Hi guys,
I have the same problem here. Im using a Gigabyte RTX 3080 Gaming OC.
The only settings that I dont get a Thermal Throttling is: TDP - 52%; Mems: -500; Core -500; Fans - 90%, this give me 72 MH/s at 190W.
It took me days to find this thread and finally discover why my RTX 3080 dont generate the 100 MH/s.
Anyone has found a solution that doesnt involve voiding the warranty? Because I cannot disassambly the cooler without voiding the warranty (at least here in my country).
Also, I have a friend with a Asus RTX 3080 TUF OC with the same problem.
Never ever go cheap with cards and expect wonders... If you cant disassemble the crap coolers and lower the clocks is no option anymore, increase the fan speed further and/or use external 3000(+) high pressure fans in addition (like the Noctua Industrial). But this will hurt the lifespan of the original fans even more. So expect to replace them in the near future.
Go cheap? Its a bloody RTX 3080 card, even the entry level RTX 3080 is far cry from cheap, It SHOULD work properly. I dont really care about the lifespan of the fans, since this card has a 4 year warranty here.
I haven't watched all of hardware unboxed videos to see their reviews on this specific Gigabyte model but if you're dropping TDP down to 1/2 and lowering the clocks across the board you're basically only using 1/2 the card. I would just sell the card locally for a nice markup and find a different model of card or just buy the crypto directly. The Gigabyte doesn't have mixed caps so it was probably going to have issues right off the bat.
A friend got the Asus RTX 3080 TUF OC with the fancy MLCC caps and has the same problem. I dont think that is a cap issue, Gigabyte uses the 470u caps instead of the 220u caps (ZOTAC like), more than the double specified by nvidia.
It`s revolting that they have done something like that.
As Phil noted this is a case where mixed caps seems to filter multiple frequencies so that the heat buildup does get to extremes (see futher back in the thread for a chart he linked). Both the Asus and Gigabyte are thus bad models to use for mining because they're going with a single cap type. Obviously nVidia will remedy this with Rev 2 cards from AIB just like Asus had to RMA all those bad RX 5700 XTs that made poor thermal contact. Honestly, this is all nVidia's desire to say they were first to market even if it was a paper launch. Their Founders Editions had 6+ months to test out configs that worked and it looks like AIBs got screwed by having to rush to production.
Sometimes you get lucky buying the first batch because you might get better binned VRAM or higher binned chips (whereas lower yield chips would be saved for some cut-down refresh card). Sometimes you get screwed by meeting design defects and get to be the guinea pig. In this I couldn't even be a guinea pig because I could get my hands on one.
It still games well though as no game is going to run a constant 300w TDP through it. If you want the hashes, sell the card for a profit and just pick up a Founders edition. Phil sitting on 94MH/s sounds decent.
I have trouble getting my head around this explanation. The caps issue had to do with the cards being unstable when it boosted to high clocks - a remedy was to reduce the max clock speed.
Igorslab has done a test and showed high memory temps on the FE. The slowdown in hashrates is very similar to my sapphire rx5700 when the memory temp reaches 104-106C.
It wouldn't surprise me if the AIB models that have this hashrate drop issue cheaped out on thermal pad quality.
Perhaps the reason why you are not seeing hashrate drops on higher end, mixed cap models is because they didn't cheap out on the memory cooling.
Well the original idea first brought out was mixed caps allow for filtering more noise on more frequencies not my idea. My thread had all the links. others have done some teardowns and added high quality thermal pads. with some decent results. later not first.
So proper testing is get 3 shitty gpus models with one cap style
get 3 good gpus models with mixed cap style
do teardown on all six pull all thermal pads on all six.
run all six at lowest power settings and low core setting and low ram setting.
then measure ram memory temps on all six.
if they all over heat equally then it is not mixed caps
After reading newer evidence from guys like you about pads and older evidence from the links in my thread
I suspect caps and pads are the issue. with a third possibility that the actual ram itself could be better quality in the evgas I use.
Time will tell and it will get teased out , but I am thinking the buying mixed capped evga 3080's seems like best solution short term.
