Second opinion please

  • Thread starter Thread starter GT
  • Start date Start date
G

GT

I'm experimenting with a potential file/print server for my home network
using old PC hardware from the cupboard.

Got an old athlon 2400 (AXDA 2400 DKV 3C) and trying to get it as cool as
possible.

Multiplier is 15x. Correct FSB is 133MHz. CPU should be at 1.65v. Designed
TDP is 65w.

Am using a Zalman heatsink - forget its name, but its a decent, fanless
one - fan was on a bracket above the heatsink, when required.

I'm playing in the BIOS, turning it down and down in stages and have so far
got it down to FSB 100, x15, voltage 1.2 (reporting 1.17v in bios screen). I
haven't found the fail point yet either (will run memtest later)! The
heatsink is barely warm and BIOS reports ABSII temp of 42 deg C. Temp probe
in middle of heatsink reports 38 deg C. Can someone confirm my following
calculations:

Actual TDP (heat watts) = designTDP * (clock_new/clock_old) *
(volts_new/volts_old) * (volts_new/volts_old)
So 65 * (1500/2000) * (1.2/1.65) * (1.2/1.65)
= 25.8w from a 65w CPU!

Does this formula / calculation seem right?
How is the processor operating at such a low voltage - its a desktop CPU not
a ULV mobile?
Was there a difference in the CPU between the mobile and the desktop
variants, or was it just marketing hype?
 
GT said:
I'm experimenting with a potential file/print server for my home network
using old PC hardware from the cupboard.

Got an old athlon 2400 (AXDA 2400 DKV 3C) and trying to get it as cool as
possible.

Multiplier is 15x. Correct FSB is 133MHz. CPU should be at 1.65v. Designed
TDP is 65w.

Am using a Zalman heatsink - forget its name, but its a decent, fanless
one - fan was on a bracket above the heatsink, when required.

I'm playing in the BIOS, turning it down and down in stages and have so
far got it down to FSB 100, x15, voltage 1.2 (reporting 1.17v in bios
screen). I haven't found the fail point yet either (will run memtest
later)! The heatsink is barely warm and BIOS reports ABSII temp of 42 deg
C. Temp probe in middle of heatsink reports 38 deg C. Can someone confirm
my following calculations:

Actual TDP (heat watts) = designTDP * (clock_new/clock_old) *
(volts_new/volts_old) * (volts_new/volts_old)
So 65 * (1500/2000) * (1.2/1.65) * (1.2/1.65)
= 25.8w from a 65w CPU!
Click to expand...

Update: Ran memtest for 15 minutes before I quit and it seems to be stable
at 1.15v (reported as 1.12v) at 15x100, which is 1500MHz (about athlon 1800
speed) running at 23.7W !!. CPU temp looks like it has settled at 40 deg C -
sitting 'open' on the desk no fans anywhere. Heatsink probe says 35 deg C
which feels about right - can barely feel any heat with the back of my
finger at the base of the heatsink! How is this possible with old discarded
equipment??
 
GT said:
Update: Ran memtest for 15 minutes before I quit and it seems to be stable
at 1.15v (reported as 1.12v) at 15x100, which is 1500MHz (about athlon 1800
speed) running at 23.7W !!. CPU temp looks like it has settled at 40 deg C -
sitting 'open' on the desk no fans anywhere. Heatsink probe says 35 deg C
which feels about right - can barely feel any heat with the back of my
finger at the base of the heatsink! How is this possible with old discarded
equipment??
Click to expand...

Your equation looks good.

The processor manufacturer, does have some control over the manufacturing
process. (There was an IEEE article about the things Intel can do, that
was written more than five years ago, and it hints at the capabilities.)
It would be the combination of what they do to the wafer, plus "binning",
that picks out the good, low power or low voltage ones.

I think you should move on to Prime95, to find out how hot it gets
with a real load on it. That might get you closer to a "TDP" situation.
Do a bit more testing, before you get too satisfied with that 1.15V :-)

I'd probably run Prime95 (on a smaller block of memory) and 3DMark2001SE
benchmark, at the same time, as a stability test. I've had cases, where
Prime95 was clean, until I starting running 3DMark. I like Prime95,
because it's actually looking for errors.

Paul
 
Back
Top