USB Temperature

[Denorios]

I don't know whether there just something odd about my computer, but adding and removing USB devices seems to have a marked effect on the CPU temperature. I just removed my 128MB flash stick, and the temperature dropped 5C fairly sharply. I plugged it in again and the temperature spiked back up. I'm not even transferring data, it's just plugged in

Has anyone had a similar experience? Does anyone know why this effect is so pronounced?

[acaurora]

Resources are used to maintain the connection. However I am surprised that the CPU temperature rises. For me, rather, it is the device. For example, I have a Sandisk Titanium 512 MB Flash Drive (uber fast), and it gets a little warm, even after plugging it in and not doign anything.

[dukla2000]

The most plausible explanation I have is that
1) Nomally the CPU at idle is entering some HLT (or even STPGNT) state
2) The USB disk driver uses a loop (rather than interrupts) to monitor the device once it is inserted, and that loop is tight enough to prevent the 'usual' benefit ffrom HLTs.

[Denorios]

Trouble is, I'm running CPUIdle on the advanced cooling setting, so the USB can't be disrupting the HLT state that much, can it?

The effect is more pronounced with USB 2.0 devices than with USB 1.1

[Denorios]

Just tried it again. When I put the stick in, the CPU climbed from a stable Idle of 41C to 47C in just over a minute

I'm going to see what the delta is when CPUIdle isn't running.

[Denorios]

Look's like you're right, dukla. When I disabled CPUIdle and let the temperature stabilise, inserting and removing the memory stick had no effect on the temperature at all. However, when running, CPUIdle is showing 99% idle.

Bugger. How irritating. I wouldn't have thought that the serial bus needed that much monitoring - after all, the IDE interface doesn't.

[Wraith]

Has anyone had a similar experience?

Yes. We recently got a USB2 scanner, Canon 3200F I believe. Plugging in the USB made CPU temps rise. 5C or better, I'd guess. Even though I wasn't scanning, and it isn't one of the scanners that's powered through USB.

I don't run CPUIdle. Just running MBM5 and TrendMicro Internet Security.

[dukla2000]

Look's like you're right, dukla.

Not again

I wouldn't have thought that the serial bus needed that much monitoring - after all, the IDE interface doesn't.

I haven't noticed the temp effect, but with a new 2.5" drive in a USB 2 external box I am noticing all kinds of other interesting things. Like under Linux about 20 processes are spawned when I plug the thing in. And on some machines USB 2 ports run at USB 1 speed because of IRQ problems, and also can run at USB 1 speed because of other USB peripherals plugged into the same controller.

All in all it seems to me USB delivers plug n play, but in general includes some really sloppy coding/cludges/compromises to get there. When I was a younger and worked on real machines ALL devices spoke to the CPU via interrupts, and the particular systems I worked on allowed for 4 * 256 hardware IRQs!! The basic PC limitation is the Intel 86 architecture is stack based as opposed to (hardware) interrupt orientated. And yes, again in my old days, people did software interrupts: they were measured in milliseconds compared to hardware interrupts in nanoseconds. My conclusion is that USB is delivering a software solution to what is an architectural hardware limitation.

[kesv]

My conclusion is that USB is delivering a software solution to what is an architectural hardware limitation.

A plausible conclusion. As I understand it, USB is a design with 'dumb' hardware i.e. it needs quite a bit of software to work. Apparently this is what separates it from Firewire for example.
Kind of like hardware- vs. software-modems.

[Ducky]

My conclusion is that USB is delivering a software solution to what is an architectural hardware limitation.

A plausible conclusion. As I understand it, USB is a design with 'dumb' hardware i.e. it needs quite a bit of software to work. Apparently this is what separates it from Firewire for example.
Kind of like hardware- vs. software-modems.

Correct! USB requires software to do the actual negotiation and data transfers -- the only reason Intel pushes for USB is because it's:

1. Cheap to make.
2. Requires more CPU cycles to work, which would make people upgrade their computers more often.

That's one of the reasons why I almost always recommend people use Firewire devices instead of USB, but nobody believes me...

[josephclemente]

I wonder if a USB mouse or keyboard is heating up our computers as well... I'll have to try unplugging some stuff...