On Tue, 7 Feb 2006, Delbert Franz wrote: > I have just noticed that inserting a usb key in my computer, a > MSI K8T Neo (MS-6702) MB, using the Via K8T800 chipset, with > AMD 64 3700+ CPU, causes the reported CPU temperature, > via lm_sensors, to increase by about 4 degrees C over the space > of a minute of so. The CPU load does not change in the > gkrellm display while this temperature increase is under way. > I am running Debian Sarge with kernel 2.6.15.2 compiled from > source from kernel.org. > > I did an install of W2K SP4 on a 6 GB scratch IDE drive and > found that the same temperature change occurs under MSW, > suggesting that it is not OS dependent. I suspect it is > in the MB firmware.
Why do you suspect that? The firmware doesn't get involved when you insert a USB device. > I then connected a watt meter to the computer and found that > the idle power usage under Linux was about 106 watts. Inserting > a usb key in one of the built in USB ports causes the power > usage to jump by about 30 watts, to a total of about 136. Of course > it does not take long for the CPU temp to rise by about 4 degrees C. > However, the CPU load shows no change. The power usage and the > temperature rise remain when the usb key is removed. Only a reboot > will reset the system. Removing the usb-storage related modules > might also cut the power. I have not tried that yet. Need to recompile > every USB item as a module to do that. Probably you would need to remove only the host controller drivers (uhci-hcd and ehci-hcd). > If I load the system with a large-scale simulation of > flow in a river, we have a peak power usage of about 157 watts. > The CPU temperature increases by about 4-5 degrees C > from the level it reaches with only the USB key inserted. > > Finally, I added a PCI card with 5 USB ports to the system, and inserted > a usb key to one port on this card. No extra power usage > appears, except for a small blip after the usb key is inserted. > The temperature of the CPU is unchanged. Under full load, the > max temperature is the same as when the key was inserted in > a built in port. After the application ends, the temperature > quickly returns to its idle value. We get a temperature change > of about 10 degree C from idle to full load using a number crunching > application. This probably makes most of the CPU active-integer, floating > point, and addressing. > > I get the same behavior on another computer, built on an Asus > A8V which uses a K8T800Pro chipset. Both computers use an AMI > Bios and both use a VT8237 chip for USB. I guess it should not > be a surprise that they both display similar behavior on this issue. The VIA USB host controller chips have some very questionable behavior. Look at the email threads on this page: http://marc.theaimsgroup.com/?a=113508574400006&r=1&w=2 > Questions: > > 1. Has anyone else detected this behavior on these or other > motherboards? > > 2. Is it reasonable to assume that the behavior stems from some > implementation detail in the chipset/BIOS? Chipset, yes. BIOS, no. > 3. Why does the insertion of a usb key in a builtin port > require 30 watts more power from the CPU???? This is more than > half the increase from idle caused by loading the CPU with a large > numerical simulation application:)! Why do you think the increased power is all being used by the CPU? Why couldn't it be used by the USB host controller (and partly by the USB key itself)? > 4. What is going on in the CPU that the CPU load meter does not > change even though 30 watts more power is being dissipated? Insertion of a USB device prevents the CPU from entering certain low-power states, because it forces the USB host controller to start doing DMA. So even though the CPU isn't doing any more work computationally, it's still using more power. > I currently plan to use PCI based USB ports for usb-storage > connections--I see no need to heat the CPU so much just to do > some I/O on the USB bus. Try using non-VIA components. :-) > Thanks for any insight into this strange behavior. I don't believe > there is anything amiss in lm_sensors and that the problem lies in > the USB subsystem as implemented in the VIA chipset. Quite likely. Alan Stern ------------------------------------------------------- This SF.net email is sponsored by: Splunk Inc. Do you grep through log files for problems? Stop! Download the new AJAX search engine that makes searching your log files as easy as surfing the web. DOWNLOAD SPLUNK! http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642 _______________________________________________ [email protected] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-users
