Basics RTX 3080/3090 Mining Down Perfomance Explication

jgonzi

Mitglied
Mitglied seit
Okt 3, 2020
Beiträge
10
Bewertungspunkte
3
Punkte
2
Hi,

This is my 1st post... I will do it in English, weil ich die Technische Nammen in Deutsch nicht verstehe.

I suppose some you are getting some loss of perfomance while mining...
All this problem is because Memory is in near Temp Junction, as Igor informs as with his article. That's the big problem.

I have Gigabyte RTX 3080 Gaming OC. I was seen that the Fans goes up to 100% in Windows 10 and HiveOS in Auto. Fan.. I was thinking drivers Problem, because the GPU temp was perfect... 55C-57C... or that my GPU was witha defect.

Making some Benchmarks in Windows without any MSI Afterburner or Aourus Engine the fan was OK... so the card should by OK, but when I put a miner and FAN 100%...

After all this information I found the problem... I see that reducing the memory speed the fan was staying under 70% automaticly... only a little downclock of -500Mhz,

So after that I decided to open the card, because with Gigabyte you don't loose Garantie and I make a Thermal Pad Mod too with the backplate. I haven't found any FLIR image of the Gigabyte so I decided to try with Igors Image and some other Photos of other articles and it always seems to be 2-3 grads more top left of the GPU (Backplate Side)

So I opened my Gigabyte to check that all memory pads where ok, and I found 1 of them was only covering 85%, I changed and I decided to put Thermal Pads in Backplate... I did with shit Thermal Pads I haven't got good ones yet... ( I have ordered now), the space in Gigabyte is 2.7mm

After that Thermal Mod... No more 100% Fans... And I don't need to lower Memory and I can push it high without arriving to 100% Fans.

In my opinion, the Bios is checking the Memory Temp and increase Fans Speed at least in Gigabyte Fans Auto Curve. So you can have a perfect GPU Temp but the Memory is getting to hot... I don't know if other cards are doing this check, I hope so.

I have added a Screen of my MSI... In the Green I push the Memory and Automaticly the Fans goes UP... In the Red I return to previos Memory and Fans go Down. I hope that could help someone to modify his GPU to reduced Temp in Mem too. It is true that Mining is goes of the largest load in Memory is all time 100%... comparing to games.

Many thanks, Vielen Dank
 

Anhänge

  • MemFan.jpg
    MemFan.jpg
    245,5 KB · Aufrufe : 15
Thanks for posting this. I observed similar behavior and had GPU throttling even at 48C-52C (gpu core clocks drop) on heavy memory workloads.
 
Good to know and bear in mind. My EVGA 3090 fans also revv up somewhat annoyingly with GPU temps even below 70C, but I will only start tinkering with it once water blocks from Watercool are available (incl active backplate). ;) Hope that will solve it then.
 

Here's an interesting observation. I ran a gpu core intensive load and that didn't throttle with Thrm even post 56 degrees but a memory intensive load immediately throttles past 52 degrees. This most likely indicates it might have something to do with GDDR6X temperatures?

I wonder if new batches of Gigabyte has done this kind of poor quality thermal pad implementation.
From the video link it actually looks like this might be highly possible if the thermal pads are memory chips can run significantly hotter.

I conducted some tests. The card throttled even at 48C under heavy memory workload and pointing a fan on the backplate side helped to temporarily gain the correct core clocks back.

Checked the thermal pad contacts too on memory chip. I do see some impression.
 
Similar results with ASUS TUF Gaming 3080. Thottling on 50-55C, NOT throttling on 47C. I haven`t opened my card yet, I will try and keep it mining for a few days to see results.
However, I thought I mention I have only seen this behaviour after updating driver to latest. Before, I had NO throttling issues at 50C degrees.

Current driver 460.89, driver before throttling isues = 457.51 (I think).
 
Similar results with ASUS TUF Gaming 3080. Thottling on 50-55C, NOT throttling on 47C. I haven`t opened my card yet, I will try and keep it mining for a few days to see results.
However, I thought I mention I have only seen this behaviour after updating driver to latest. Before, I had NO throttling issues at 50C degrees.

Current driver 460.89, driver before throttling isues = 457.51 (I think).
That's interesting that you didn't have throttling issues on older driver. I was going to attribute it to the cards that throttle are having only one set of caps. The cards that have mix capacitors don't seem to throttle. My EVGA card and FE model didn't throttle whereas I read reports on zotac, gigabyte, asus tuf series throttling.

Weirdly my ASUS card doesn't throttle.
 
Oben Unten