NVIDIA RTX 50 SERIES - Technical/General Discussion

[Rusty0611]

I don't get, what people don't get, about undervolting/ overclocking. You can do both at the same time and double win.

The daily conservative undervolt + overclock profile I've settled on for my 5080 is 0.91v at 2950Mhz core and +1400Mhz on the Vram (Default core volts are ~1.01v and core clocks 2750Mhz).

So I'm overclocking AND undervolting, and get more performance Vs stock whilst shaving off about 80watts of power which results in less heat and noise. What is not to like about this?

I've also done a super aggressive PBO undervolt / overclock setup on my 9700X (superb CPU by the way) so it runs at low volts whilst boosting all cores to max. Again, more performance Vs stock whilst drawing less power, and having less heat and noise. My air cooled rig is really quiet now and flies.

Agreed. Both go hand in hand if you want to get the most from your card. It’s definitely not an either/or situation.

I really want to reduce the voltage more for my 24/7 clock speeds and see if I can unlock more headroom but it’s so demoralising and time consuming when it crashes and you’re on the limit that I think I’m just gonna leave it where it is . Mine runs at 52-60c and doesn’t hit close to 600W except for Steel Nomad.

Anyone else sniffing their power cable for burning when it’s maxed out?

 

[Varys]

Agreed. Both go hand in hand if you want to get the most from your card. It’s definitely not an either/or situation.

I really want to reduce the voltage more for my 24/7 clock speeds and see if I can unlock more headroom but it’s so demoralising and time consuming when it crashes and you’re on the limit that I think I’m just gonna leave it where it is . Mine runs at 52-60c and doesn’t hit close to 600W except for Steel Nomad.

Anyone else sniffing their power cable for burning when it’s maxed out?

so you want to undervolt afterall

 

[Vito]

Agreed. Both go hand in hand if you want to get the most from your card. It’s definitely not an either/or situation.

I really want to reduce the voltage more for my 24/7 clock speeds and see if I can unlock more headroom but it’s so demoralising and time consuming when it crashes and you’re on the limit that I think I’m just gonna leave it where it is . Mine runs at 52-60c and doesn’t hit close to 600W except for Steel Nomad.

Anyone else sniffing their power cable for burning when it’s maxed out?

I see you have a LC, without a fan on it . Got the Astral LC 'just now' so not needing to worry about the power cable due to the monitering, runs well within spec.
Would you mind sharing your curve, I can't seem to find it from yor posts? I've tried about 20 varations, trying to get the highest boost for lowest voltage (stable gaming) so I can to keep fans low. The rad fans are quite enough tbh, but the 1 on the card gets audible above 1,100 rpm or so. When I turned it down the memory junction temps went upwards os 84°C. Not having it for 3.3k, so might try to get a cheaper card and just put a WC loop back on, only need block for GPU...

 

[CornyOC]

Agreed. Both go hand in hand if you want to get the most from your card. It’s definitely not an either/or situation.

I really want to reduce the voltage more for my 24/7 clock speeds and see if I can unlock more headroom but it’s so demoralising and time consuming when it crashes and you’re on the limit that I think I’m just gonna leave it where it is . Mine runs at 52-60c and doesn’t hit close to 600W except for Steel Nomad.

Anyone else sniffing their power cable for burning when it’s maxed out?

Just buy a thermocouple and tape it to the power cable GPU end.

 

[Cobras]

Ok, so I have a Palit 5090 on the way and have a few questions.

1. Can I still use NVCleaninstall and the old Nvidia control panel with the 5090 or does it require Nvidia APP?

2.Has the old Nvidia control panel been updated to include support for multi frame gen and DLSS swapping or is this only on the Nvidia APP?

3.What is the most stable driver currently for the 5090?

 

[SeeNoWeevil]

The way I see it:

- overclock: increasing the clock speeds over stock (for performance).

- undervolt: decreasing the voltage compared to stock (primarily for for temps or noise).

They used to be ‘opposing approaches’ but now more than ever they are complementary; an undervolt that keeps temperatures lower can help enable / sustain a mild overclock.

Undervolted cards clock higher because you've basically told them they need less voltage for a given clockspeed. Nothing to do with temps.

A spec comparison I haven't seen elsewhere, here are the weights of various 5090 I pieced together from listings and reviews (weights do not include the extra fans on some of the higher-end models e.g. Aorus):

Aorus - 3.100kg
Astral - 3.042kg
Suprim - 2.840kg
Vangaurd - 2.407
Triple Fan - 2.380kg
TUF - 2.370kg
Solid - 2.287kg
Trio - 2.119kg
Gamerock - 2.222kg
Gainward Phantom - ~2.000kg
Windforce - 2.000kg
Ventus - 1.882kg
Founders Edition - 1.814kg

Note how weight largely, and unsurprisingly, correllates with thermal performance (and tier) of each model. Takeaway is, for all the gusto about airflow cooling design (they all use 3x fan setups except the FE), it's really the size of the good-old aliminum sink and copper pipes that count. The only real outlier is the Aorus on which I did a double take on (and reconfirmed with a few listings) -- use a support backet with that one, kids.

