How can graphic card memories overheat?

[ultron]

http://www.anandtech.com/show/9883/gddr5x-standard-jedec-new-gpu-memory-14-gbps

It sounds like a very dumb question but i don't have understand one thing:

According to this chart each 7 Gbps GDDR5 module consumes approximetly 2.75 watt. A chip that uses this amount of energy shouldn't overheat easily. But on the graphics cards without proper cooling solution, users have to deal memory overheat related problems. I can't understand how it happens.

a/For some reason the heat can't transfered to outside the chip properly(I heard some Intel CPUs has this kind of problems, some users delid their cpus)

b/it's about how the memory modules works. The numbers on the charts are average values, the chip heats instantaneously etc. etc.

Could someone explain this fact with layman's terms.

Sorry for my english.

 

[Silverforce11]

This is why graphics cards that are open air designs, can leave the GDDR5 chips bare. They will run cool just from passive airflow.

Things that normally overheat are VRM, Mosfets.

 

[Magic Carpet]

I wonder, how hot the mem chips get from overclocking? Anybody got a temp gun?

 

[KaRLiToS]

This is why graphics cards that are open air designs, can leave the GDDR5 chips bare. They will run cool just from passive airflow.

Things that normally overheat are VRM, Mosfets.

^^ This

• VRM, Mosfets
• GPU chip (die)

 

[ultron]

This is why graphics cards that are open air designs, can leave the GDDR5 chips bare. They will run cool just from passive airflow.

Things that normally overheat are VRM, Mosfets.

But a lot people suffer from VRAM overheating. Why this happens?

And the second question: Are R9 290X's memory modules and 960's memory chips produced in same manufacturing process. Because 290X's each module consumes 1.875 watt and 960's memory chips consume 2.5 watt. Normally chips' power consumptions have to effect from frequency exponentially. (This is the reason how overclock cause power consumption increases.) But even %40 increase on the speed,the power usage goes up only %33.

Does chip capacity affect the power usage?

And the last question: R9 290X has relatively slow memory frequency. 390X has better on this subject. But why it doesn't use 7 Gb/s memories. Is the power consumption the biggest problem? It doesn't seem like as far as i learned from the charts.

 

[MrTeal]

On their own, memory modules aren't likely to overheat. As you say, a few watts isn't much. Generally GPU memory cooling is a bit like DDR memory cooling; it's nice to look at.
However, you can create scenarios where memory can get quite hot. If there isn't much airflow over the chips themselves and they don't have a thermal path into the heatink, the chips will dissipate heat into the PCB. Normally not a big problem, but to move heat off the chip you still need a temperature delta; the chips are going to be hotter than the PCB.

Looking at the image, the delta between the chips and the board around them isn't huge. I'd guess ~10C. However, when the PCB is 90C, without additional cooling on them the GDDR5 has to be higher than that. They might not be high draw devices, but if you stick them in a place where everything else is hot, expect them to be the same.

 

[Silverforce11]

Titan X is a special case, because it has vram chips on both sides of the PCB, going for 12GB. This means the already hot PCB has to get heat from both GDDR5 chips on either side, rather than one side (the rear usually) acting as a dissipation surface.

Since Titan X aren't dropping dead, we can assume good GDDR5 can operate beyond 100C...

 

[IEC]

Unless you have zero passive airflow or are going for world record overclocks... you don't need anything extra on GDDR5.

 

[CakeMonster]

Not all errors from memory overclocking needs to be from overheating necessarily.

 

[Magic Carpet]

Anyway, so what's the operating range of those GDDR5 mem chips? up to 100C, 125C, 150C?

Datasheets please.

 

[KaRLiToS]

https://www.micron.com/~/media/documents/products/data-sheet/dram/gddr5/4gb_gddr5_sgram_brief.pdf

• Operating temperature
– Commercial (0°C ≤ TC ≤ +95°C)

 

[Vellinious]

The cooler you keep it, the better it'll run. There are numerous examples of it. People going from air cooling on their Maxwell cards to watercooling, and being able to clock their memory higher. The cooler, the better.

On Maxwell, Samsung memory was the best, and had a tendency to clock higher than the others. Elpida was also very good, but would heat up quite a bit, and run worse as they got hotter. Same with Hynix, except they wouldn't clock as high. On Elpida, especially, if your cooling was adequate, you could see clocks as high as 2200 base clock. That's a pretty significant increase.

For instance, on like settings in the custom bios, both of my 970s would hit around 2055 on air, but increasing it any more than that would cause artifacts and crashes. Once under water, they'd both easily hit 2100. The one card was fantastic, and would clock as high as 2180 for benchmarking runs. Good stuff.

 

[StrangerGuy]

Unless you have zero passive airflow or are going for world record overclocks... you don't need anything extra on GDDR5.

How dare those stupid EEs design and build things that don't work right when users push them way out of spec.

 

[LTC8K6]

How dare those stupid EEs design and build things that don't work right when users push them way out of spec.

How dare those users use the overclocking software they were given by the manufacturer!

 

[Creig]

Since Titan X aren't dropping dead, we can assume good GDDR5 can operate beyond 100C...

Can GDDR5 operate above maximum recommended temperatures ? Sure. But you risk the possibility of memory related errors and reduced expected lifespan. It'll be interesting to see if Titan X cards suffer from increasing failure rates as time goes on due to running their memory at temps beyond the recommended maximum specification.

 

[Face2Face]

http://www.anandtech.com/show/9883/gddr5x-standard-jedec-new-gpu-memory-14-gbps

It sounds like a very dumb question but i don't have understand one thing:

According to this chart each 7 Gbps GDDR5 module consumes approximetly 2.75 watt. A chip that uses this amount of energy shouldn't overheat easily. But on the graphics cards without proper cooling solution, users have to deal memory overheat related problems. I can't understand how it happens.

a/For some reason the heat can't transfered to outside the chip properly(I heard some Intel CPUs has this kind of problems, some users delid their cpus)

b/it's about how the memory modules works. The numbers on the charts are average values, the chip heats instantaneously etc. etc.

Could someone explain this fact with layman's terms.

Sorry for my english.

Who has been complaining about VRAM overheating problems? There are only a handful of cards that even allow for VRAM temperature monitoring. Of course they can use methods such as a FLIR or something similar, but can you send us some links to the users that are complaining about this?

 

[ultron]

Who has been complaining about VRAM overheating problems? There are only a handful of cards that even allow for VRAM temperature monitoring. Of course they can use methods such as a FLIR or something similar, but can you send us some links to the users that are complaining about this?

Most of the people have to deal with this kind of artifacts, and some sources claimed that these issues caused by bad cooling on VRAM/dead VRAM/overvoltage on VRAM:

https://www.youtube.com/watch?v=Fntj4v-fmmI

 

[ultron]

And there's another thing. As i learned from the article GDDR5X can make HBM2 be unnecessary. Because HBM on Fury X doesn't scale very well:

Theoretical BW/Real BW ratio is not OK(%70) in comparision with GDDR5's on R9 390X(%78)