• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA 461.09 Driver on a 3090 and the secret memory PerfCap

Joined
Jan 2, 2011
Messages
9 (0.00/day)
Something changed with this driver and my ability to overclock the memory on my 3090.
I believe they changed the thermal limit of the memory, reducing it significantly.

With previous drivers, I was able to run my memory at 10350mhz under full stress-test load. With the new driver, I pretty much have to run stock at 9501mhz. Even a slight overclock will cause my card to throttle for Thermal PerfCap reason according to GPUz.

NVIDIA does not allow us to see the memory temperature, nor the memory temperature limit. I believe this is because it runs way too hot, basically at the limit of the spec of the memory, and they want to hide this fact. There is indeed a thermal sensor on the memory though, and there is indeed a thermal limit for the memory. This limit is separate from the GPU thermal limit. This is very easy to prove. My GPU runs at 55C under full load with a thermal limit of 83C. So I am no where near the thermal limit. At stock speeds, GPUZ will show the PerfCap reason as "Pwr". When I overclock the memory, it will eventually slow the whole card down and the PerfCap Reason will change to Thermal. Performance takes a giant nose dive of course when this happens.

This driver has caused me to take a massive performance hit by the stealth change to the secret memory thermal limit. I actually have two different 3090s from two different manufacturers. They both suffer the same fate from this new driver.

I mine ETH with these cards, and this new driver is literally costing me money...

If you are running with a high memory overclock, or hell, even ANY memory overclock, run a stress-test or mine some ETH and see what GPUz says in the "Sensors" tab for PerfCap Reason. I bet you will not be happy.
 
Joined
Oct 15, 2011
Messages
1,920 (0.42/day)
Location
Springfield, Vermont
System Name KHR-1
Processor Ryzen 9 5900X
Motherboard ASRock B550 PG Velocita (UEFI-BIOS P3.40)
Memory 32 GB G.Skill RipJawsV F4-3200C16D-32GVR
Video Card(s) Sapphire Nitro+ Radeon RX 6750 XT
Storage Western Digital Black SN850 1 TB NVMe SSD
Display(s) Alienware AW3423DWF OLED-ASRock PG27Q15R2A (backup)
Case Corsair 275R
Audio Device(s) Technics SA-EX140 receiver with Polk VT60 speakers
Power Supply eVGA Supernova G3 750W
Mouse Logitech G Pro (Hero)
Software Windows 11 Pro x64 23H2
Looks like same BS as the with GPU! On my Turing, I genuinely think in my case, it's because of the laughably low watt limit of my GeForce GTX 1660 Super.

"Pwr"-> That usually is the power limit. For example, with the GeForce GTX 1660 series, this commonly appears to be less than 135W. A lot of those, appear to have the default at 125W.

You don't have to use FurMark or the like to exceed the limit! In normal games, I get "pwr" all the fracking time!
 
Last edited:

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
26,957 (3.71/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Have you tested switching back to the previous driver to rule out any other cause?

Do you have a way to test if it's your VRM or memory temperature that's holding you back?
 
Last edited:
Joined
Oct 22, 2020
Messages
41 (0.03/day)
Processor AMD Ryzen 7 5800X
Motherboard ASUS X570 TUF
Cooling Cougar Aura WC 240mm
Memory 32 GB 3200Mhz CL15
Video Card(s) RTX 3080 TUF Shunt modded - RTX 3060Ti EVGA XC (will shunt mod) - RTX 2080Ti Zotac AMP! Extreme
No issues on my case with VRAM OC changes, I have a 3060Ti, now, I had to use DDU since the newer driver bugged my Hz on my screen, I had tearing even on desktop :( At least is working fine though lol
 
Joined
Jan 2, 2011
Messages
9 (0.00/day)
Confirmed that reverting the driver resolves the problem.

Also, it turns out that the cards will even Thermal PerfCap at STOCK speeds after a few hours of ETH mining. Awesome....

I would love to hear from another 3090 owner. This is horrible.
 
