Fastest CPU in April

On Thu, 14 Dec 2006 08:05:43 -0500, George Macdonald
wrote:
The biggest change was probably in the specification rather than the
chip itself. When the initial Pentium D 900 series was released they
used the maintream rating of 95W TDP for only the low-end 920 (2.8GHz)
and 930 (3.0GHz) chips, using the "performance" rating of 130W TDP for
their higher end 940 (3.2GHz) and 950 (3.4GHz) chips. Then, with the
release of the C1 stepping they moved all of those chips to the
"mainstream" rating of 95W but introduced the new 960 (3.6GHz) chip
with their 130W rating. The D0 stepping just bumped that 960 chip
down to the 95W rating.

I dunno where you're getting your info but the doc above and the 900 series
data sheet clearly show the *large* reduction in max power with the C1 to
D0 transition for the 940 thru 960.

I'm getting my info straight from the horses.. umm.. mouth:

http://processorfinder.intel.com/list.aspx?ProcFam=2112

All C1 stepping Pentium D chips that Intel sold had a TDP of 95W
except for the SL9AP, which was a Pentium D 960 C1 stepping rated for
130W.

Now obviously these chips aren't going to exactly match their rating,
a 3.6GHz C1 stepping isn't going to use 35W more power than a 3.4GHz
chip each and every time. There is always some variation to these
things and as the process matures and is tweaked power consumption
will tend to come down. The D0 stepping might have made a few more
tweaks to the chip to improve power consumption further, but it's
unlikely to be a major change. These new steppings are pretty normal
even as a chip is past it's prime.

I have to believe the data sheet - a 20% reduction in *max* power current
draw is more than a tweak.

A 20% reduction in actual max power is is more than a tweak, a change
in specification by 20% might not be much more though. Intel, much
like AMD, is not longer providing an exact spec for the power of
individual chips but rather a couple maxmum power consumption points.
This is entirely targetted at OEMs so that they know what they need to
design their systems for and know exactly what chips they can and
cannot use. This is why you don't see any power rating between 95W
and 130W because Intel simply hasn't defined such any in-between
ratings for their Pentium D line.

Basically this means that the Pentium D 960 C1 stepping might have
consumed only 90W max (or less) most of the time. However there is
going to be a range of power consumption (most likely some sort of
bell curve), and if the top 1% of the chips consumed 96W then Intel
would need to spec the entire sSpec at the next higher rating point,
and that is 130W. In this example if Intel were to bring their
worst-case power consumption down by just a watt or two they could fit
it within their 95W spec. It's quite possible that average power
wouldn't even change much, just a tightening of manufacturing could
reduce the variation between average and worst-case.
--
Tony Hill
hilla underscore 20 at yahoo dot ca

On Fri, 15 Dec 2006 01:37:29 -0500, Tony Hill
wrote:

On Thu, 14 Dec 2006 08:05:43 -0500, George Macdonald
wrote:
The biggest change was probably in the specification rather than the
chip itself. When the initial Pentium D 900 series was released they
used the maintream rating of 95W TDP for only the low-end 920 (2.8GHz)
and 930 (3.0GHz) chips, using the "performance" rating of 130W TDP for
their higher end 940 (3.2GHz) and 950 (3.4GHz) chips. Then, with the
release of the C1 stepping they moved all of those chips to the
"mainstream" rating of 95W but introduced the new 960 (3.6GHz) chip
with their 130W rating. The D0 stepping just bumped that 960 chip
down to the 95W rating.

I dunno where you're getting your info but the doc above and the 900 series
data sheet clearly show the *large* reduction in max power with the C1 to
D0 transition for the 940 thru 960.

I'm getting my info straight from the horses.. umm.. mouth:

http://processorfinder.intel.com/list.aspx?ProcFam=2112

All C1 stepping Pentium D chips that Intel sold had a TDP of 95W
except for the SL9AP, which was a Pentium D 960 C1 stepping rated for
130W.

Now obviously these chips aren't going to exactly match their rating,
a 3.6GHz C1 stepping isn't going to use 35W more power than a 3.4GHz
chip each and every time. There is always some variation to these
things and as the process matures and is tweaked power consumption
will tend to come down. The D0 stepping might have made a few more
tweaks to the chip to improve power consumption further, but it's
unlikely to be a major change. These new steppings are pretty normal
even as a chip is past it's prime.

I have to believe the data sheet - a 20% reduction in *max* power current
draw is more than a tweak.

A 20% reduction in actual max power is is more than a tweak, a change
in specification by 20% might not be much more though. Intel, much
like AMD, is not longer providing an exact spec for the power of
individual chips but rather a couple maxmum power consumption points.
This is entirely targetted at OEMs so that they know what they need to
design their systems for and know exactly what chips they can and
cannot use. This is why you don't see any power rating between 95W
and 130W because Intel simply hasn't defined such any in-between
ratings for their Pentium D line.

I think it's fair to assume that, with D0, the chips which already met the
"mainstream" FMB specs, on C1, dropped down even further and were well
below the reqts; IOW yield distribution moved upwards. I note that the 960
also transitioned from a fixed voltage, with C1, to variable with D0...
pre-empting the OCers?:-)

Basically this means that the Pentium D 960 C1 stepping might have
consumed only 90W max (or less) most of the time. However there is
going to be a range of power consumption (most likely some sort of
bell curve), and if the top 1% of the chips consumed 96W then Intel
would need to spec the entire sSpec at the next higher rating point,
and that is 130W. In this example if Intel were to bring their
worst-case power consumption down by just a watt or two they could fit
it within their 95W spec. It's quite possible that average power
wouldn't even change much, just a tightening of manufacturing could
reduce the variation between average and worst-case.

Without a statistically significant set of measurements, it's impossible to
know what the 960 was running at, worst case, with C1, and your bell curve
is an awfully big assumption - given the highly non-linear behavior here
one can easily assume the other way. Of course one would assume that Intel
*does* the measurements and decides which chips can be pushed on (a fixed)
voltage and hit the higher performance... and which have to be
"downgraded".

--
Rgds, George Macdonald

dMn wrote:
wizzywiz wrote:
wrote:
SNIP

When my program takes days to execute, there is definately a need for
more cpu power.

Or a better program/programmer.

dMn

Quad is only a giant leap for the ordinary desktop. Some scientific and
commercial applications require far more power, and have done for
years.