You are not logged in.
I recently bought a Foxconn NetBox nT330i, which has an nVidia ION Chipset and an Intel Atom 330. Install went fine, all hardware is up and running, including the driver from the "nvidia" package.
The (kind of weird) problem now is that the box heats up very quickly after bootup and the integrated fan kicks in, keeping the temperatures around ~36°C (96°F). When I don't start Xorg, the fan keeps active at full speed, even if there is absolutely no workload on the box. Starting Xorg (twm), the temperatures get significantly better, and the fan slows down for some short periods.
I suspect the nVidia driver to make the difference, but really don't understand what's going on here. Any ideas?
Thanks,
lynix
Offline
I'd say that the kernel doesn't do any power managment on the ion chipset (while nvidia driver does), and it defaults to full performance on boot, causing extra heat.
- Did you checked if is there some setting to operate in the bios?
- What if you just do a "modprobe nvidia"?
- Maybe you could just start X at boot and kill him, if you don't want X at all, hoping that the driver initialize ION hardware the way you want.
...Just guessing.
Help me to improve ssh-rdp !
Retroarch User? Try my koko-aio shader !
Offline
Thanks for your thoughts, kokoko3k
I do already load the nvidia-driver on bootup via /etc/rc.conf, which does not seem to have any effect on temperatures. It really only depends on runnin Xorg; when I kill it, the temps instantly go up again - weird ^^
Unfortunately the bios is very minimalistic: no options concerning hardware monitoring or fan control. When powered on, the box says "Quiet mode enabled", but I see no option for that in the bios - I don't know what that "Quiet mode" is at all.
Offline
See [wiki]cpufrequtils[/wiki].
EDIT: I installed Arch on an AMD laptop recently, and until I set up CPU frequency scaling on it, the keyboard would become too hot to rest my fingertips on within minutes of turning it on.
Last edited by rransom (2010-06-17 11:48:15)
Offline
Do you mean gpufrequtils?
Just joking,
but i doubt cpu governor or mhz has something to do with nvidia power managment.
Help me to improve ssh-rdp !
Retroarch User? Try my koko-aio shader !
Offline
Thanks for the tip rranson, but according to Wikipedia, the Atom 330 does not support frequency scaling.
Offline
Have you tried using fan speed control? Could be that none of the sensors will work, but worth a shot.
I also have an ion 330 based machine running Arch and found that replacing the thermal compd on the heatsink helped with temps (not GPU temps but CPU temps). You just need to purchase a couple of copper shims (found mine on ebay.) That seller is good and shipped promptly. He also offers 2 packs for like 5-6 bucks shipped.
On my MB the Zotac IONITX A-D, you just remove it from the case, take out 4 screws, and the HS comes off. One extra step was taking some 220 then 400 grit sand paper to the heat sink to remove the hardened thermal material from the factory where the GPU mated up to the heat sink. I used an old tube of Arctic Silver 5 on the GPU+shim and CPU+shim, then reinstalled. Below are the before and after temps. I heated it up using prime95 (4 threads) for Linux x86_64 v 25.11 using large FFts for 20 min.
Link to photos: http://bluesky.jalbum.net/IONITX/
Before (factory thermal pad)
(Idle/Load)
GPU: (33/35)
MB: (28/32)
CPU: (40/52)
Core0: (43/54)
Core1: (41/52)
After
GPU: (33/34)
MB: (28/30)
CPU: (36/40)
Core0: (38/43)
Core1: (36/41)
Differences
GPU: (0/-1)
MB: (0/-2)
CPU: (-4/-12)
Core0: (-5/-11)
Core1: (-5/-11)
Pretty sweet mod for a few copper shim ($6 shipped).
Last edited by graysky (2010-06-17 20:44:00)
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
The powertop utility can save some wattage: pacman -S powertop
Run it several times as it will cascade through several levels of very helpful instructions -- but only as things get fixed. ;-) Works fairly well for Atom.
Offline
Thanks graysky for your detailed reply. I tried the fan control but no sensors were discovered (as expected).
Then I did some benchmarking an testing and made some nice graphs:
The copper shim method you described ops out as there already is such a chopper shim between fan and the heating parts.
In the end the cause of the heat seems to be the main memory which heats up the whole case. I think the easiest solution will be to make some holes in the top of the case so that the fan will get fresh air to blow at the parts it cools.
Found some good blog post to show how it looks like in that box: link.
@soloport: thanks, I already tried that
Last edited by lynix (2010-06-24 20:09:20)
Offline
@lynix - GPU is WAY too hot. Spend the $6 on the shims and do the mod.
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
Okay, it got on my nerves to leave the case open all the time to prevent the fan from blowing at maximum speed.
So I opened the case, removed the fan and that's what it looked like:
Then I replaced the bulk thermal compound with Arctic MX-3. Unfortunately that did not improve the heat situation, so I'm now thinking about your chopper-mod, graysky.
Do you think I can do it with that cooling unit, too? It's already got some chopper planes, connected to an internal heat pipe.