Joined
Oct 15, 2011
Messages
1,920 (0.42/day)
Location
Springfield, Vermont
System Name KHR-1
Processor Ryzen 9 5900X
Motherboard ASRock B550 PG Velocita (UEFI-BIOS P3.40)
Memory 32 GB G.Skill RipJawsV F4-3200C16D-32GVR
Video Card(s) Sapphire Nitro+ Radeon RX 6750 XT
Storage Western Digital Black SN850 1 TB NVMe SSD
Display(s) Alienware AW3423DWF OLED-ASRock PG27Q15R2A (backup)
Case Corsair 275R
Audio Device(s) Technics SA-EX140 receiver with Polk VT60 speakers
Power Supply eVGA Supernova G3 750W
Mouse Logitech G Pro (Hero)
Software Windows 11 Pro x64 23H2
No issues on my case with VRAM OC changes, I have a 3060Ti, now, I had to use DDU since the newer driver bugged my Hz on my screen, I had tearing even on desktop :( At least is working fine though lol
Is that with 461.09?
 
Joined
Jan 2, 2011
Messages
9 (0.00/day)
Have you tested switching back to the previous driver to rule out any other cause?

Do you have a way to test if it's your VRM or memory temperature that's holding you back?
I do not. I am basing the claim that it is likely memory temperature on a few things.

0. GPU-Z shows "PerfCap Reason Thrm" and the performance takes a hit. So we know that something in the card is getting too hot, because the card is saying so. The GPU is cool as a cucumber when this happens.
1. With any driver previous to 461-09 I could leave all settings stock, then overclock the memory to 10350mhz. No voltage changes. Aggressive fan curve that would pretty much keep the fan at 100% under any load at all. Both of my cards could run at these settings for weeks at a time, 24/7, mining ETH (which puts extreme load on the memory)
2. With any driver previous to 461-09, memory speeds higher than 10350 would start to cause performance nose-dives, with GPU-Z reporting PerfCap Reason Thrm.
3. Room temperature, CPU temperature, Case Temperature, GPU Temperature would all stay super cool while mining. Both rigs are extremely well ventilated. None of these temperatures have changed with the new 461-09 driver showing constant PefCap Thrm all the time.
4. With either driver, a laser thermometer shows that the area on the heat plate above memory on top of the card gets really hot, much hotter than anything else that can me measured on the surface of the card. Much hotter than anything that can be read from any internal sensor in the card. I can see 71-72C on the outside of the plate under load. The memory on the bottom is sharing a giant active heatsink/fan with the GPU. Since this card has so much memory, they had to put some of it on top. The top memory has no active cooling. Even the passive cooling is barely that. It is simply a flat metal plate with no fins. This bad design lets the memory on top of the card just sit there and cook.
5. Some research into the issue lead me here: https://www.igorslab.de/en/nvidia-c...-3080-fe-to-mimic-the-hotspot-for-the-gddr6x/
You will see that Igor has some secret NVIDIA tool that actually lets him read the memory temperature, and it is indeed maxing out. This pre-production card that he had could be improved by a few degrees with some thermal tape changes which were rolled in to the production cards. But still, the memory on top of the card is getting extremely hot and has nothing but a flat metal plate to assist with heat mitigation.

I have two separate rigs, each with a different card from a different manufacturer. Same problem is observed with both. The two cards are:
Gigabyte Eagle 3090
MSI TRIO OC 3090
 
Last edited:
Joined
Mar 14, 2014
Messages
1,274 (0.35/day)
Processor i7-4790K 4.6GHz @1.29v
Motherboard ASUS Maximus Hero VII Z97
Cooling Noctua NH-U14S
Memory G. Skill Trident X 2x8GB 2133MHz
Video Card(s) Asus Tuf RTX 3060 V1 FHR (Newegg Shuffle)
Storage OS 120GB Kingston V300, Samsung 850 Pro 512GB , 3TB Hitachi HDD, 2x5TB Toshiba X300, 500GB M.2 @ x2
Display(s) Lenovo y27g 1080p 144Hz
Case Fractal Design Define R4
Audio Device(s) AKG Q701's w/ O2+ODAC (Sounds a little bright)
Power Supply EVGA Supernova G2 850w
Mouse Glorious Model D
Keyboard Rosewill Full Size. Red Switches. Blue Leds. RK-9100xBRE - Hate this. way to big
Software Win10
Benchmark Scores 3DMark FireStrike Score : needs updating
OP, any way you can get more air over the card to test it? A box fan or something?
You're also on GDDR6 without chips on the back of the card. OP has GDDR6X with chips on the back only being passively cooled.
 
