NVIDIA RTX 50 SERIES - Technical/General Discussion

[heatonpkmassive]

But... but... ITS THE PRINCIPLE!

I'm sure Nvidia are well aware of what principles are

 

[Copesman]

I think it will get patched now there's a stink

 

[andy85wsm]

Literally just installed the card, looks and feels very good quality. All metallic no plastics. Updating drivers ready for afew benchmarks.

Has all 112 Rops

Thanks! That's the one I've asked for, just waiting for a quote now.

Everything is in stock so I'm keen to get it ordered and built!

 

[Clarence Boddicker]

Does anybody have any up to date info on how many gpu’s a factory can produce per week/month. A post in a thread from 2020 claimed up to 400’000 cards per month could be produced by factory.

It’s 6 weeks since launch and 4 weeks since cny ended. Just trying to work out why retailers seem to be getting such small batches of cards and why so few FE’s.

 

[Joxeon]

Does anybody have any up to date info on how many gpu’s a factory can produce per week/month. A post in a thread from 2020 claimed up to 400’000 cards per month could be produced by factory.

It’s 6 weeks since launch and 4 weeks since cny ended. Just trying to work out why retailers seem to be getting such small batches of cards and why so few FE’s.

Nvidia stopped producing the 40 series a while back and shifted all the production to AI and have only swapped 1/5 of that production back to the 50 series.

 

[Ayahuasca]

Does anybody have any up to date info on how many gpu’s a factory can produce per week/month. A post in a thread from 2020 claimed up to 400’000 cards per month could be produced by factory.

It’s 6 weeks since launch and 4 weeks since cny ended. Just trying to work out why retailers seem to be getting such small batches of cards and why so few FE’s.

 

[Trox]

Does anybody have any up to date info on how many gpu’s a factory can produce per week/month. A post in a thread from 2020 claimed up to 400’000 cards per month could be produced by factory.

It’s 6 weeks since launch and 4 weeks since cny ended. Just trying to work out why retailers seem to be getting such small batches of cards and why so few FE’s.

It’s not nvidias main focus now

 

[Rogueywon]

Does anybody have any up to date info on how many gpu’s a factory can produce per week/month. A post in a thread from 2020 claimed up to 400’000 cards per month could be produced by factory.

It’s 6 weeks since launch and 4 weeks since cny ended. Just trying to work out why retailers seem to be getting such small batches of cards and why so few FE’s.

From the way we have stopped hearing about people receiving cards with missing ROPs, having heard of a lot of early cases, I would guess that a lot of cards working through the supply chain were held up for checking.

 

[SeeNoWeevil]

There's a lot of juice left in 5080s out the box. Most seem to do +550Mhz on the core at around 290w. Can't really get much more performance for no-melt-wattage.

 

[IT Troll]

I'd like to undervolt the 5090 to tame the power draw and improve efficiency.
I guess best results are had by adjusting the voltage curve in Afterburner.
However, for a quick tweak, I see derbauer just reduced the power target. Is this method now working OK in the nVidia app?

 

[no_1_dave]

I'd like to undervolt the 5090 to tame the power draw and improve efficiency.
I guess best results are had by adjusting the voltage curve in Afterburner.
However, for a quick tweak, I see derbauer just reduced the power target. Is this method now working OK in the nVidia app?

The Nvidia app didn't work at first, but it's fixed now

 

[SeeNoWeevil]

I'd like to undervolt the 5090 to tame the power draw and improve efficiency.
I guess best results are had by adjusting the voltage curve in Afterburner.
However, for a quick tweak, I see derbauer just reduced the power target. Is this method now working OK in the nVidia app?

Increase the Core until you find the highest possible value where everything is still stable. This will give you the most efficient curve for your particular chip. If you stopped there, you'd have higher boost clocks with the same power draw as stock. However, you could then increase the power limit for even more performance and power draw OR drop the slider for less performance and less power draw. When dropping the limit, there will be a point (let's say 85%) where you will have the same performance as stock AND lower power draw. Obviously none of this is guaranteed though. It's possible (unlikely) that the chip has no headroom whatsoever and increasing the core makes it unstable. You could still drop the power limit on its own and save power at the expense of some performance.

 

[Quartz]

That's fair enough, but I get the impression that you're in a big minority that they're not really bothered about. I'm not saying that's right, but I think that's their thinking.

Compare this to DX9 support on Arc

 

[Quartz]

It’s just one game? So it’s literally just borderlands 2 that’s affected?

There are many. The one I play is Borderlands 2

 

[IT Troll]

The Nvidia app didn't work at first, but it's fixed now

When dropping the limit, there will be a point (let's say 85%) where you will have the same performance as stock AND lower power draw.

Thanks, that would be the ideal. I'll give it a go when it arrives.

 

[Grim5]

I normally like GN videos, but this is a bit of a nothing burger.

Will you say it's a nothing burger in 10 years when RTX and DLSS stops working on current games because future Nvidia architecture doesn't have fixed function Tensor cores???

Sounds like a nothing burger alright!!

This is the reality of exclusive hardware features like this, Nvidia is only interested in pushing it when it suits them and will deprecate it later. With PhysX that's not too major as not many games used it but so many games now use DLSS and RTX fixed function hardware features that will
Completely stop working when Nvidia moves to a new architecture philosophy in future

 

[heatonpkmassive]

Will you say it's a nothing burger in 10 years when RTX and DLSS stops working on current games because future Nvidia architecture doesn't have fixed function Tensor cores???

Sounds like a nothing burger alright!!

This is the reality of exclusive hardware features like this, Nvidia is only interested in pushing it when it suits them and will deprecate it later. With PhysX that's not too major as not many games used it but so many games now use DLSS and RTX fixed function hardware features that will
Completely stop working when Nvidia moves to a new architecture philosophy in future

Again, that's just how it is. I'm not saying it's right in any way, it's not. But they get away with it because they know gamers can do NOTHING about it.

And why do they know that? Because stupid people are more than happy to pay way over the odds for a card so they know they can get away with it.

 

[heatonpkmassive]

Compare this to DX9 support on Arc

Yeah, another good example. And why? Due to the fact that they know they can get away with it. I'm not saying any of this is good, it's not.

But it's how it is unfortunately.

 

[Joxeon]

Will you say it's a nothing burger in 10 years when RTX and DLSS stops working on current games because future Nvidia architecture doesn't have fixed function Tensor cores???

Would be a shame to cut RT support in 10 years when Nvidia is just on the cusp of finally delivering 4k 60fps native in Cyberpunk.

 

[RavenXXX2]

I like going back and playing older titles on new hardware, this makes blackwell even less of a desirable purchase with the plethora of other issues.

Tides are a changing with the 9070 outselling the 5070 massively, nvidia got complacent, Jensen will hate seeing gamers desert green for red. I expect a super refresh with a 5070 16gb.