Hardware questions
-
I ended going with:
Biostar NM70I-1037U
Intel 320 120GB SSD
Intel PRO/1000 PT Dual Port NIC
Antec ISK 300-150The problem I have now is I am getting high power consumption for this system. I did enable powerD and I can see the CPU speed will throttle all the way down to 200MHz, so I know powerD is working. But I see virtually no difference in power consumption from when I am at the BIOS screen vs when PfSense is running.
I checked the BIOS settings and all power saving features are enabled. I have disabled on-board audio and LAN.
Currently it idles at 35W and load is over 40W, which is way over what the system should be. The TDP of the CPU is only 18W and the SSD can't be more then 1W. I also noticed that even after enabling powerD, that the power consumption did not decrease.
I have two other SFF systems. One is a an AMD E-350 which idles at 15W and the other is a Celeron 520 that idles at less then 30W.
From the BIOS POST screen to the full boot-up of Pfsense I do not see the power consumption go below 35W.
-
Try adding the following to /boot/loader.conf.local
hint.p4tcc.0.disabled=1 hint.acpi_throttle.0.disabled=1 est_load="YES" cpufreq_load="YES"
Restart and try again. EIST is much better than P4TCC.
-
Try adding the following to /boot/loader.conf.local
hint.p4tcc.0.disabled=1 hint.acpi_throttle.0.disabled=1 est_load="YES" cpufreq_load="YES"
Restart and try again. EIST is much better than P4TCC.
Thanks for the help. But after enabling these settings and rebooting the power consumption did not change.
-
Thanks for the help. But after enabling these settings and rebooting the power consumption did not change.
I just realized that the Antec ISK300 uses the regular AC type PSU which isn't quite as efficient at low loads compared to the DC types. Are your other SFF units also using the same PSU?
The PT dual port will soak up about 3 - 5W before PSU efficiency losses. The SSD will take about 1W - 2W, ditto for each stick of RAM.
Not sure about the NM70 chipset though but that may rather depend on the power regulation circuitry specific to each board model. Also, do you have a monitor attached to the unit when running this test? What happens if you attach a monitor but turn it off? -
Thanks for the help. But after enabling these settings and rebooting the power consumption did not change.
I just realized that the Antec ISK300 uses the regular AC type PSU which isn't quite as efficient at low loads compared to the DC types. Are your other SFF units also using the same PSU?
The PT dual port will soak up about 3 - 5W before PSU efficiency losses. The SSD will take about 1W - 2W, ditto for each stick of RAM.
Not sure about the NM70 chipset though but that may rather depend on the power regulation circuitry specific to each board model. Also, do you have a monitor attached to the unit when running this test? What happens if you attach a monitor but turn it off?Removing everything except Ethernet and power does not make any difference. One of my SFF PCs is using a APEX MI-008 case, which has a 250W PSU and this is only pulling 15W with an AMD E-350. If anything the 250W would be less efficient with such a small load then the 150W that is in the Antec case. I am just using a watt meter to measure the power consumption.
-
Thanks for the help. But after enabling these settings and rebooting the power consumption did not change.
I just realized that the Antec ISK300 uses the regular AC type PSU which isn't quite as efficient at low loads compared to the DC types. Are your other SFF units also using the same PSU?
The PT dual port will soak up about 3 - 5W before PSU efficiency losses. The SSD will take about 1W - 2W, ditto for each stick of RAM.
Not sure about the NM70 chipset though but that may rather depend on the power regulation circuitry specific to each board model. Also, do you have a monitor attached to the unit when running this test? What happens if you attach a monitor but turn it off?Removing everything except Ethernet and power does not make any difference. One of my SFF PCs is using a APEX MI-008 case, which has a 250W PSU and this is only pulling 15W with an AMD E-350. If anything the 250W would be less efficient with such a small load then the 150W that is in the Antec case. I am just using a watt meter to measure the power consumption.
So I did some more testing and I did find for some reason the PSU in the Antec case is adding 10W additional to the system.
Even still it seems as though the system does not use any power saving features. I tested the system with the PSU from my other SFF case that is known to go lower then 15W. The total draw with the Intel NIC(5W), and SSD(1W) is 26W at the BIOS. After Pfsense has booted, with powerd enabled and the additional boot-up perimeters suggested in this forum it's still at 25W. I do notice the CPU speed clocking down to below 800MHz in the dashboard of PFsense.
Even with no NIC or SSD in the system it's at about 20W. According to PFsense the CPU does throttle down, but there is no change in power draw. The TDP of the CPU 18W, I would say 2W is the motherboard, so if the system was running at full speed all the time 20W seem to be about right. I would suspect that if the CPU is throttling down that this should drop at least a couple of watts.
Any other ideas?
-
Use the instructions above to disable p4tcc and enable eist. Regular p4tcc doesn't help save any significant power. Your running frequency should not go below 800mhz if powerd is using eist.
If there is an option in bios to reduce the gpu frequency, use that. On that chip, the lowest is either 400mhz or 350mhz. -
Use the instructions above to disable p4tcc and enable eist. Regular p4tcc doesn't help save any significant power. Your running frequency should not go below 800mhz if powerd is using eist.
If there is an option in bios to reduce the gpu frequency, use that. On that chip, the lowest is either 400mhz or 350mhz.I have used the settings you provided and enabled powerD with Hiadaptive mode.
