On Thu, 26 Mar 2015, Jacob Pan wrote:
> RAPL energy hardware unit can vary within a single CPU package, e.g.
> HSW server DRAM has a fixed energy unit of 15.3 uJ (2^-16) whereas
> the unit on other domains can be enumerated from power unit MSR.
> There might be other variations in the future, this patch adds
> per cpu model quirk to allow special handling of certain cpus.
So I have a Desktop Haswell machine (model 60) that is instrumented to
measure actual DRAM power at the DIMM (with a sense resistor).
We are consistently getting RAPL results roughly a factor of 2 smaller
than the actual, measured results.
Does desktop Haswell have a similar units issue to Haswell-EP?
I wasted a bunch of time trying to decode the
DRAM_ENERGY_SCALEFACTOR_MCHBAR values described in the
Desktop 4th Generation Intel Core Processor Family
datasheet but possibly that value in the MCHBAR is unrelated to the one
exported by the RAPL interface.
Vince
On Mon, 15 Jun 2015 13:32:01 -0400 (EDT)
Vince Weaver <[email protected]> wrote:
> On Thu, 26 Mar 2015, Jacob Pan wrote:
>
> > RAPL energy hardware unit can vary within a single CPU package, e.g.
> > HSW server DRAM has a fixed energy unit of 15.3 uJ (2^-16) whereas
> > the unit on other domains can be enumerated from power unit MSR.
> > There might be other variations in the future, this patch adds
> > per cpu model quirk to allow special handling of certain cpus.
>
> So I have a Desktop Haswell machine (model 60) that is instrumented
> to measure actual DRAM power at the DIMM (with a sense resistor).
>
> We are consistently getting RAPL results roughly a factor of 2
> smaller than the actual, measured results.
>
> Does desktop Haswell have a similar units issue to Haswell-EP?
>
Not that I know of. perhaps has to do with the difference of what
and where we are measuring. Let me find a instrumented system to verify
and get back to you.
> I wasted a bunch of time trying to decode the
> DRAM_ENERGY_SCALEFACTOR_MCHBAR values described in the
> Desktop 4th Generation Intel Core Processor Family
> datasheet but possibly that value in the MCHBAR is unrelated to the
> one exported by the RAPL interface.
>
This scale factor is not directly related to the energy counter. The
MMIO DRAM RAPL energy status uses the same unit as the MSR interface,
the resolution is ~61uJ.
> Vince
[Jacob Pan]