Joined
Jan 2, 2011
Messages
9 (0.00/day)
OP, any way you can get more air over the card to test it? A box fan or something?
That is an interesting idea. Based on that idea, I did some testing.

* I was mining ETH continuously through all of these tests to put thermal load on the memory *

Test 1. Stock clocks. It had been thermal throttled all day, then I added the box fan on high, basically replacing my case side panel.
Result: No more thermal throttling

Test 2. Before this new driver, I could run +850mhz memory overclock on this card without thermal throttling. Now I can't even run stock unless I have a box fan on it. Lets test +206mhz WITH the box fan.
Result: Thermal throttling.

Test 3. I grabbed a crappy old aluminum CPU heatsink and set it on top of the card near the memory. No thermal paste. No tape. Just gravity. Box fan still blowing. Card still at +206mhz
Result: No more thermal throttling.

Pictures attached of all three of these tests.

This now proves beyond reasonable doubt that the new driver absolutely does change the secret Memory PerfCap. The question is... why?
I have two guesses:

1. Maybe they are getting too many RMAs from people killing there cards, with the memory giving out due to long periods of high temperature. The current surge of crypto prices probably make for a lot more cards being used 24/7 for mining, like mine. This change saves money in warranty replacements.
2. Maybe a percentage of cards don't handle the obcenely high temperatures very well and have stability issues. They nerf EVERY card with a global driver update, and the problem cards are now fixed.

Test4: No box fan. Case closed up. +150mhz overclock on the memory. Old aluminum CPU heatsink still resting on top of the card, sitting over the memory.
Result: No more thermal throttling.

I guess it is time to find a good heatsink to permanently mount on top of the card. Something with a wide base, and active cooling. If I get this much relief from a random-ass passive heatsink, held on by gravity, and no thermal interface material; A good sized active heatsink with thermal compound should provide tremendous support. Actually, if you look at the attached picture, the heatsink I put on there seems to be almost exactly the right size.

Who though it would be a good idea to put heat generating components on top of the card with no heat mitigation system beyond a flat metal plate? Seriously.... wtf. This is an irresponsible design.
 

Attachments

  • test1.png
    test1.png
    964.2 KB · Views: 1,695
  • test2.png
    test2.png
    970.5 KB · Views: 1,192
  • test3.png
    test3.png
    1 MB · Views: 1,323
  • test3_heatsink.jpg
    test3_heatsink.jpg
    175.2 KB · Views: 1,677
  • heatsink.png
    heatsink.png
    2 MB · Views: 1,808
Last edited:

Sacmac

New Member
Joined
Dec 16, 2020
Messages
1 (0.00/day)
I placed a heatsink 200x69x39mm with no thermal pads on the backplate and scored 17865 on TimeSpy on 456.71 driver. After updating to 461.09, I noticed my Timespy score dropped by 300 points with +110 on core and +600 on mem settings same as before the update.

GPU is PNY 3090 Revel with KFA2 390w bios flashed.
 
Last edited:
Joined
Nov 11, 2016
Messages
3,045 (1.13/day)
System Name The de-ploughminator Mk-II
Processor i7 13700KF
Motherboard MSI Z790 Carbon
Cooling ID-Cooling SE-226-XT + Phanteks T30
Memory 2x16GB G.Skill DDR5 7200Cas34
Video Card(s) Asus RTX4090 TUF
Storage Kingston KC3000 2TB NVME
Display(s) LG OLED CX48"
Case Corsair 5000D Air
Power Supply Corsair HX850
Mouse Razor Viper Ultimate
Keyboard Corsair K75
Software win11
I guess it is time to find a good heatsink to permanently mount on top of the card. Something with a wide base, and active cooling. If I get this much relief from a random-ass passive heatsink, held on by gravity, and no thermal interface material; A good sized active heatsink with thermal compound should provide tremendous support. Actually, if you look at the attached picture, the heatsink I put on there seems to be almost exactly the right size.