You are correct that the CPU speed does throttle down to 800MHz and not any lower, but the power consumption does not decrease. I have every power saving feature in the BIOS enabled and the GFX settings are on the lowest. I also turned off all the on-board functions that I am not using like audio and LAN.
I have posted a link to a PDF with the BIOS settings below, maybe I am missing something.
Thanks for the help.
-
Yep disable P4tcc (unless you're running a P4-M CPU). You should also disable ACPI throttling which does nothing useful for power consumption. (Edit: which you've done) There's a great explanation of it here:
https://wiki.freebsd.org/TuningPowerConsumptionMy experience with newer CPUs is that if they support lower C states, and are enabled to do so, then it's very difficult to see the power saving achieved by running lower P states. I have speculated that if the CPU spends any time at idle then the savings due to switching to a low C state are substantially greater than P state saving. The only time you see such a saving is when the CPU is running with a continuous moderate load preventing it using C states.
I'm open to correction on that though. ;)
Check the sysctls mentioned in that link to see if the correct C and P states are being reported by the bios. Check the boot log to see that the est driver is attaching to the CPU.
Steve
-
Yep disable P4tcc (unless you're running a P4-M CPU). You should also disable ACPI throttling which does nothing useful for power consumption. (Edit: which you've done) There's a great explanation of it here:
https://wiki.freebsd.org/TuningPowerConsumptionMy experience with newer CPUs is that if they support lower C states, and are enabled to do so, then it's very difficult to see the power saving achieved by running lower P states. I have speculated that if the CPU spends any time at idle then the savings due to switching to a low C state are substantially greater than P state saving. The only time you see such a saving is when the CPU is running with a continuous moderate load preventing it using C states.
I'm open to correction on that though. ;)
Steve
I will have to do some testing with the additional C states and see if that helps.
-
@raidflex: It would appear that your PSU and/ or board may be the limiting factor in this endeavour. For best results, use a DC/DC type PSU. Those offer the best power efficiency at lower (<100W) loads.
My experience with newer CPUs is that if they support lower C states, and are enabled to do so, then it's very difficult to see the power saving achieved by running lower P states. I have speculated that if the CPU spends any time at idle then the savings due to switching to a low C state are substantially greater than P state saving. The only time you see such a saving is when the CPU is running with a continuous moderate load preventing it using C states.
Pretty much accurate. Intel markets it as Idle States/ Idle-64 but C-states does benefit most CPUs significantly more than EIST/ CnQ especially when the OS isn't able to enable/ utilize the latter. In short, it gates certain areas of the CPU that aren't in use to save power but doesn't reduce the Vcore.
There are certain situations where EIST can enable even more power savings. EIST allows for VID (Vcore) drops when the multiplier is reduced. This can allow for more power savings since the active sections of the core will also consume less power.
However, most ULV or LV chips are already operating at close to the minimum possible Vcore so these won't benefit as significantly from EIST. The Celeron 1037U with the U suffix actually belongs to this group.
For a 17W TDP ULV chip, being able to gain 1-2W of savings is probably as good as it gets. Older 65nm or 45nm Core 2 Quads are probably the ones that benefit the most from EIST v-drops. -
Many C states supported by newer CPUs do reduce the core voltage. And by new I mean above 486. ;)
http://www.hardwaresecrets.com/article/611
Edit: Totally read that wrong. Late P4 models and above support C1E. ::)I was unable to measure any difference when trying to enable eist on an E4500 Core2Duo. However I was having to modify the bios to do it so it's entirely possible I got it wrong!
Steve
-
I was unable to measure any difference when trying to enable eist on an E4500 Core2Duo. However I was having to modify the bios to do it so it's entirely possible I got it wrong!
Steve
Too extreme for my blood - BIOS modding. What chipset are you working off anyway?
-
Chipset, hmmm. Looks like it's G41 Express: https://forum.pfsense.org/index.php/topic,43574.msg259735.html#msg259735
By modifying the various bios registers neccessary and working up my own custom DSDT table I was able to make the est driver attach to the CPU and report useful values but never actually see any change in power consumption. I attempted to read the CPU core voltages directly from the CPU and they seemed to changing appropriately. CPU frequency definitely changed. However, as I said, I was learning about how to do any of this as went along so the chances of some huge error creeping in were huge. :P
https://forum.pfsense.org/index.php/topic,43574.msg264887.html#msg264887
Steve
-
@Steve: Considering the hardware you have and the PSU, I doubt you can get any lower short of replacing the PSU with a more efficient unit. The G3x/ 4x chipsets do seem to soak up quite a fair bit of power. When I had my E5200, moving from the G35 to an Nvidia Quadro FX470 (9400mGPU) board saved me nearly 15W at the wall. My bet will be on PSU efficiency though.
-
I completely agree, it's never going to be a super low power box. You pretty much need a mobile chipset and a DC power supply. I still expected to see a few Watts difference Other boxes I have enabled EIST on have dropped 5-6W at idle, well within the tolerances of my power meter for instance. I saw nothing at all. I concluded either I was doing it wrong or the effects of enhanced C states were swamping any differences.
Steve
-
Maybe you got "lucky". Some cores are from good yields where the VID would already qualify them for LV or ULV chips but the demand might not be there to package them as such. Really depends on the actual VID of your chip though. Alternatively, the board uses a fairly inefficient VRM so you don't see much difference. I do know the 65nm chips tend to guzzle a fair bit of power (by today's standards), at least my E2160 did until so.