ALCPU Forums

Hi!

The CoreTemp tool seems to be alive after a long standby period... That's great!
The latest 1.2 version claimed to have Intel Xeon CPUs supported, but it seems to me some models are still missing.

I have two Xeon D based systems nearby, D-1527 and D-1548, they are not detected properly.
I'm sending the screenshots and register dumps from both of them.
I suppose the entire Xeon-D lineup will behave the same.
Hope to see the improvements in Core Temp .

Regards.

Interesting bunch of CPUs.
Thanks for the heads up, support added to the next release.

Thank you!

In future I'll get the boards with D-1567 and Pentium D1508 & D1517, and will check them too.

Checked version 1.3 on D-1527 and D-1548, both are okay. Thank you for doing that so fast!

I have another suggestion for discussion: in fact, there are two "energy values in joules" available for any Intel CPU via MSR values.
The first is generally a computing core consumed energy, while the second value depends on the CPU segment - for desktop/mobile models it's energy consumed by graphics core (GT), while for server/workstation models (without graphics) it's energy consumed by memory controller.

How do you think, is it difficult to add an additional field to the program interface that will show the second power value respective to the CPU model? I think it might be useful...

I think it shouldn't be very difficult to add that value, I just never really thought it would be very useful. I'll consider adding this in a future version, thanks.

The Coolest wrote:I think it shouldn't be very difficult to add that value, I just never really thought it would be very useful. I'll consider adding this in a future version, thanks.

Performed some experiments today, sharing the result (for anyone who is interested):
The reference document is "Intel® 64 and IA-32 Architectures Software Developer’s Manual" located at http://www.intel.com/content/www/us/en/ ... 25462.html
The MSRs responsible for power metering are described in "14.9.1 RAPL Interfaces" section.

In fact there are four energy values that are being measured
-Package (total energy consumed by the CPU)
-PP0 (IA execution cores)
-PP1 (Uncore block, but I think it's on-chip Graphics Engine (GT))
-DRAM (DRAM controller)
As I understand, Package energy = sum of (PP0 + PP1 + DRAM)

I added the following MSR registers to the RW Everything:
MSR_PKG_ENERGY_STATUS=0x611
MSR_DRAM_ENERGY_STATUS=0x619
MSR_PP0_ENERGY_STATUS=0x639
MSR_PP1_ENERGY_STATUS=0x641

When my 4790K is operating with a discrete videocard, and the onboard HD4600 is disabled, the PP1 counter shows zeroes and doesn't increment.
When the onboard graphics is enabled, all four counters are active, so it seems PP1 is responsible for graphics rather than for other Uncore components (like CPU PCIe controller).

The use for additional power meters might be the following:
-to see when the DRAM is heavily loaded or not (good to see when Linx is allocating the RAM and when it performs active calculations)
-to see how much the GT cores consume in different scenarios (for mobile-U-Y CPUs).

Thanks

Thanks, I've got that doc always up to date and open when I work on the Intel side of things.
Sadly I don't have a Xeon to try this out on, but I did check both PP0 and PP1 on an i3 Broadwell-U and it worked pretty well. PP0 registered a lower power usage than PKG and PP1 maxed out at around 1W
I'm not convinced that it's a very useful piece of information to be added to the GUI, as with the current way things are designed will require another line, which takes more real estate on the screen. I'll try to think if there's an intuitive way to somehow put the power values side by side, similar to the Min/Max/Load values.
My 2600K didn't show anything for DRAM_STATUS, I guess that this MSR is only available in the server variety, perhaps in the later chips Intel has enabled them for consumer parts as well.

After some time spent in Paint..... Take a look



I tried to keep the GUI style, and it seems it's pretty good

The only thing is the former TDP fiels, it should be filled with something else in my case

Heh, that's good. I just think it wastes a lot of space vertically that way. I really want to have the power readings fit completely into one line (similar to what you did) but in the main area of the readings.