Who though it would be a good idea to put heat generating components on top of the card with no heat mitigation system beyond a flat metal plate? Seriously.... wtf. This is an irresponsible design.

Well it's no wonder the GDDR6X on your 3090 is overheating, it's pulling 187W. Under gaming workload the GDDR6X memory only use around 80-110W (MVDDC Power Draw), this is at 10000mhz clock (20Gbps)
gpuz.png


Seems like active cooling is required for the memory modules on the back side of the 3090 when mining, did you measure the temp on the backplate with an IR thernometer ? The backplate on my 3090 reach around 53C max under gaming workload.

I actually gained OC headroom with the 461.09 driver
 
Last edited:
Joined
Jan 11, 2005
Messages
200 (0.03/day)
That is an interesting idea. Based on that idea, I did some testing.

* I was mining ETH continuously through all of these tests to put thermal load on the memory *

Test 1. Stock clocks. It had been thermal throttled all day, then I added the box fan on high, basically replacing my case side panel.
Result: No more thermal throttling

Test 2. Before this new driver, I could run +850mhz memory overclock on this card without thermal throttling. Now I can't even run stock unless I have a box fan on it. Lets test +206mhz WITH the box fan.
Result: Thermal throttling.

Test 3. I grabbed a crappy old aluminum CPU heatsink and set it on top of the card near the memory. No thermal paste. No tape. Just gravity. Box fan still blowing. Card still at +206mhz
Result: No more thermal throttling.

Pictures attached of all three of these tests.

This now proves beyond reasonable doubt that the new driver absolutely does change the secret Memory PerfCap. The question is... why?
I have two guesses:

1. Maybe they are getting too many RMAs from people killing there cards, with the memory giving out due to long periods of high temperature. The current surge of crypto prices probably make for a lot more cards being used 24/7 for mining, like mine. This change saves money in warranty replacements.
2. Maybe a percentage of cards don't handle the obcenely high temperatures very well and have stability issues. They nerf EVERY card with a global driver update, and the problem cards are now fixed.

Test4: No box fan. Case closed up. +150mhz overclock on the memory. Old aluminum CPU heatsink still resting on top of the card, sitting over the memory.
Result: No more thermal throttling.

I guess it is time to find a good heatsink to permanently mount on top of the card. Something with a wide base, and active cooling. If I get this much relief from a random-ass passive heatsink, held on by gravity, and no thermal interface material; A good sized active heatsink with thermal compound should provide tremendous support. Actually, if you look at the attached picture, the heatsink I put on there seems to be almost exactly the right size.

Who though it would be a good idea to put heat generating components on top of the card with no heat mitigation system beyond a flat metal plate? Seriously.... wtf. This is an irresponsible design.

In addition to what it said above, Your memory power draw and your SRC (Input power plane source power) are both way past their internal power limits when you are getting the thermal flag.

This is based on the Gigabyte 3090 OC Bios dump.

The SRC max power limit is 175W for SRC1 and 2. You are past that.

The MVDDC Power Limit is 103.125W and you are WAY past that.

What is interesting is that the SRC is even at its power limit to begin with. However I've seen this before. Most likely, because MVDDC's limit is exceeded, SRC, which is the power balancing shunt, looks like it leeches power onto its read then reaches its throttle point and signals a throttling flag. (it's these reasons why when doing shunt mods, the SRC shunt must be modded!). And as I suspected, the 8 pin power limit max is 123W, but you're at 151W, which means my theory I found on the FE is correct--the 8 pins use the same internal power limit as the SRC (there are # of SRC power limits equal to the # of 8 pin connectors) and their own power limits seem to be ignored.

For the MSI 3090, the 8 pin power limit is 121W, and the VRAM power limit is even lower--100W. SRC is 150W default and 175W max just like before, but since this is a three 8 pin card, there are three SRC power limits (all set to the same value).

