Drwaffles

machie

My card really doesn't like 2100MHz, even with 1.1V to the GPU core, so I had my card locked at 2085MHz with 1.075V. When pinned at 1.1V GPU core voltage, the card hits the pwr limitter so often that the board power draw, core voltage and core clock fluctuates so much that I don't think you'll get good benchmark off of it anyway. All I'm trying to do here is draw as much power as I can, not set the best frequency/voltage curve for a benchmark run.
 
I did managed 19.2k graphics score in time spy at with the card locked at 2070MHz with 1.05V to the GPU core, and that's pretty the best my card can do.

As far as I understand it, the benches count how many frames you can draw in a given amount of time.. More frequency, more frames (Provided no errors); the power draw is just a consequence of the processor doing it's thing.
As you hit power limits, the frequecy will drop further as the voltage drops..
 
Yes, lower voltage = less power limiter bashing, but also = lower frequency and still dips to the same level anyway when you hit the power limits. Score wise you're typically still better off having it high as possible, even if it's got dips (So long as it doesn't introduce stutters etc) It's a bit like driving in the wrong gear all the time because you don't want to get close to the redline.
 
When you say pinned at 1.1v, do you mean you locked it on the curve in afterburner? Or you adjusted the curve so it actually ran at 1100mv? And you don't have to run 2100mhz, but 1100mv might be enough for 2085 stable 1.093 might be enough for 2070mhz stable etc
 
 
Granted, this is difficult without seeing your exact curve but it's just that your +100 core doesn't appear to be doing anything currently. If the curve you've set doesn't increase in clocks at that maximum voltage point you want.. It'll just run at lowest voltage for the highest frequency you've set like this. But maybe you did that and i'm misunderstanding you.

DrwafflesWouldn't necessarily recomend the XOC on air (Nor my solution) 420w+ is hard to keep cool. Even with 85% fan duty and an open case I'd see 65-72c.. It can help if you're trying to get up the leaderboards on timespy, but I'd only do one bios and keep the other standard, at least you've got the option.
I've got an EK block now and it's 46-52c under heavy load depending on my water temperature.

cennis2018
What are your serial number first 4 digits? For people who can/cannot reach 450W?
 
From the 3090 forums,
I am seeing the 2119 batch reaches 500W and 2012 batch not being able to reach 500W
 

Drwaffles
Do remember they never promise gains, overclocking is luck of the draw. You could but a Strix card that doesn't do 15mhz over the 1900 boost.

What voltage and temps is it running when crashing at 2025-2040m
Try using the curve instead of the offset, helped mine.

Drwaffles
Do remember they never promise gains, overclocking is luck of the draw. You could buy a Strix card that doesn't do 15mhz over the 1900 boost.
They don't deserve blame for silicon lottery.. The FTW3 PCB was a big let down compared to my old 1080, but that's another story of cost cutting (I pre ordered on faith it'd be similar quality, but it was substantiality worse)

What voltage and temps is it running when crashing at 2025-2040m
Try using the curve instead of the offset, helped mine.

Drwaffles
Try 1.1v (Use the curve at 1.1v, that point needs to be a higher frequency than the 1.093v point)

Also remember that as you hit a power limiter the voltage drops, so even though you set 1093mv, it might be down at 1063 or lower.. You need to see what it's doing in GPUZ under a load like timespy or port royal.

60-62c is excellent, that's not overheating.

That's a presumption at this stage.. And again this isn't what we're discussing.. Core frequency achievable is LUCK.. We're talking about power limit variation, not OC headroom.

cennis2018
What are your serial number first 4 digits? For people who can/cannot reach 450W?
 
From the 3090 forums,
I am seeing the 2119 batch reaches 500W and 2012 batch not being able to reach 500W
 

kraade
I have had 2 3080s , both experienced un-even draw across the 3 8pins my first config had a daisy chain on 1 and 3 and I first thought this was due to the daisy chain, but with a new PSU and new independent cables, I think 3 has been lower on both cards. one SN was 2012 and the current is 2014

Drwaffles
The FTW3 PCB was a big let down compared to my old 1080

Drwaffles

MarlowX1m
 
 
The 3rd 8-pin according to this information was supposed to be a 6-pin and 8-pin was a last minute change. Might explain why we are are limited to about 400w and how aggressive the firmware is throttling trying to stop it from jumping to higher frequencies . Apparently there is a transient response problem

 
Certinally would explain why now that's i'm not bouncing off the power limiter on the MSI bios (and able to maintain a fixed voltage) all my crashing issues went away while overclocked.
I've half a mind to RMA my card due to the inabilty manage its power draw, but i'm not sure the issue is actually resolved.. That and Australia like EU are getting pretty well ignored supply wise, it'll probably be a 3 month wait.
 
I've got another one on order for a mate, he won't run above stock setting so if his hits 440+ at 1.1v i'll swap my block across. 
Certinally all very interesting though.. Seems like a lot of cost cutting is starting to bite them; It's never a good thing when Accountants get in charge of design.

Attached Image(s)