News Copper Plate Mod Reduces RTX 3080 GDDR6X Memory Temps by 25 Degrees

InvalidError

Titan
Moderator
Someone publishes a piece about using copper shims, someone else makes a custom-fit copper shim that does fundamentally the same thing in a tidier form factor.

Since the DRAM packages are plastic/epoxy at 1-2W a pop, they could have used steel or aluminum and achieved pretty much the same cooling performance for a fraction of the material cost.
 

VforV

Respectable
BANNED
Oct 9, 2019
578
287
2,270
Nvidia doing pathetic things as always, charging the highest prices for their GPUs only to make Ampere run extremely hot and power hungry for that +5-10% (if any) over RDNA2 to claim supremacy.

I don't see any aricles on RDNA2 about needing extra hardware mods after you bought it to lower it's horribly high temps, why? Because it does not have this problem.

RX 6950X XT refresh will one again prove it's a more efficient high end GPU vs that laughable 3090 Ti and I really can't wait for RDNA3 to feed nvidia a little bit of humble pie, although I doubt they will admit that even if they lose the gaming crown... meh.
 

watzupken

Reputable
Mar 16, 2020
1,007
507
6,070
Nvidia doing pathetic things as always, charging the highest prices for their GPUs only to make Ampere run extremely hot and power hungry for that +5-10% (if any) over RDNA2 to claim supremacy.

I don't see any aricles on RDNA2 about needing extra hardware mods after you bought it to lower it's horribly high temps, why? Because it does not have this problem.

RX 6950X XT refresh will one again prove it's a more efficient high end GPU vs that laughable 3090 Ti and I really can't wait for RDNA3 to feed nvidia a little bit of humble pie, although I doubt they will admit that even if they lose the gaming crown... meh.
Objectively, I feel the problem with Nvidia’s high end Ampere line up is mostly due to the power hungry GDDR6X memory. It is faster than normal GDDR6, but at the expense of power and therefore, heat output. To get a sense how power hungry it is, you just have to look at the difference between a RTX 3070 and 3070 Ti.

It is actually not all bright and sunny on AMD’s RDNA2 when it comes to high heat output. After trying out both the RX 6800 XT and 6600 XT, I generally find that the GPU hotspot temps can get pretty high. The delta between the skin vs hotspot temp can be between 15 to 25 degrees celcius. Whereas on Apmere cards, my testing on the RTX 3070 Ti and 3080 shows a smaller delta between skin and hotspot temp of 10 degrees celcius and below.
 

InvalidError

Titan
Moderator
Objectively, I feel the problem with Nvidia’s high end Ampere line up is mostly due to the power hungry GDDR6X memory. It is faster than normal GDDR6, but at the expense of power and therefore, heat output.
The GDDR6X memory only accounts for about 50W, a relatively small fraction of the 400+W peak board power. Nowhere near 'most' power. Most of the power comes from pushing the GPU die to its limits. The closer to limits you get, the worse power efficiency gets.
 
  • Like
Reactions: VforV

watzupken

Reputable
Mar 16, 2020
1,007
507
6,070
The GDDR6X memory only accounts for about 50W, a relatively small fraction of the 400+W peak board power. Nowhere near 'most' power. Most of the power comes from pushing the GPU die to its limits. The closer to limits you get, the worse power efficiency gets.
I actually don’t think it is just 50W. If you look at the difference in power consumption between the RTX 3070 and 3070 Ti, that slight increase in CUDA cores is not going to result in a 70W difference in power on top of the power required by GDDR6. That is a 220W part, “upgraded” to 290W. When you move up to RTX 3090, you have an extra 16 pieces of GDDR6X, running at a higher clock, which will demand more power. As we move up again to the RTX 3090 Ti, the number of memory chips have dropped, but pushed even harder for both the GPU and VRAM. So no surprises the power consumption went up. After all, I believe the sweet spot clockspeed of Ampere sits close to 1.9 to 2. Once you push it beyond 2 GHz, it starts to pull in a lot more power.
 

VforV

Respectable
BANNED
Oct 9, 2019
578
287
2,270
Objectively, I feel the problem with Nvidia’s high end Ampere line up is mostly due to the power hungry GDDR6X memory. It is faster than normal GDDR6, but at the expense of power and therefore, heat output. To get a sense how power hungry it is, you just have to look at the difference between a RTX 3070 and 3070 Ti.

It is actually not all bright and sunny on AMD’s RDNA2 when it comes to high heat output. After trying out both the RX 6800 XT and 6600 XT, I generally find that the GPU hotspot temps can get pretty high. The delta between the skin vs hotspot temp can be between 15 to 25 degrees celcius. Whereas on Apmere cards, my testing on the RTX 3070 Ti and 3080 shows a smaller delta between skin and hotspot temp of 10 degrees celcius and below.
So it's worse for +15 degrees on the hot spot vs Ampere?

Sure, let's say I agree, but the fact remains none of the RDNA2 get to 100 degrees or more like Ampere does on their most expensive GPUs they have... not to mention power consumption, not to mention price...

Ampere is horrible in the high end. End of Story.
 
May 3, 2022
1
0
10
Buyer Beware

I’m posting my experience with CoolMyGPU.com so that potential buyers will be aware of the risks associated with purchasing their products.

I have an EVGA RTX 3090 FTW3. I purchased their compatible front plate and bracket backplate.

I have swapped pads numerous time on my GPU’s as I have a mining farm, and I’ve become quite proficient at it.

This particular 3090 had Gelid Extreme pads on the VRam and the power modules (1 mm to replace the oem thermal puffy). The mosfet pads are stock. With the Gelids, my mem temps fluctuated between 86C-88C at 72F ambient temp.

I installed the CoolMyGPU plates strictly following their instructional video.

Upon testing, I discovered that my Temps jumped to 108C. My first guess was that contact was not adequate. I compressed the card to prove my hypothesis that poor contact was made with these plates, dropping my temperatures to 102C.

I opened the GPU to troubleshoot and Found nothing impeding contact, so I reapplied thermal paste, and tried it again.

Temperatures were still over 106C! After opening this card up about 6 times to test different methods of remedying contact issues, I discovered that the only way to get the temps under 100C with this was to add 1mm Gelid pads on top of each plate. The lowest temp accomplished with this configuration was 98C.

After days of troubleshooting, and over $50 in thermal paste and pads, I finally decided to remove the plates all together and put Gelid Extremes back into the card.

This immediately brought my temps back down to 86C-88C.

It was clear that the plates provided by CoolMyGpu.com were either the incorrect plates, or they were milled too thin.

I reached out to CoolMyGPU via email. No response.

I reached out again via email. No response.

I then reached out via discord, and was able to make contact with their customer support after a couple of days.

This person informed me that the owner was the only person who could approve refunds, and that he reached out to him about my order.

A week later, still no response. Finally, after informing support that I was sick of being jerked around on this, he referred me to the owner’s Discord.(NoSo-CMG)

I have sent multiple messages to the owner about this, including a testimonial from another customer that his plate was too thick, and have yet to get a response.

I’ve looked these guys up, it does not look like they are an officially registered business in the state of Georgia, and their shipping address is a little townhouse.

Any legitimate business would provide refunds or at least try to fix the issue to help customers. I hope potential buyers take this information into account when making their buying decision.