Either way, the PWR flag is from the SRC hitting its max limit (which it's hitting due to VRAM *AND* 8 pins exceeding their internal limits)
That doesn't explain the thermal throttling behavior though. I do see you're at 178W SRC when you have that flag and 175W when you are not...



gigabyte_dump.jpg
 
Joined
Jan 2, 2011
Messages
9 (0.00/day)
I added three active heat-pipe M2 coolers over the memory. They are instantly getting very hot, which means they are transferring heat off of the card quite nicely. I wish I could read the memory temperature for the card, but ya know.... that is privileged information that NVIDIA does not think we should see.

The card no longer Thermal Caps while mining at stock speeds. I seem to be able to stay un-capped with a 200mhz memory overclock now too. I also flashed the EAGLE OC BIOS onto this Eagle non-oc card, so now I can raise the power limit to 110%, because I obviously need more heat :)

Overall, I wish these heatsinks more more effective than they are. But they seem to be much better than nothing.

Well it's no wonder the GDDR6X on your 3090 is overheating, it's pulling 187W. Under gaming workload the GDDR6X memory only use around 80-110W (MVDDC Power Draw), this is at 10000mhz clock (20Gbps)
When I play COD with this card and have the memory clocked to only 9700, my MVDDC will float around 135W. This is with max settings, 4k/120/VRR + RT
 

Attachments

  • mem_hs.jpg
    mem_hs.jpg
    250.1 KB · Views: 1,336
  • mem_hs2.jpg
    mem_hs2.jpg
    486.7 KB · Views: 1,374
  • mem_hs3.png
    mem_hs3.png
    34.1 KB · Views: 1,293
Joined
May 12, 2006
Messages
1,554 (0.24/day)
Location
The Gulag Casino
System Name ROG 7900X3d By purecain
Processor AMD Ryzen 7 7900X3D
Motherboard ASUS Crosshair X670E Hero
Cooling Noctua NH U12A
Memory 64Gb G.Skill Trident Z5 neo RGB 6400@6000mhz@1.41v
Video Card(s) Aorus RTX4090 Extreme Waterforce
Storage 990Pro2Tb-1TbSamsung Evo M.2/ 2TbSamsung QVO/ 1TbSamsung Evo780/ 120gbKingston Now
Display(s) LG 65UN85006LA 65" Smart 4K Ultra HD HDR LED TV
Case Thermaltake CoreX71 Limited Edition Etched Tempered Glass Door
Audio Device(s) On board/NIcomplete audio 6
Power Supply Seasonic FOCUS 1000w 80+
Mouse M65 RGB Elite
Keyboard K95 RGB Platinum
Software Windows11pro
Benchmark Scores [url=https://valid.x86.fr/gtle1y][img]https://valid.x86.fr/cache/banner/gtle1y-6.png[/img][/url]
reads thread, takes a look at dead 3090... spits dummy 30ft...

Interesting thread... Keep an eye on the behavior of your cards at low power with new drivers. :toast:
 

wolololo34

New Member
Joined
Jan 18, 2021
Messages
8 (0.01/day)
IMG_0004.JPG


Filed down the stand-offs of an old rx480 sapphire heatsink and attached it to the top

85% power/ 1200mhz core locked at 725 / +1200hmz memory/ 80% fan = 120 hash

can get 95%power / 1200hmz locked at 725/ +1500mhz memory / 100% fan = 124 hash

I am mining with this card until I can afford it!

I hope the drivers get better for this card because my vega 64 LC (now in a water loop) seems to have better frame timing and feels smoother.
 

natx808

New Member
Joined
Jan 20, 2021
Messages
1 (0.00/day)
Thanks for posting this. I moved my aorus 3090 down a slot to make room for a passive heatsink but still hitting thermal throttle. i was able to squeeze a few additional mh out of it though. I know it's more effort, but I assume you have seen the vids of the 3080 eagle where they replace pads with thermal grizzly rated at W / mK 8.0 ? supposedly the pads in the gigabyte 3080 eagle are junk. so I wonder if all 3000 series gigabyte cards pads are junk....haven't inspected mine yet but will report back.