Note: I could find no reviews of PNY or Inno3D card weights anywhere.

AIB cards are pointlessly large. None of that extra size translates into any meaningful performance out the other end once past a certain point. I've had 2 5080s. One massive (Asus TUF) and one 'small' MSI Inspire. Both overclocked to within spitting distance of each other. The TUF ran a few degrees cooler but that's irrelevant and almost certainly just BIOS fan curve. The MSI was actually quieter. It also didn't have horrendous coil whine. So yeah, all that bulk and 'performance' inductors just resulted in a big, whiny GPU.

 

[Nitefly]

Undervolted cards clock higher because you've basically told them they need less voltage for a given clockspeed. Nothing to do with temps.

Historically some people have undervolted cards so that they use less power and run cooler, thereby requiring less GPU / CPU / case fan RPM = less noise.

Depends on the purpose.

Same performance + lower temps = good reason to undervolt.

 

[Rusty0611]

so you want to undervolt afterall

I think you're very confused . Mines always been running with less voltage than default which I have said repeatedly...

Process so far:

- work out top end of where I want core speed to be at (3250 MHz ish for me)
- apply and make sure it's stable
- scale the voltage back a bit and make sure it's still stable

Next up:

- work out exactly where the breakpoint is on voltage which means....

I see you have a LC, without a fan on it . Got the Astral LC 'just now' so not needing to worry about the power cable due to the monitering, runs well within spec.
Would you mind sharing your curve, I can't seem to find it from yor posts? I've tried about 20 varations, trying to get the highest boost for lowest voltage (stable gaming) so I can to keep fans low. The rad fans are quite enough tbh, but the 1 on the card gets audible above 1,100 rpm or so. When I turned it down the memory junction temps went upwards os 84°C. Not having it for 3.3k, so might try to get a cheaper card and just put a WC loop back on, only need block for GPU...

Apologies, have just gone back to square one on my curve to fine tune it to within an inch of its life so don't have it anymore. But it was running at whatever the default was for this card minus a few chunks downwards.

84c on memory junction is like 15c higher than mine which is surprising!! Though haven't been really monitoring it too much so will leave HWInfo on next time I'm gaming and see what it is over a period of time.

Just buy a thermocouple and tape it to the power cable GPU end.

I am not really that worried was just a joke . But definitely happier when it's topping out in the low 500s than running at 600W capped. Most games, so far, seem to be like 450-500W

 

[CornyOC]

Ok, so I have a Palit 5090 on the way and have a few questions.

1. Can I still use NVCleaninstall and the old Nvidia control panel with the 5090 or does it require Nvidia APP?

2.Has the old Nvidia control panel been updated to include support for multi frame gen and DLSS swapping or is this only on the Nvidia APP?

3.What is the most stable driver currently for the 5090?

1. I installed Nvidia app and it also installed control panel. No idea if you can use it without the Nvidia app.

2. I don't see MFG option in control panel.

3. I run the latest 572.83 and I only crashed from unstable overclock, the driver itself seems stable. Be aware though, even the latest version will not ensure all games gonna run well on your 5090, you can expect poor performance in some games, mostly older ones.

 

[McCarthy]

In terms of undervolting, what is the current best game to test with. I remember the 3090 was Quake II RTX. It was brutal!

 

[SuperElectric]

I see you have a LC, without a fan on it . Got the Astral LC 'just now' so not needing to worry about the power cable due to the monitering, runs well within spec.
Would you mind sharing your curve, I can't seem to find it from yor posts? I've tried about 20 varations, trying to get the highest boost for lowest voltage (stable gaming) so I can to keep fans low. The rad fans are quite enough tbh, but the 1 on the card gets audible above 1,100 rpm or so. When I turned it down the memory junction temps went upwards os 84°C. Not having it for 3.3k, so might try to get a cheaper card and just put a WC loop back on, only need block for GPU...

That's a shame, I thought the Astral LC would be near silent. I still have half a mind to go with one of the liquid 5090s, but only if they're siginificantly quieter than the Vanguard (and I can turn down the case fans).

What rpm did you turn it down to for the 84°C mem temp?

 

[Vito]

That's a shame, I thought the Astral LC would be near silent. I still have half a mind to go with one of the liquid 5090s, but only if they're siginificantly quieter than the Vanguard (and I can turn down the case fans).

What rpm did you turn it down to for the 84°C mem temp?

Only on Alan Wake 2, not Indy...800rpm Cant hear it. No audible coil whine when gaming, low when benching. It was a high 3050 boost +3000 mem. Have Suprim LC on order but no date.
I ll play around more this week.

 

[razorpakk]

In terms of undervolting, what is the current best game to test with. I remember the 3090 was Quake II RTX. It was brutal!

