Re. power consumption I found this snippet in the Technical Guide interesting
(listed under Sustainability):
"VCL is a co-optimization between hardware and firmware to dynamically optimize
and
micro-tune the voltage applied to each processor chip for the current operating
conditions.
This reduces the net leakage current and power consumption of the machine. (On
prior
generations of hardware this voltage level was set at a fixed margin for all
parts shipped to the
field.)
The VCL algorithm monitors power and thermal sensors within each processor core
to detect
when margins are approached on the voltage level. This might result in rare
instances of
processor pipeline throttling and adjusts the voltage level to alleviate said
conditions."
On the one hand, that seems like smart engineering. On the other hand it seems
like a complication that adds some ("rare") risk of performance degradation for
what seems like would be an extremely small savings compared to the overall
system power consumption.
I may be overly sensitive to the issue though as I'm chasing a problem on my
main development PC that appears to be related to the similar dynamic power
adjustments for the AMD processor. Making a slight increase in the power (PBO)
"curve" seems to help it avoid sudden unexpected shutdown, especially at low
utilization times. (My theory is that it's letting the power drift too low.) Of
course I expect IBM to have done a lot better engineering for the z17!
Scott Chapman
On Thu, 17 Apr 2025 00:47:30 +0200, Radoslaw Skorupka <[email protected]>
wrote:
>W dniu 16.04.2025 o 19:17, Paul Gilmartin pisze:
>> On Wed, 16 Apr 2025 11:11:00 -0500, Charles Mills wrote:
>>> The detail reasons have been posted by others. Fast cycle time = more power
>>> = more heat = big problem on a small piece of real estate. Size (length of
>>> electrical signal), heat dissipation and cycle speed work against each
>>> other.
>>>
>> The Intel processors reduce power during wait.
>Not exactly true. It depends on many factors. Note: laptop is not good
>comparison to server.
>
>> I know; I have one. When it's busy
>> the chassis gets hot and the fan turns on.
>See above. It can be desktop PC as well. Look at CPU-ID program and
>watch GHz speed. It's changing constantly. For servers it is usually
>disabled.
>
>
>> Long ago I learned here that IBM processors
>> never reduce power. Is that still true?
>No, it's not. Look at POWER presentations. Or simply break cooler in
>your CPC.
>
>--
>Radoslaw Skorupka
>Lodz, Poland
>
>----------------------------------------------------------------------
>For IBM-MAIN subscribe / signoff / archive access instructions,
>send email to [email protected] with the message: INFO IBM-MAIN
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
