Hardware & Technical Latest Nvidia Ampere Rumours

Messing with my memory a bit more and have it back to 1800/3600. A few timings, especially relating to writes, seem to be the main limiting factor.

What do you mean by "over the bare heatsink"? Does it have a heatsink on the backplate, or you put the fans sideway of the card?

I pulled the stock fans and shroud off entirely, exposing the heatsink fin stack. A slot further down (I can't mount it flush cause the card plus bare heatsink is almost 2.5 slots thick), I have this holding three of these, blowing at the card.

If I flash higher power firmware, I'll probably mount three Noctua 92mm fans directly to the GPU heatsink.

BTW as an easy upgrade, would it make sense for me to remove the backplate and put thermal pads where the VRAM is then putting back the backplate?

If I recall correctly the MSI 3080 Suprim already has pads behind the memory and VRM, so there probably isn't much to be gained by swapping them out. TIM performance here is less critical than with the interface to the main cooler.

That said, if you don't already have pads on your backplate, adding them should knock off at least a few degrees, even without underfil, but that will depend on the gap to the backplate, what the backplate is made of, and if there is any airflow over it. A significant portion of the GDDR6X's cooling comes from the PCB it's mounted to, cool that and you cool the memory. Good putty is ideal, but there are some pretty soft pads in the 3-6W/mK range that will also work well enough. Ideally you want to use something compressible/conformable (rated for at least 30% elongation and no more than ~40 on the Shore 00 Hardness scale, if possible), due to all the surface mount components on the back of the PCB. I recommend cutting small squares of about 15mm*15mm, of half a mm thicker than the distance you want it to fill, and putting one behind each memory IC, then a small strip behind the VRM power stages. Using too much surface area is redundant and makes it harder to concentrate pressure on the pads.
 
Messing with my memory a bit more and have it back to 1800/3600. A few timings, especially relating to writes, seem to be the main limiting factor.



I pulled the stock fans and shroud off entirely, exposing the heatsink fin stack. A slot further down (I can't mount it flush cause the card plus bare heatsink is almost 2.5 slots thick), I have this holding three of these, blowing at the card.

If I flash higher power firmware, I'll probably mount three Noctua 92mm fans directly to the GPU heatsink.



If I recall correctly the MSI 3080 Suprim already has pads behind the memory and VRM, so there probably isn't much to be gained by swapping them out. TIM performance here is less critical than with the interface to the main cooler.

That said, if you don't already have pads on your backplate, adding them should knock off at least a few degrees, even without underfil, but that will depend on the gap to the backplate, what the backplate is made of, and if there is any airflow over it. A significant portion of the GDDR6X's cooling comes from the PCB it's mounted to, cool that and you cool the memory. Good putty is ideal, but there are some pretty soft pads in the 3-6W/mK range that will also work well enough. Ideally you want to use something compressible/conformable (rated for at least 30% elongation and no more than ~40 on the Shore 00 Hardness scale, if possible), due to all the surface mount components on the back of the PCB. I recommend cutting small squares of about 15mm*15mm, of half a mm thicker than the distance you want it to fill, and putting one behind each memory IC, then a small strip behind the VRM power stages. Using too much surface area is redundant and makes it harder to concentrate pressure on the pads.

Wow, never seen such a solution to cool the card!

I have to check, but I think only the 3090 (and the 3080ti?) Suprim X have pads behind the backplate, so I might take a look into it!
 
Can't even source a current gen RTX, but rumors are the next gen cards (Lovelace (LOL), and RDNA 3) are coming in October '22.
 
Can't even source a current gen RTX, but rumors are the next gen cards (Lovelace (LOL), and RDNA 3) are coming in October '22.
The question is whether they will arrive in Oct with enough stock at something approaching MSRP.
 
Can't even source a current gen RTX, but rumors are the next gen cards (Lovelace (LOL), and RDNA 3) are coming in October '22.

Everyone is saying mid-2022 for solid availability of Ampere (probably refreshes) and RDNA2, probably helped by Intel Xe offering more options. Wouldn't surprise me to see a Q4 2022 launch of Lovelace (why the "lol"?) and RDNA3, but I'm not expecting supply to be great on launch, or prices to ever be good.

The question is whether they will arrive in Oct with enough stock at something approaching MSRP.

I strongly suspect the supply situation will have improved by then, but I don't think it will be quite normal. Most newly announced fab capacity won't be online until 2023-24 and there are still unresolved supply chain issues with raw materials and components.
 