i also have a gigabyte 3080 and it doesn't even have thermal pads on the back of the memory modules and the backplate. can't get that past 45% power before it throttles. have pads on order for that card. will compliment with heatsinks if still throttling after doing the pads.
 

wolololo34

New Member
Joined
Jan 18, 2021
Messages
8 (0.01/day)
Thanks for posting this. I moved my aorus 3090 down a slot to make room for a passive heatsink but still hitting thermal throttle. i was able to squeeze a few additional mh out of it though. I know it's more effort, but I assume you have seen the vids of the 3080 eagle where they replace pads with thermal grizzly rated at W / mK 8.0 ? supposedly the pads in the gigabyte 3080 eagle are junk. so I wonder if all 3000 series gigabyte cards pads are junk....haven't inspected mine yet but will report back.

i also have a gigabyte 3080 and it doesn't even have thermal pads on the back of the memory modules and the backplate. can't get that past 45% power before it throttles. have pads on order for that card. will compliment with heatsinks if still throttling after doing the pads.
I think they need to completely rethink how heatsinks work. You would ideally want the pcb to be in a heatsink sandwich with fans on both sides. Because of this I wonder if one sided block water cooling actually works?

I have now put a 1.5mm thermal pad to keep the heatsink in place, it works really well. I think this GPU needs to be in an open case with fans blowing onto it. Honestly I'm surprised that they put the card (both this model and the 3090 in general) into production, this design should not have passed QC. I mean I have dealt with Vega hotspots reaching 110 but at least AMD are honest enough to allow GPU-Z to read this.


I'm sure the pads are junk but I don't want to void my warranty* , getting at the backplate requires complete disassembly.


*this is contested, but breaking seals allows companies to drag their feet with claims etc
 
Last edited:
Joined
Feb 7, 2020
Messages
16 (0.01/day)
I actually had great success with this setup, but sadly had to give it up because the far edge of the fan was pushing down on the card. I initially thought there was a clean gap between the card and the bottom fan. Fan speed dropped from mid-80% to about 60%.

I might try the looking for a low profile heatsink to the top next.

MSI advertises this uses special thermal pads for the memory and separate core pipes to take heat away from the memory modules and this Suprim X has the largest heatsink yet. And it's struggling, too.




 
Last edited:

wolololo34

New Member
Joined
Jan 18, 2021
Messages
8 (0.01/day)
Ha. Whats VIP cooling? Does it serve you a pina colada every 15minutes.
Those Gigabyte 3080's without thermal pads are not VIP, no pina colada for them.

Tbh I'd advise running the gpu outside of the case.
  1. Pci riser (those thermaltake 1m ones will do)
  2. Noctua 200mm fan blowing air back to front (opposite of what a case would do)
  3. An old gpu heatsink with a 1.5mm thermal pad on-top with a static 120mm fan (blowing towards with heatsink).
I'm now getting:

+ 1200mem / 70% fan / 120 hash

+ 1500mem /80% fan / 123 hash

I could sleep next to it if i wanted to, really quiet.
 
Last edited:
Joined
Aug 16, 2004
Messages
3,275 (0.46/day)
Location
Sunny California
Processor Intel Core i9 13900KF
Motherboard Asus ROG Maximus Z690 Hero EVA Edition
Cooling Asus Ryujin II 360 EVA Edition
Memory 4x16GBs DDR5 6800MHz G.Skill Trident Z5 Neo Series
Video Card(s) Zotac RTX 4090 AMP Extreme Airo
Storage 2TB Samsung 980 Pro OS - 4TB Nextorage G Series Games - 8TBs WD Black Storage
Display(s) LG C2 OLED 42" 4K 120Hz HDR G-Sync enabled TV
Case Asus ROG Helios EVA Edition
Audio Device(s) Denon AVR-S910W - 7.1 Klipsch Dolby ATMOS Speaker Setup - Audeze Maxwell
Power Supply EVGA Supernova G2 1300W
Mouse Asus ROG Keris EVA Edition - Asus ROG Scabbard II EVA Edition
Keyboard Asus ROG Strix Scope EVA Edition
VR HMD Samsung Odyssey VR
Software Windows 11 Pro 64bit
I added three active heat-pipe M2 coolers over the memory. They are instantly getting very hot, which means they are transferring heat off of the card quite nicely. I wish I could read the memory temperature for the card, but ya know.... that is privileged information that NVIDIA does not think we should see.

