Reports of Bricked NVIDIA GeForce RTX 5090 and RTX 5090D Surge

[Vayra86]

I guess if you compared cores from different generations sure, but if we compared the 5080 to 5090, 4080 to 4090 and the sli performance 680 to 690 the performance was almost linear. Also while you state the core density on the 5090 is 33% higher the bandwidth is unnecessary doubled which is also not stopping people from further oc the over killed vram.
I would request vram speed scaling but it seems Nvidia gave everyone 5 minutes to test the cards so there goes that. I would test it myself but it's a vapor launch rumored to be available in September.
The sli performance in the past scaled perfectly and now a chip with double the cores scales at best to 50%.
Besides the scaling what I realized Nvidia paper launched this by purpose with all the effort that went in to the Blackwell gaming cards launch. Nvidia wanted its competition to perceive that they are still focused on gaming.
Circling back with the cherry on top the proof is in the article that they aren't focused on the gaming cards with no supply and drivers that allegedly brick cards. Is it cognitive or hyper cognitive enough for you?
Now I take a hit of that caffeine. View attachment 383313

Euh... I think there's a language thing going on, but scaling in terms of SLI or in terms of GPU size is the same thing; it is how much return you see on the hardware investment: does the additional amount of shaders you have now (also goes for SLI, doubled shader count) scale linearly, or better, or worse than that? