RAM temperature sensors needed soon.
 

Cryptocoins may start using algorithms which can be executed fast only on CPU-RAM chips.

Problem with this is current hardware has no RAM temperature sensors ?!?!?

Therefore I am requesting this feature to be added to new hardware as soon as possible.

Bye,
Skybuck.

RAM temperature sensors needed soon.
 

This is probably a golden opertunity for companies like ASRock which have patents on or licenses patents on overheat protection circuitry for motherboards.

This type of invention/circuitry could and probably should be extended to RAM chips... to protect RAM chips against overheat.

Full system shutdown is recommended.

Also windows/bios and such need to be updated to give feedback of what happened after the shutdown and restart/reboot to diagnose the problem.

Would be nice if windows event log has an entry like:

"System shut down due to RAM overheating".

Bye,
Skybuck.

RAM temperature sensors needed soon.
 

wrote:
This is probably a golden opertunity for companies like ASRock which have patents on or licenses patents on overheat protection circuitry for motherboards.

This type of invention/circuitry could and probably should be extended to RAM chips... to protect RAM chips against overheat.

Full system shutdown is recommended.

Also windows/bios and such need to be updated to give feedback of what happened after the shutdown and restart/reboot to diagnose the problem.

Would be nice if windows event log has an entry like:

"System shut down due to RAM overheating".

Bye,
Skybuck.

The idea has been around for a while.
That article is from 12 years ago. You could
tie a thermal sensor into the SMBUS that the
SPD EEPROM uses. Someone has to write code to use it.

https://www.eetimes.com/document.asp?doc_id=1299439

Intel has also had a RAM throttling capability, but
I'm not aware of how, or if, it is used on normal
motherboards.

Module power dissipation (Kingston datasheet), is based
on an industry standard cycle mix. Module power dissipation
would rise, if there was a way to do continuous read or
write cycles. NOP cycles or refresh cycles use less power.
It's the fact that a DIMM does not run flat out, that keeps it
cool.

The DIMM burst size is designed for CPU cache lines.
A burst might be in the 2/4/8 range. But maybe the
RAM chip itself supports 128 or 256 cycles in a row.
The chipset probably doesn't have such a mode designed
into it, so it cannot abuse the DIMM on a whim. You could
design a memory controller in an FPGA, to abuse a conventional
memory chip if you wanted.

But I thought Intel had some ability to measure cycle
density, and signal when throttling might be needed.
I've just never seen any exposure of this feature.
It's never mentioned in any kind of technical
discussions.

The hottest memories were RAMBUS. But there was a
reason for that. Each memory chip was sufficient to
answer all queries. You could "focus" your query
on a single chip on a RAMBUS RIMM, and that chip
would get hot because of the *4W* power dissipation.
The other seven chips in the "rank" would remain
cool by comparison. It is because of this "hot chip"
situation, only on RAMBUS, that RAMBUS RIMMs had
heatsinks *riveted* to the DIMM. So a user could
not remove the heatsink, and cause a chip to smoke.
The metal cover was as much a heat spreader, as
a chip cooler.

The memories we use now, don't do that. All chips
rise to the same temperature in a rank (assumes normal
random computer patterns). The cooler is there to
cool all the chips, rather than spread the heat
away from a single hot chip. As a result, the
cooler plates are not riveted. A user can remove
them if they want.

I saw a recent reference to this topic again,
perhaps on some server motherboards. But the concept
has been around long enough, I no longer pay attention
to such articles.

Paul

RAM temperature sensors needed soon.
 

RAM temperature sensors needed soon.
 

On Monday, November 5, 2018 at 4:15:04 PM UTC+8, Paul wrote:

The hottest memories were RAMBUS. But there was a
reason for that. Each memory chip was sufficient to
answer all queries. You could "focus" your query
on a single chip on a RAMBUS RIMM, and that chip
would get hot because of the *4W* power dissipation.
The other seven chips in the "rank" would remain
cool by comparison. It is because of this "hot chip"
situation, only on RAMBUS, that RAMBUS RIMMs had
heatsinks *riveted* to the DIMM. So a user could
not remove the heatsink, and cause a chip to smoke.
The metal cover was as much a heat spreader, as
a chip cooler.

The memories we use now, don't do that. All chips
rise to the same temperature in a rank (assumes normal
random computer patterns). The cooler is there to
cool all the chips, rather than spread the heat
away from a single hot chip. As a result, the
cooler plates are not riveted. A user can remove
them if they want.

I saw a recent reference to this topic again,
perhaps on some server motherboards. But the concept
has been around long enough, I no longer pay attention
to such articles.

Paul

I presume server ECC FBDIMMs run hotter - I have seen pizza boxes with fans
directed over the memory modules.

RAM temperature sensors needed soon.
 

wrote:
On Monday, November 5, 2018 at 4:15:04 PM UTC+8, Paul wrote:
The hottest memories were RAMBUS. But there was a
reason for that. Each memory chip was sufficient to
answer all queries. You could "focus" your query
on a single chip on a RAMBUS RIMM, and that chip
would get hot because of the *4W* power dissipation.
The other seven chips in the "rank" would remain
cool by comparison. It is because of this "hot chip"
situation, only on RAMBUS, that RAMBUS RIMMs had
heatsinks *riveted* to the DIMM. So a user could
not remove the heatsink, and cause a chip to smoke.
The metal cover was as much a heat spreader, as
a chip cooler.

The memories we use now, don't do that. All chips
rise to the same temperature in a rank (assumes normal
random computer patterns). The cooler is there to
cool all the chips, rather than spread the heat
away from a single hot chip. As a result, the
cooler plates are not riveted. A user can remove
them if they want.

I saw a recent reference to this topic again,
perhaps on some server motherboards. But the concept
has been around long enough, I no longer pay attention
to such articles.

Paul

I presume server ECC FBDIMMs run hotter - I have seen pizza boxes with fans
directed over the memory modules.

Here's a comment from someone.

https://www.phoronix.com/forums/foru...ry-cooling-mod

"After having burnt myself slightly on a heatspreader
for an AMB chip on a FB-DIMM module today... first
off I was surprised but
"

I think at least the first generation controller chips
ran pretty hot. It got better with generation.

I've never had any of those here.

Paul