Given how much thermal headroom I had with my modded Arorus 3080 Master, I decided to try higher power limit firmware on it. Found a 450w one that still supported ReBAR for another Aorus part with the same VRM controller and device ID. Firmware seemed to work at first, but it quickly became clear that only the reported, not actual, power consumption had increased...clock speeds were down and temperatures hadn't increased at all.

Evidently the power limit on this rev 1 part with only two 8-pin power connectors is limited in hardware. Will probably need a shunt mod to actually increase the power limit...which I'm not going to bother with just to get another 25-50MHz on an air cooled part.
 
Decided to go ahead and wire my 92mm fans directly to the heatsink on my Gigabyte 3080:
7rL1gTZ.jpg


5-10C improvement over having them on the slot-mounted cooler with no shroud.
 
At 60% fan speed in 26C ambient room temp at maximum power limited GPU load (~360w):

GPU: 62C
GPU hotspot: 73C
GDDR6X junction: 84C

100% fan speed knocks off another ~10C.
This is very similar to my stock Suprim X.
BTW checked and I was wrong, it seems to have pads behind the backplate too.
 
This is very similar to my stock Suprim X.
BTW checked and I was wrong, it seems to have pads behind the backplate too.

It's not far off stock for this part, either, except for memory junction temp, which started slightly worse and wound up at 110C as the original pads degraded. Main difference is noise level. The aftermarket fans only need to spin about half as fast to produce equivalent temps.

Did you replace the pads on the front of your card? MSI seems to be afflicted by the same sort of issues my Gigabyte had, unless they had already switched to different pads by the time your sample was made.
 
It's not far off stock for this part, either, except for memory junction temp, which started slightly worse and wound up at 110C as the original pads degraded. Main difference is noise level. The aftermarket fans only need to spin about half as fast to produce equivalent temps.

Did you replace the pads on the front of your card? MSI seems to be afflicted by the same sort of issues my Gigabyte had, unless they had already switched to different pads by the time your sample was made.

I bought mine in January, so probably not.
At 200W mining load the memory junction temp is about 96 +-2 degrees depending on ambient, not sure how it fares to newer cards. Pretty sure better pads would make a big difference though, and perhaps changing them on my backplate should be simple enough to do...
 
I bought mine in January, so probably not.
At 200W mining load the memory junction temp is about 96 +-2 degrees depending on ambient, not sure how it fares to newer cards. Pretty sure better pads would make a big difference though, and perhaps changing them on my backplate should be simple enough to do...

New pads on the front would help a lot, and you'd probably want to replace the ones on the back at the same time. Leaving the existing ones to bleed silicone oil onto the new ones probably wouldn't be good for their longevity.
 
New pads on the front would help a lot, and you'd probably want to replace the ones on the back at the same time. Leaving the existing ones to bleed silicone oil onto the new ones probably wouldn't be good for their longevity.

To be honest I'm not really a DIY guy so I have close to zero experience in tinkering with these stuff. Would you still advise to essentially fix what isn't broken?
 
You're best off leaving it alone unless the memory is regularly cracking 100C.
My more aggressive mining profile does 90-91 MH/s @ 207W, at 65% fan speed the memory junction tops out at 97-98C depending on ambient temperature. Interestingly ambient temperature has a huge impact on the GPU die temps, but memory temp-elasticity is low.

99% of the time however I'm running a 199W 85 MH/s profile, that usually runs at 96C on the memory. I tried even lower settings but it felt like it doesn't decrease temps much and efficiency dropped a lot.

At this point I have recovered my upgrade costs and more, so fully OK with this. Planning to move some of my stock market capital to crypto however. :)
 
You're best off leaving it alone unless the memory is regularly cracking 100C.
Off topic, but google did not yield a 1:1 answer... I bought a refurbished PC for media purposes, mainly file sharing (old cartoons for the kids otherwise not available on any streaming service) - if network discovery is off on all devices, is that enough to prevent any worms to spread on my home network?
 
Off topic, but google did not yield a 1:1 answer... I bought a refurbished PC for media purposes, mainly file sharing (old cartoons for the kids otherwise not available on any streaming service) - if network discovery is off on all devices, is that enough to prevent any worms to spread on my home network?
Nope
 
Off topic, but google did not yield a 1:1 answer... I bought a refurbished PC for media purposes, mainly file sharing (old cartoons for the kids otherwise not available on any streaming service) - if network discovery is off on all devices, is that enough to prevent any worms to spread on my home network?

No.

If you haven't done so, you should wipe the system and put a fresh OS install on it.
 
Back
Top Bottom