2.7GHz, but they didn't.
- Double that, and you get 130W - any extra IMC power for a 2.6GHz K8 QC part.
- But K8L core surely dissipates more power than K8 core.
- Also, added L3 power dissipation, probably low.
I think the last two probably more than make up for the "- extra IMC" term, leaving a 2.6GHz Barcelona over the 125W power budget allowed. Hence 2.5GHz max speed @ 125W for QC.
I think this sort of analysis has the benefit of being based on the power of a part already on the same 65nm process. Can you refute parts of it?
---
As to energy density, are you saying we are at that limit? and are you including cache area in "core area"? I also think your 40W / core is probably high, given that Intel are currently at 30W / core for CMW @ 2.66GHz.
My calculations for a DC 45nm die size for Penryn are here:
Message 23085773
So ~105mm^2 for Intel 45nm DC, compared to 143mm^2 for Intel 65nm DC Conroe. Including all L2 per core then, density is only up 36%...
vr-zone claim from some source, for what it is worth, was 57W, DC, 6MB L2, 3.5-4GHz.
That is 28.5W/core. Now, using my die size estimate, that works out to 1.36 * 28.5W = ~39W/core equivalent energy density at 65nm Conroe. Interesting, close to your 40W/core density.
Maybe vr-zone is right. |
