Disappointing 3090 FE performance - any ideas

73C does seem hot for a watercooled 3900 but it's less than the throttle threshold. GPU performance is generally proportional to its clock and if it can boost to its max (and memory clock is OK) then it will be at its maximum performance.
 
Core is now idling at 33c with a coolant temp of 30. max It got today while gaming was 49c, significantly better than the 70+. stock settings

Still can't quite my head around undervolting and overclocking, but i'm sure i'll get there.
 
Core is now idling at 33c with a coolant temp of 30. max It got today while gaming was 49c, significantly better than the 70+. stock settings

Still can't quite my head around undervolting and overclocking, but i'm sure i'll get there.

Much better temps. Your average clocks on your recent run look much more inline, nearly 200 Mhz extra on average.

Guide for undervolting to watch to understand basics. In a nutshell every chip will vary so about learning your behaviour. You will note under load your card typically uses say 1v+ in use. What you want to do is set say 0.8v for example and see the clock speed your card can hit. Generally speaking Nvidia out the box apply more then extra needed voltage to ensure stability, even when your chip is a dud. But if you fine tune things can see a real nice save in power usage / heat. Or you can use slightly more voltage and turn it into a overclock while saving power. You will see as you increase clock speed, takes more and more voltage.

 
That looks much better now :) https://www.3dmark.com/compare/spy/28605027/spy/28618082#

You can see the difference in your average GPU clocks by switching the block over, the GPU hotspot temp was likely causing throttling due to a bad contact. First time I put my block on the 3080 it had a bad contact resulting in a similar experience and was really evident when I took it apart it due to the lack off pressure showing on one side of the TIM.
 
many thanks. I solved my sheered bolt issue so now await some new bolts from EK. Meanwhile i'm going to read up on hotspots and the like. I noticed a discrepancy in the height of the pads on the rear, meaning that the memory couldn't make contact with the backplane. EK only supply one thickness whereas OC supply different thicknesses. i'll pay more attention when refitting the EK block.

I do like the optimum tech videos though. That video describes undervolting I think, but does it explain overclocking? or is it a single action within the curve editor, ie drag the curve down, then take one point up to say 1950 and apply?

cheers
 
73C does seem hot for a watercooled 3900 but it's less than the throttle threshold. GPU performance is generally proportional to its clock and if it can boost to its max (and memory clock is OK) then it will be at its maximum performance.
He mentioned Hotspot was 105c , with air cooling hotspot delta should ideally be 10-15c over GPU temp

Probably mounting issue or paste not fully spreading
 
many thanks. I solved my sheered bolt issue so now await some new bolts from EK. Meanwhile i'm going to read up on hotspots and the like. I noticed a discrepancy in the height of the pads on the rear, meaning that the memory couldn't make contact with the backplane. EK only supply one thickness whereas OC supply different thicknesses. i'll pay more attention when refitting the EK block.

I do like the optimum tech videos though. That video describes undervolting I think, but does it explain overclocking? or is it a single action within the curve editor, ie drag the curve down, then take one point up to say 1950 and apply?

cheers
He explains undervolting with results You can actually gain more with undervolting doesn't mean you need to lose performance while lowering power use and lowering the temps . Each card is different so what results he gets doesn't mean you will get the same could get better or worse due to silicone lottery you can adjust slightly till its stable you have base to play around

Overclocking will just use more power and you won't notice the few FPS increase in gaming unless you chasing benchmark numbers for how much power increase its generally not worth it everyday gaming use
 
Again thanks. I noted that my last run used only 350 watts as opposed to over 400 in earlier tests.

My preferred 'game' is MS FS2020, which is, of course, limited to a single core, though we all hope DX12 and DLss is going to save the world :)

I'll continue tweaking, but am comfortable I've resolved my basic missunderstanding of the subject generally.

next up is which thermal pads I should use for the EK rebuild
 
He mentioned Hotspot was 105c , with air cooling hotspot delta should ideally be 10-15c over GPU temp

Probably mounting issue or paste not fully spreading
Agreed.

Sounds like it's sorted now.

I can't recall the hot spot calculation, is it at least 10c above average core temp, and includes the VRM temp as part of the calc?

My 3080 is 13c delta when not much is going on, and once things heat up if it fluctuates between 10-12c delta.

Highest I get is 77 hotspot after an hour of quake rtx.

So pleased my I went to water-cooling.
 
Again thanks. I noted that my last run used only 350 watts as opposed to over 400 in earlier tests.

