NVIDIA GeForce RTX 5080 Spotted with Missing ROPs, Too

[lilhasselhoffer]

So...let me get this straight. Nvidia immediately comes up with the 0.5% figure. They immediately cite three models. It's almost like they already knew that some percentage of "good" product that they released was in fact testing badly. At that failure rate, we are looking at 5 in 1000, or one in 200. That doesn't sound like a lot...but these are thousand dollar+ cards that just have some number of ROPs that don't work. If I'm cynical, they are looking at their AI frame generation to make-up for this failure, effectively "hiding" it behind what could theoretically be silly levels of interpolated frames, meaning instead of 1 real in 6 frames their algorithm might have to generate 1 real in 5 frames...but the difference of effectively 220 fps and 180 fps would be indistinguishable on a 120 Hz refresh rate monitor.

While I won't defend the scuminess, I do think that someone somewhere along the chain knew the fallout was happening and thought it'd be a reasonable risk. I also think that the 0.5% figure is shenanigans. When you hear half a percentage you are prone to not consider it your problem. If they announced that 2% of units would experience drops then everybody would test...because that's 1 in 50 people. Given that they have no legal obligation to actually disclose failure rates expressly to anyone (including their shareholders if this is slid into their nominal silicon failure rates), then we'll never actually know how badly they tried to screw consumers.

 

[TheGeekn°72]

So...let me get this straight. Nvidia immediately comes up with the 0.5% figure. They immediately cite three models.

They only cited the 5090(D) and the 5070Ti and then a 5080 with the same issue popped up

It's almost like they already knew that some percentage of "good" product that they released was in fact testing badly.

At this point into the shit show, I wouldn't put it past them that they had a terrible yield (due to some X factor ?) and knew what they sold/provided to their AIBPs and kept their mouth shut hoping no one would notice (or that it wouldn't even show up on diagnostic tools)

Either that or people were supposed to initially get the dies with the lower ROP count and the higher ROP ones were supposed to be Super variants ? At this point, unless there's an insider leak, no one can know...

 

[AnotherReader]

Wonder how much does the pure architecture improvement help between Ampere and Ada. Like IPC and other stuff beside just raw numbers compared.

There's no IPC increase for the shaders. All of the improvements are due to very high clock speeds relative to Ampere.

 

[Darc Requiem]

@lilhasselhoffer I think Nvidia knew about the defective dies, shipped them anyway, and expected minimal blow back. If this was the only issue with the 50 series, they may have been right but this is one of several. I don't think they'd be dumb enough to think no one would notice. The type of consumer that buys a new GPU day one is more likely to notice problems.

 

[R0H1T]

So...let me get this straight. Nvidia immediately comes up with the 0.5% figure.

Don't be surprised if that figure is "in fact" also a lie

 

[lilhasselhoffer]

@lilhasselhoffer I think Nvidia knew about the defective dies, shipped them anyway, and expected minimal blow back. If this was the only issue with the 50 series, they may have been right but this is one of several. I don't think they'd be dumb enough to think no one would notice. The type of consumer that buys a new GPU day one is more likely to notice problems.

Don't be surprised if that figure is "in fact" also a lie

I...think both of you might have missed the point where I say that exact thing. Specifically, they released knowing it was bad but assuming that "only 8 ROPs missing" would be covered up by the performance added by the new features of DLSS4. IE, in a vacuum there's be no perceived loss of performance because nobody would be checking them.


Tin foil hat time, I'd suggest that 3-7% of their released hardware probably has this bug based upon reporting. That number isn't coming from my backside. That's a 1 in 200 going to a 1 in 34 or a 1 in 14. That fits in line nicely with Nvidia sampling their stock significantly less than advertised, and someone in their QC way underestimating failure rates. This corresponds to about a 1 off estimation of failure rates for planning (RPN #)...that could also be used to justify protections to save money all the way around. It reeks of quality saying "good enough" and launching...which I'd suggest is 100% because Nvidia knows whatever garbage they put on the market will move because they've created a situation where there are no good alternatives. Isn't it great to be king of crap mountain...until you start sinking in....

 

[R0H1T]

I didn't I'm just highlighting it, remember the GTX 970 issues & how/when it was reported? Now do you know how many users got partial/full refund for that especially outside the US? Where they actually had to pay something to settle ~ I'm betting close to zero or at most a rounding error

Nvidia are basically relying on user ignorance/lax consumer protection laws to get away with another highway robbery & their idiot fanboys keep defending them

 

[docnorth]

Don't panic, Blackwell refresh will fix it

That would be Ada++.

 

[Fluffmeister]

For an apparent paper launch, plenty are missing ROPs already.

 

[theouto]

Funny how nvidia only said that they investigated and found irregularities in manufacturing with 5080 chips only after the news came out. it reads something like this:
"Some 5080 GPUs have also been found with missing ROPs"
"Oh yeah, yeah we knew that too actually, we fixed it, yeah trust us yeah yeah".

How will this launch get worse? Stay tuned for the release of the RTX 5070, where things cannot possibly get worse.

 

[chrcoluk]

GN says the factories he visited check every single chip for this stuff so he has a very hard time believing it was accidental. I do suspect though he was lied to on his tour, and they just told him they check every single chip, as someone who has worked in manufacturing, that would be seen as unviable, its just too costly to check every single unit you make. In every factory I worked in, if we knew someone "special" was visiting then normal practices went out of the window for that shift, suddenly everything had to be clean, production was slowed down to allow workers to check everything they processing and so forth, then when its all done back to normal.