Alan Wake 2, arguably the "heaviest" game in terms of feautures, was stable for me, while AC: Shadows wasn't (maybe because it's buggy).

Essentially, there isn't a perfect game. The instability is often in specific spots of the curve (frequency/voltage moves up and down dynamically) and different games hit the card in different ways.
So just play what you want to play and see what happens...

I think you're very confused . Mines always been running with less voltage than default which I have said repeatedly...

Process so far:

- work out top end of where I want core speed to be at (3250 MHz ish for me)

What voltage are you using for that ridicolous frequency? Also, out of curiosity, what Steel Nomad score can you hit up there?

 

[Rusty0611]

What voltage are you using for that ridicolous frequency? Also, out of curiosity, what Steel Nomad score can you hit up there?

Can’t remember exactly as my old curve has gone, but 1.0x volts. Not helpful I know

Steel Nomad I got 15.3k but needed to reduce the voltage more to reduce power as it wasn’t boosting anywhere near fully in that benchmark.

Might re-run the benchmark when I have redone the voltage curve again to see what gains I get but it all takes time and to be honest, as my temps are basically sub 60c 95% of the time even on the old curve, not sure how much time I will spend on it.

 

[razorpakk]

Can’t remember exactly as my old curve has gone, but 1.0x volts. Not helpful I know

Steel Nomad I got 15.3k but needed to reduce the voltage more to reduce power as it wasn’t boosting anywhere near fully in that benchmark.

Might re-run the benchmark when I have redone the voltage curve again to see what gains I get but it all takes time and to be honest, as my temps are basically sub 60c 95% of the time even on the old curve, not sure how much time I will spend on it.

This 15405 is with average core of 2,817 on the Vanguard (had flashed Aorus Master bios for a moment, but that doesn't matter as the card was undervolted @0.92).

I noticed massive diminishing returns when going past that frequency in games/benchmarks.

This 15615 is at 3,025. Way more power for barely anything.

 

[Rusty0611]

This 15405 is with average core of 2,817 on the Vanguard (had flashed Aorus Master bios for a moment, but that doesn't matter as the card was undervolted @0.92).

I noticed massive diminishing returns when going past that frequency in games/benchmarks.

This 15615 is at 3,025. Way more power for barely anything.

I don’t really care about how much power it’s drawing, personally, as long as it’s not hitting 600W. Although I’m ATX 3.1 both ends still don’t want to hit 600W.

Games I see much bigger gains than Steel Nomad by having higher clocks. 5-8% generally. Worth it for me.

 

[Vito]

This 15405 is with average core of 2,817 on the Vanguard (had flashed Aorus Master bios for a moment, but that doesn't matter as the card was undervolted @0.92).

I noticed massive diminishing returns when going past that frequency in games/benchmarks.

This 15615 is at 3,025. Way more power for barely anything.

Thats great on that voltage, pretty sure I was over 1 volt to hit 5,400 odd and higher clock, so going to try that (Astral LC). Did u just raise everything below 0.92 the same amount as 0.92 @ 2, 817, then flatten the curve after 2,817. This is new to me and I'm trying to learn from more experienced members, thanks.

 

[razorpakk]

Thats great on that voltage, pretty sure I was over 1 volt to hit 5,400 odd and higher clock, so going to try that (Astral LC). Did u just raise everything below 0.92 the same amount as 0.92 @ 2, 817, then flatten the curve after 2,817. This is new to me and I'm trying to learn from more experienced members, thanks.

Up until 0.9 it was all maxed out - going as high as Afterburner allows (so for me [email protected] and 2917 I believe, for 0.9). The steps all the way to 0.92 are much more granular though, as the GPU needs exponentially more power to go above 2900Mhz.

But it's more of a fun experiment, for daily usage I don't bother going above 0.9 (I might if I wasn't worried about the connector melting, but here we are).

 

[Vito]

Up until 0.9 it was all maxed out - going as high as Afterburner allows (so for me [email protected] and 2917 I believe, for 0.9). The steps all the way to 0.92 are much more granular though, as the GPU needs exponentially more power to go above 2900Mhz.

But it's more of a fun experiment, for daily usage I don't bother going above 0.9 (I might if I wasn't worried about the connector melting, but here we are).

Thanks very much for that.

 

[CornyOC]

This 15405 is with average core of 2,817 on the Vanguard (had flashed Aorus Master bios for a moment, but that doesn't matter as the card was undervolted @0.92).

I noticed massive diminishing returns when going past that frequency in games/benchmarks.

This 15615 is at 3,025. Way more power for barely anything.

Those are impressive figures for stock cooling solution. Even LN2 benchers got only got another 1000 points out of it. I wonder though how those settings gonna work out in practice.

I was playing First Berserker: Khazan and my 14600-ish figure OC proved to be stable for like 20 hours until I got to some area, where I could replicate the crash until I added a bit more juice (from 0.895V to 0.9V).