Offline
@lynix - it's been several months, please update the thread with your results using the copper shims. BTW, coretemp is off by 20C. Edit /etc/sensors3.conf and add:
# ZOTAC IONITX-A-U Atom 330 1.6GHz Dual-Core 441 NVIDIA ION
# needs both coretemp and lm90 modules
chip "coretemp-isa-0000"
label temp1 "Core0"
compute temp1 @+20,@-20
chip "coretemp-isa-0001"
label temp1 "Core1"
compute temp1 @+20,@-20
chip "w83l771-*"
label temp1 "M/B Temp"
label temp2 "CPU Temp"
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
As I wrote above I wasn't sure about that shim-mod. That's why I asked you:
I'm now thinking about your chopper-mod, graysky.
Do you think I can do it with that cooling unit, too? It's already got some chopper planes, connected to an internal heat pipe.
You didn't answer and in the meantime I did a different mod: I drilled some holes in the top of the case and mounted a 120mm case fan:
Now the temps are low enough for the internal fan not to come up. I'll add the values you suggested, and do more temperature logging.
Next time I'm in a hardware store I'll fetch some copper shims, and finally do your mod
Offline
Sorry, didn't read carefully. I dunno what our cooling unit is exactly... a powered cooler with HS?
Last edited by graysky (2010-11-04 22:36:17)
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
Never mind
I dunno what our cooling unit is exactly... a powered cooler with HS?
Yes, it consists of a little fan and an aluminium-block with an integrated heatpipe that connects the two copper plates on CPU and GPU with the copper fins:
With your values added in /etc/sensors3.conf, I got these readings (after a full system upgrade):
coretemp-isa-0000
Adapter: ISA adapter
Core0: +45.0°C (crit = +110.0°C)
coretemp-isa-0001
Adapter: ISA adapter
Core1: +48.0°C (crit = +110.0°C)
I have no reading for the M/B, though. Do we have the same board? Your box is a Zotac, right?
Offline
Yeah, Zotac. Sorry. Your temps may or may not be off by 20C (probably not actually).
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
Related to 1st post:
When X is not started, gpu runs at the speed set by the graphic bios (usualy the 3D clocks at full voltage). So it will get hot if you don't start X. GPU only throttles down to 2D clocks after loading the nvidia drivers. <- this explains why temperature rises when you kill X.
Also your gpu is running at normal temperatures. Anything below 60ºC is perfect! Your gpu is designed to run at max of 100ºC And also your cpu.
I don't see any reason to mod anything... Copper mod doubles the sides that are exposed to air (before cpu -> copper, now cpu -> copper -> copper), it's not a good idea to do that. Thermal pads are safer when there's space between cpu and the heatsink.
Replace the original thermal compound is a good thing to do, but never put thermal compound where there's suposed to be a thermal pad.
Last edited by TigTex (2010-11-05 00:02:08)
.::. TigTex @ Portugal .::.
Offline
@TT - I don't follow your logic there with regard to sides exposed to air...
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
With that copper between the cpu and the heatsink you will double the chance of getting air bubbles because you have to apply thermal compound twice (between cpu and copper and between copper and the pc heatsink)... And also the copper may damage something if it too thick. Usualy thermal pads are safer, but not so efective. The temperature difference is not worth the effort in my opinion, and usualy components with thermal pads have low TDP and don't require such cooling mod, like ION.
.::. TigTex @ Portugal .::.
Offline
@TT - I understand your points. In general, there isn't a reason beyond being compulsive to do this
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline
Yep, I already did that mod on a laptop and it didn't improve nothing at all. But I did that just for fun/knowledge... 70ºC under full load on a mobile gpu (nvidia 9400m) isn't a reason to mod it... Undervolting the cpu helped, but that was a core 2 duo, not an atom. And also keeping the heatsink clean and apply good thermal paste like arctic mx-4 helps a lot with temperatures and fan noise.
There should be a way to set 2D clocks without having to start X. KMS with nvidia would solve that, but at the moment, there's no such thing Maybe nouveau is/will be capable of setting the card to 2D clock speed with KMS, I really don't know the current state of that powersaving feature.
So lynix's problem is more like a software problem, not hardware...
Last edited by TigTex (2010-11-06 20:31:25)
.::. TigTex @ Portugal .::.
Offline
Offline
@graysky - see? no change in temperatures
Lynix, are you still running your box without starting X?
.::. TigTex @ Portugal .::.
Offline
@graysky - see? no change in temperatures
Lynix, are you still running your box without starting X?
Dunno what to tell you. I appreciated a pretty major decrease as I posted before:
Before (factory thermal pad)
(Idle/Load)
GPU: (33/35)
MB: (28/32)
CPU: (40/52)
Core0: (43/54)
Core1: (41/52)After
GPU: (33/34)
MB: (28/30)
CPU: (36/40)
Core0: (38/43)
Core1: (36/41)Differences
GPU: (0/-1)
MB: (0/-2)
CPU: (-4/-12)
Core0: (-5/-11)
Core1: (-5/-11)Pretty sweet mod for a few copper shim ($6 shipped).
CPU-optimized Linux-ck packages @ Repo-ck • AUR packages • Zsh and other configs
Offline