My preferred 'game' is MS FS2020, which is, of course, limited to a single core, though we all hope DX12 and DLss is going to save the world :)

I'll continue tweaking, but am comfortable I've resolved my basic missunderstanding of the subject generally.

next up is which thermal pads I should use for the EK rebuild
Everyday is a school day with PC building
 
Well, that proves a point. In trying to tighten the EK block, I managed to sheer one of the threaded bolts holding the block to the 3090. I'm exploring with EK what to do. however I dug out my old OC water block and connected that. Much better performance https://www.3dmark.com/3dm/76050329. Core only got to 57, hot spot to 60 and junction 62. Core Mhz behaved much better too during the tests.

Now to fix the ek !!
How did you manage to strip a screw? Did you over tighten or force the screw in when its not going in smooth? The bolts should move freely in and out until its tight and its always advised not to overtighten.
 
How did you manage to strip a screw? Did you over tighten or force the screw in when its not going in smooth? The bolts should move freely in and out until its tight and its always advised not to overtighten.
Simple really. The instructions say use 2.5 4mm bolts to affix the card to the block , the four around the actual gou chip. But the ek video disentangle mention that. So I used the supplied 2.5 6mm, not reading instructions. RTFM and all that. The additional pressure from my tightening shared the threaded bolt in the block. Fortunately I was able to drill out the bolts without affecting the threads. Ek are sending me new threaded bolts. Difficult to describe, but no harm done. As I say RTFM, though I’m annoyed ek didn’t supply the right number of the right bolts.
 
And my last question. Any recommendations for thermal pads? I think different thicknesses are needed, though EK only supply one. Maybe 1mm and 1.5 in places around the back
 
Simple really. The instructions say use 2.5 4mm bolts to affix the card to the block , the four around the actual gou chip. But the ek video disentangle mention that. So I used the supplied 2.5 6mm, not reading instructions. RTFM and all that. The additional pressure from my tightening shared the threaded bolt in the block. Fortunately I was able to drill out the bolts without affecting the threads. Ek are sending me new threaded bolts. Difficult to describe, but no harm done. As I say RTFM, though I’m annoyed ek didn’t supply the right number of the right bolts.
Ouch, if the screw doesn't wanna go i usually stop and double check. Seems abit insane to force it when its not wanting to screw in.
 
And my last question. Any recommendations for thermal pads? I think different thicknesses are needed, though EK only supply one. Maybe 1mm and 1.5 in places around the back
I used Gelid Thermal pads which were really good for backplate cooling with heatsinks. I can't see it being any different for use on a block on memory chips etc.
 
That's me, insane, but thanks for the reminder. As I say RTFM, not the video.

TBH the EK block I found pretty meh in quality control. For me one of the screws was done in so tight from the factory when I was undoing the block, it stripped. Had to drill it out which did scuff the finish. Very annoying though not one you can see.

the EK Pads are 1mm all around. You should honestly find memory temps perfectly fine with the stock pads tbh unless you moved them / touched them a lot etc, even overclocked from what I have seen using block since they released temp wise. but yeah, if you do replace them would be 1mm all around. Don't want to go overly thick for sake of it, just enough for good contact and heat transfer.
 
TBH the EK block I found pretty meh in quality control. For me one of the screws was done in so tight from the factory when I was undoing the block, it stripped. Had to drill it out which did scuff the finish. Very annoying though not one you can see.

the EK Pads are 1mm all around. You should honestly find memory temps perfectly fine with the stock pads tbh unless you moved them / touched them a lot etc, even overclocked from what I have seen using block since they released temp wise. but yeah, if you do replace them would be 1mm all around. Don't want to go overly thick for sake of it, just enough for good contact and heat transfer.
The trouble with a single thickness of pad is that some of memory chips sit higher than others on the card, which to my mind means some might not be in contact and others over compressed. Having looked again at the OC version they supply different thicknesses to compensate.
 
TBH the EK block I found pretty meh in quality control. For me one of the screws was done in so tight from the factory when I was undoing the block, it stripped. Had to drill it out which did scuff the finish. Very annoying though not one you can see.

the EK Pads are 1mm all around. You should honestly find memory temps perfectly fine with the stock pads tbh unless you moved them / touched them a lot etc, even overclocked from what I have seen using block since they released temp wise. but yeah, if you do replace them would be 1mm all around. Don't want to go overly thick for sake of it, just enough for good contact and heat transfer.
Usually with EK the screws are too loose to the point where the orings aren't compressed and it just leaks. I've made a habit of just checking all screws before i install anything as all manufacturers fall flat on this at some point for someone.
 
Back
Top Bottom