Slashdot Mirror
Google Demands Higher Chip Temps From Intel
JagsLive writes "When purchasing server processors directly from Intel, Google has insisted on a guarantee that the chips can operate at temperatures five degrees centigrade higher than their standard qualification, according to a former Google employee. This allowed the search giant to maintain higher temperatures within its data centers, the ex-employee says, and save millions of dollars each year in cooling costs."
In the past, I've under-clocked primarily for noise benefits. Lower clock->lower temp->slower fam RPM->lower dbSPL.
I'm surprised Google isn't considering moving some of its data centres to Arctic locations where you get cool temperatures year-round.
Don't kid yourself. They probably have. But then, who did you get to work at what would probably be a very remote location?
Additionally, such remote locations may suffer from not enough bandwidth and/or electricity.
Bearded Dragon
This sounds like a scenario where lawyers are trying to act as engineers. That works about as well as you might expect.
There are these engineering things, amusingly called "Schmoo plots", that map out a chip's operating envelope of voltage versus speed versus temperature. From those an engineer can forsee how hot you can run a chip before its rise and fall time margins start to get marginal.
There is very little Intel can do to stretch thing by another 5 degrees. It's not something that can be imposed by fiat. Intel engineers have already juggled all the variables to come up with the best performance possible. SOMETHING is going to have to give. Either the chips will have to be selected and graded for speed, lowering the overall envelope for the chips everyone else gets, or they'll have to fudge some other parameters, hoping nobody will notice, or worse yet they'll tweak some variable right to the edge of raggedness, resulting in worse reliability down the road.
Lawyers and accountants generally don't know you can't have everything. let's hope the engineers educate them.
It's not just the celerons, its most CPUs. A modern CPU is quite big and contains a lot of components that aren't essential to the operation of the chip. If you disable these, you have a slightly less good chip without the engineering cost of designing a entirely new die layout. AMD takes this to extremes. Their Opterons have 4 cores, three hypertransport connections and a load of cache. If there is a manufacturing flaw in the cache, they are sold as a model with less cache. If it's in the cores, then they are sold as three, two, or single core chips. If it's in the HT controllers then they only support 2- or 4-way multichip configurations. Intel's 486SX line was just a 486 (later renamed the 486DX) where the FPU didn't work.
I am TheRaven on Soylent News
There are two issues with higher operating temp.
One is that you get less drive current from your transistors, so you get less performance (which everyone seems to understand), but this is usually a fairly small effect for 5 degree C.
The _big_ deal with 5 degree C would be electromigration in interconnect metal, which goes up very quickly with temperature. So the difference in failure rates might be quite large.
If there was any deal at all, it's likely that the Intel engineers tried to remove some conservatism from their temperature estimates to see if they could squeeze out 5 degrees from the thermal budget, or perhaps information on the workload itself to get Intel to "bless" the higher data center temperature.
What was even weirder was the i487SX. This was a i486DX with an extra pin. If you had a i486SX system and wanted a FPU you bought the i487SX and plugged it in. Then during boot the i486SX was disabled and the i487SX was used for everything.