Heatspreaders on server memory...

[PliotronX]

So in servers class we were picking apart rackmount servers and I was surprised to find heatblankets... I mean heatspreaders on the memory modules. I wish these things would go away. Number one the chipset controller for the registration is a different height than the memory chips themselves and B (intentionally changed to lettering scheme for hilarity) spreaders restrict airflow which memory needs a lot more of than a piece of aluminum. I always take spreaders off and place a quiet fan on them and it has always resulted in better overclocking and longer lifetimes. In such a constricted space I would think airflow would be paramount. Sorry for the rant.

Added bonus, I had pulled the heatsink from one of the CPU's on my 1U and there wasn't a syringe of thermal grease in the lab. So I had swabbed it with alkeehawl and left a note on my networking professor's nice new 1U server that due to a mishap there was no thermal interface material on CPU2. I redeemed myself this evening in networking class by laying a nice application of ceramique ii on that bad boy. I told him that I would bet him that CPU would run about three degrees cooler than the other. Now I'm super curious and wanna fire it up.

 

[Cerb]

FB-DIMMs, which were mostly used by Intel instead of plain DDR2, needed heatsinks, particularly on the first slot.

 

[PliotronX]

Ahhhh that makes sense, according to Wiki the memory buffer communicates with the chips themselves in serial so that one chip is accessed at a time a la RDRAM which is the type of RAM that actually needed spreaders. Thanks for that.

 

[aigomorla]

So in servers class we were picking apart rackmount servers and I was surprised to find heatblankets... I mean heatspreaders on the memory modules. I wish these things would go away. Number one the chipset controller for the registration is a different height than the memory chips themselves and B (intentionally changed to lettering scheme for hilarity) spreaders restrict airflow which memory needs a lot more of than a piece of aluminum. I always take spreaders off and place a quiet fan on them and it has always resulted in better overclocking and longer lifetimes. In such a constricted space I would think airflow would be paramount. Sorry for the rant..

which servers were u picking apart?
because i'll know to AVOID them.

if ur heat spreaders arent working right... that server was designed by 9 yr olds... or a really old server...

When done right it should be like this:

Which u see is ducted from this:

Also here is another one... u can see having heat spreaders on this type of package wouldnt of made a difference either.. i bet the ram isnt high performance ECC either due to the lack of heat spreaders. Higher Cap ram typically wont have heat spreaders because they are typically not clocked as tight as the lower cap ram. (timings)

in which i can tell you for sure heat spreaders will prevent your memory from failing when running too hot which ECC Registered happen to do when being hammered with work.
ECC Registered runs HOT... hotter then your typical desktop ram... there is never a "excess or waste" when it comes to enterprise... its called Fail Safe.. as enterprise doesnt care about bling... or noise... (u should know how loud Rack Cases can get...)

And ram which is overclocked also run hot and require heat spreader as well to prevent them from failing.
I have both heat spreaders.. BIG massive ones.. aka dominator heat spreaders with a Fan on top of them on my desktop.

If heat spreaders didnt work, they wouldnt be mass used on the higher performance ram... and ECC on the performance side will rarely come without heat spreaders.
If your not getting enough air to the ram... well back to my first statement... which server u guys taking apart that was built by a 9yr old and didnt follow thermal map?

 

[PliotronX]

I get what you're saying Aig, Cerb's blurb helped me realize why registered FB-DIMM's require them. There is a specific need for them to distribute the heat because each chip is sequentially on so that particular chip is more prone to failure due to heat. This is not so with desktop RAM where the chips are accessed in parallel. IME heat in desktop RAM is dissipated more effectively when naked and used with a fan as that heat is dissipated through the PCB. A thin piece of aluminum or copper with a subpar TIM such as fragtape more often than not acts as a heat blanket. I saw much better results with just a quiet fan in Memtest #5 and #8 with this setup over the years. Those fancy spreaders with heatpipes and sinks are a different story but I was just surprised to see them still in use and now understand why.

The server was a Dell PowerEdge 1950.

 

[Smoblikat]

which servers were u picking apart?
because i'll know to AVOID them.

if ur heat spreaders arent working right... that server was designed by 9 yr olds... or a really old server...

When done right it should be like this:

Which u see is ducted from this:

Also here is another one... u can see having heat spreaders on this type of package wouldnt of made a difference either.. i bet the ram isnt high performance ECC either due to the lack of heat spreaders. Higher Cap ram typically wont have heat spreaders because they are typically not clocked as tight as the lower cap ram. (timings)

in which i can tell you for sure heat spreaders will prevent your memory from failing when running too hot which ECC Registered happen to do when being hammered with work.
ECC Registered runs HOT... hotter then your typical desktop ram... there is never a "excess or waste" when it comes to enterprise... its called Fail Safe.. as enterprise doesnt care about bling... or noise... (u should know how loud Rack Cases can get...)

And ram which is overclocked also run hot and require heat spreader as well to prevent them from failing.
I have both heat spreaders.. BIG massive ones.. aka dominator heat spreaders with a Fan on top of them on my desktop.

If heat spreaders didnt work, they wouldnt be mass used on the higher performance ram... and ECC on the performance side will rarely come without heat spreaders.
If your not getting enough air to the ram... well back to my first statement... which server u guys taking apart that was built by a 9yr old and didnt follow thermal map?

How does that system have 9 DIMM slots per cpu?

I thought it had to be even, that thing is running in like 4.1 channel memory config.

 

[aigomorla]

How does that system have 9 DIMM slots per cpu?

I thought it had to be even, that thing is running in like 4.1 channel memory config.

Gainestown / Westermere / gulftown... had tri channel memory controllers.
3x3 = 9 each cpu.