Question about CPUs. (Warning: A stupid one)

[Zac67]

A CPU running slower after some time is complete nonsense.

But:
- I had to downclock my Athlon XP 1700+ after ~5 years from 1866 to its nominal 1533 MHz - the system kept rebooting in spite of proper cooling. Probably a voltage raise would have removed the problems, but the board didn't allow for that and more voltage speeds up electromigration thus aging.
- I noticed my P4 Prescott at work was only throughputting 2 Mkeys/s in distributed.net's RC5-72 - the cooler was pretty clogged up. After a little bit of cleaning the noise was massively reduced and the speed was back up to ~7 Mkeys/s.

@CLS2086:
Have you defragmented your drives completely or formatted them prior to measuring?
My drives run 24/7 for years (usually 4-5) and I've never observed a significant (maybe barely measurable, but surely unnoticeable) decrease of speed (seek, throughput)...
They're not RAIDed however, in certain RAID modes you may see significant speed decreases when the drives are not spindle synchronized.

[Zac67]

Xamiche wrote:

@the guy who's computer benchmarked slower some years later with the same set up. If the heat sink and fan on the CPU or VPU were functioning less effectively, quite likely considering the age, then those chips where mostly likely running hotter and that would result in poorer CPU/VPU performance and therefore a lower benchmark score.

Only few chips are actually capable of downclocking when overheated (e.g. Pentium 4). Most simply crash.

[Zac67]

It not a matter of downclocking. You still lose performance when it over heats because of errors in calculations, or rather operations, and time spend doing the operation again because of the error.

This is complete nonsense - how's the CPU supposed to know when it's wrong? There's no way to do it other than do everything twice - and nobody's designing that way (not in consumer world).

Take a working Pentium system, run a first person shooter with a frames per second read-out and then slow the CPU fan down. Watch the frames per second drop.

Yes, a P4 will downclock when it's sensor shows overtemp - not because of 'errors in calculation'. A lot of calculation done by CPUs is that of addresses and jumps, when something goes wrong there, there's nothing to do for recovering. Either the program trashes data or crashes.

I say use a Pentium chip because it more resilient to heat than an AMD chip. It will survive where an AMD will most likely fry.

Even more nonsense. You should have paid more attention or maybe your teacher has told you wrong: take a PIII and stop the fan. Voilá: crashed. Again: P4s monitor their temp and skip clock cycles (on purpose) when overheated.
AMD were said to fry easily because the Socket A generation had the heatsink directly attached to the naked die. When the HS was not fixed accurately, the chip fried. This does not happen any more with Socket754/939/940/AM2 CPUs.

[Zac67]

Depends how you look at this but a graphics processor like the Geforce 6800 has pipelines that are part of it's speed. So if a couple of pipelines should breakdown over the course of a few years you could turn off those pipelines and still have a working card but slower.

You don't have the option to deactivate single pipelines without altering the driver (depending on design, modding the driver may not be enough). If one of the pipelines fails, you'd get garbled output and bin the card.

This is basically the difference between several models of the 6800. They are just missing some pipelines.

True. But the different version are detected by the driver and programmed accordingly. If there are no hidden tweaks in the driver and no way to simulate another version by opening/closing jumpers, reflashing BIOS etc., there's no way to make the card stop using the broken pipes.