• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

NVIDIA RTX 50 SERIES - Technical/General Discussion


this is one ugly looking card, how can you go from the gorgeous 4090 matrix to this abomination of a card??? someone mentioned here it looks like a certain male body part, now I can't unsee it lol.
 
Probably mentioned in previous pages:

This guide really helped me understand vf point undervolting:


Takes a good deal of testing for stability. Dialed in 0.900mv @ 2805MHz over several weeks and various stress tests and games, testing lots of different variables.

Thought it was stable for weeks with slightly lower voltage, then one game comes up and needs a little extra bump or two. Doesn't trigger power limits and is pretty cool running.

Recently found I could get +390 @ 0.935v @ 2917MHz.

The voltage increase per clock speed bump is definitely not linear, i.e. +405 works @ 2805MHz = 0.900v but +405 @ 2910MHz = 0.925v wasn’t stable. Very good starting point for working out the rest of your voltage curve going from low to high to try and get your personal chosen settings that meets temp and power preferences and game demands. I can keep the 2917MHz setting under 50c with my water cooled Palit card and mostly avoid hitting power limits.

I have just started testing the whole voltage curve, because I have pretty much tried every point at a clock speed and voltage that is almost on the money.

I have gotten to +390 @3007Mhz = 0.960v. This starts hitting power limits, but in many of the games I play it doesn't get massively hit and I can keep clocks at about 2900+ at slightly lower GPU usage and under certain game settings (Frame Gen, DLSS etc.) This only pushes the temps up a little bit and can function as my higher end profile for certain games. Still needs a bit of testing, but all pretty good info.

If it was shunt modded then I could keep all the clock speed I lose through early power limiting. Then I would go above 500w regularly though, which is not the goal.

All good fun anyway.
 
Last edited:
Probably mentioned in previous pages:

This guide really helped me understand vf point undervolting:


Takes a good deal of testing for stability. Dialed in 0.900mv @ 2805MHz over several weeks and various stress tests and games, testing lots of different variables.

Thought it was stable for weeks with slightly lower voltage, then one game comes up and needs a little extra bump or two. Doesn't trigger power limits and is pretty cool running.

Recently found I could get +390 @ 0.935v @ 2917MHz.

The voltage increase per clock speed bump is definitely not linear, i.e. +405 works @ 2805MHz = 0.900v but +405 @ 2910MHz = 0.925v wasn’t stable. Very good starting point for working out the rest of your voltage curve going from low to high to try and get your personal chosen settings that meets temp and power preferences and game demands. I can keep the 2917MHz setting under 50c with my water cooled Palit card and mostly avoid hitting power limits.

I have just started testing the whole voltage curve, because I have pretty much tried every point at a clock speed and voltage that is almost on the money.

I have gotten to +390 @3007Mhz = 0.960v. This starts hitting power limits, but in many of the games I play it doesn't get massively hit and I can keep clocks at about 2900+ at slightly lower GPU usage and under certain game settings (Frame Gen, DLSS etc.) This only pushes the temps up a little bit and can function as my higher end profile for certain games. Still needs a bit of testing, but all pretty good info.

If it was shunt modded then I could keep all the clock speed I lose through early power limiting. Then I would go above 500w regularly though, which is not the goal.

All good fun anyway.

Yes, that guide is GOLD! So much wrong info on undervolting out there on the internet it's unreal.

I've been undervolting my cards for years and thought I had it nailed down and got blasé about it. It seems the 50 series is slightly different though and I was actually doing my profile curves wrong for my new 5090FE.

You absolutely have to, like he shows, do a range first by holding SHIFT between 2 points - a low point (eg 0.815v so card still idles correctly) and the desired peak volt point (eg 0.895v) and then raise and flatten, which creates a different, correct 'shape' giving higher core boosts!

Before I was just dragging up the desired peak volt ONLY and then flattening the curve, which kept correct idle behaviour and volts etc, but created a much steeper shape and a slightly reduced performance/ lower boosting than it should.

For example, 3D Mark Steel Nomad I was only scoring ~137fps with my old 'incorrect' 0.895v profile whereas with the proper curve method and my 0.895v profile it's ~145fps! (Stock is ~140fps)

The vid above is a bit long and a deep dive. A good very short 3 min video I found shows the right way to do the curve also :

 
Last edited:
Yes, that guide is GOLD! So much wrong info on undervolting out there on the internet it's unreal.