The card no longer Thermal Caps while mining at stock speeds. I seem to be able to stay un-capped with a 200mhz memory overclock now too. I also flashed the EAGLE OC BIOS onto this Eagle non-oc card, so now I can raise the power limit to 110%, because I obviously need more heat :)

Overall, I wish these heatsinks more more effective than they are. But they seem to be much better than nothing.


When I play COD with this card and have the memory clocked to only 9700, my MVDDC will float around 135W. This is with max settings, 4k/120/VRR + RT

Interesting find, do you think those heat sinks would help an FE card? Where did you buy the sinks? Thanks for reporting this odd behavior.
 
Joined
Oct 17, 2012
Messages
9,781 (2.34/day)
Location
Massachusetts
System Name Americas cure is the death of Social Justice & Political Correctness
Processor i7-11700K
Motherboard Asrock Z590 Extreme wifi 6E
Cooling Noctua NH-U12A
Memory 32GB Corsair RGB fancy boi 5000
Video Card(s) RTX 3090 Reference
Storage Samsung 970 Evo 1Tb + Samsung 970 Evo 500Gb
Display(s) Dell - 27" LED QHD G-SYNC x2
Case Fractal Design Meshify-C
Audio Device(s) on board
Power Supply Seasonic Focus+ Gold 1000 Watt
Mouse Logitech G502 spectrum
Keyboard AZIO MGK-1 RGB (Kaith Blue)
Software Win 10 Professional 64 bit
Benchmark Scores the MLGeesiest
Where did you buy the sinks?

on amazon, you can find an assortment of heatsinks, ones intended for use with M.2, or chipset cooling. THESE can be adjusted to be positioned the way thats most convenient for your set up. but there are many different ones.

i used to order custom made fans of pretty good quality from a guy on ebay. he custom 3D printed them, & used great motors, all the way down to super tiny fans. if you search RAM , chipset, or VRM cooling , im sure you'll find a heatsink that you can use.

This ones a nice small full copper heat pipe design. its pricey due to the copper, but thats normal

ferd.PNGCapture.PNG
 
Last edited:
Joined
Feb 7, 2020
Messages
16 (0.01/day)
Ha. Whats VIP cooling? Does it serve you a pina colada every 15minutes.
Those Gigabyte 3080's without thermal pads are not VIP, no pina colada for them.

Tbh I'd advise running the gpu outside of the case.
  1. Pci riser (those thermaltake 1m ones will do)
  2. Noctua 200mm fan blowing air back to front (opposite of what a case would do)
  3. An old gpu heatsink with a 1.5mm thermal pad on-top with a static 120mm fan (blowing towards with heatsink).
I'm now getting:

+ 1200mem / 70% fan / 120 hash

+ 1500mem /80% fan / 123 hash

I could sleep next to it if i wanted to, really quiet.
I saw a Reddit post about the Gigabyte thermal pads disintegrating in a couple of weeks. Not VIP
 

wolololo34

New Member
Joined
Jan 18, 2021
Messages
8 (0.01/day)
I saw a Reddit post about the Gigabyte thermal pads disintegrating in a couple of weeks. Not VIP
Could you send me the link to that reddit post?

Yeah, I don't look forward to having to replace the thermal pads for this card, especially considering they are probably the spongy type of pad. Measuring the compression of spongy pads is a nightmare when converting to normal good pads. And this card has a lot of pads.
 
Joined
Feb 7, 2020
Messages
16 (0.01/day)
Could you send me the link to that reddit post?

Yeah, I don't look forward to having to replace the thermal pads for this card, especially considering they are probably the spongy type of pad. Measuring the compression of spongy pads is a nightmare when converting to normal good pads. And this card has a lot of pads.
This one:

https://www.reddit.com/r/EtherMining/comments/ky37id
The Vision one appeared to need pads of different thickness.
 
Top