I've been undervolting my cards for years and thought I had it nailed down and got blasé about it. It seems the 50 series is slightly different though and I was actually doing my profile curves wrong for my new 5090FE.

You absolutely have to, like he shows, do a range first by holding SHIFT between 2 points - a low point (eg 0.815v so card still idles correctly) and the desired peak volt point (eg 0.895v) and then raise and flatten, which creates a different, correct 'shape' giving higher core boosts!

Before I was just dragging up the desired peak volt ONLY and then flattening the curve, which kept correct idle behaviour and volts etc, but created a much steeper shape and a slightly reduced performance/ lower boosting than it should.

For example, 3D Mark Steel Nomad I was only scoring ~137fps with my old 'incorrect' 0.895v profile whereas with the proper curve method and my 0.895v profile it's ~145fps! (Stock is ~140fps)

The vid above is a bit long and a deep dive. A good very short 3 min video I found shows the right way to do the curve also :

Just watched the vid. That is a lot simpler!

I am the same as you. Always undervolted recent cards but never bothered with the curve as it seemed too complicated.

I've got way to into it now and I have gotten to 0.975v @ 3052 MHz. This serves as my highest profile that just starts to touch power limits in a game that isn't too GPU demanding, or something like Cyberpunk 2077 or MSFS 2024 where DLSS and frame gen are pretty much a requirement for getting good FPS.

I put a 90% power limit on just for safety, as I confirmed that this profile can hit 520W in a hard stress test, uncapped. I sometimes remove that if needed and I see it isn't going to go over that in a game.

This keeps me in the mid to high 2900 clock speeds and scores about 15100 in steel nomad, without touching memory overclocking. I may look at memory again at some point.

I think if I was running the card on air I would just keep the original 0.900v 2805MHz undervolt in place.

I am not really undervolting anymore, but more like taming the higher wattage and voltage so it doesn't get insane at the higher clock speeds.
 
Last edited:
^ Nice one. You're watercooled no @0121danwilliams84 ?

With my (very new) 5090FE only air cooled and this whole melting cables debacle, I'm a tad worried so I'm really leaning into the undervolting game now.

For the stuff I'm playing at the moment, even intense VR stuff, I'm finding I really don't have to push the card at all though and it's just superb. I'm running most games with my very conservative 0.85v profile as the really demanding ones have framegen anyway, and you still actually need framegen if you want to hit above 100fps at max settings 4k, as I like to do.

Basically, you can run a game like Cyberpunk maxed out with the GPU at high volts and maxed out at 550watts+ and still 'only' get about 75fps and a risk of combustion. Or, you can run the GPU at a nice 0.85v undervolt curve, barely use more than 300watts, still get about 60fps anyway and then framegen that up to 120. Which actually reduces power draw even more on the GPU. I know which I prefer to do....
 
Last edited:
^ Nice one. You're watercooled no @0121danwilliams84 ?

With my (very new) 5090FE only air cooled and this whole melting cables debacle, I'm a tad worried so I'm really leaning into the undervolting game now.

For the stuff I'm playing at the moment, even intense VR stuff, I'm finding I really don't have to push the card at all though and it's just superb. I'm running most games with my very conservative 0.85v profile as the really demanding ones have framegen anyway, and you still actually need framegen if you want to hit above 100fps at max settings 4k, as I like to do.

Basically, you can run a game like Cyberpunk maxed out with the GPU at high volts and maxed out at 550watts+ and still 'only' get about 75fps and a risk of combustion. Or, you can run the GPU at a nice 0.85v undervolt curve, barely use more than 300watts, still get about 60fps anyway and then framegen that up to 120. Which actually reduces power draw even more on the GPU. I know which I prefer to do....
Yes, that's a good point. I set myself a limit of around 500W when I originally got the card.

The higher profile is actually slightly more beneficial in some games that I play regularly and I can always switch around. Some regular games I play don't even come close to 500w, without using any upscaling or other tech.

If the cable is installed correctly then there is a low risk of melting a cable, but cables have been shown to melt at many power levels.

How do us mere mortals without Astrals know if it is connected correctly? We don't. If it's going to melt, it will probably melt.

Yeah, running watercooled and keeping it under 50c, so that should also assist the cable staying cooler than say 70,80c + with air, which should assist also.

If the load on the power pins in inbalanced then yes a lower power limit should in theory assist, but it can't completely stop catastrophic failure.
 
Last edited:
Back
Top Bottom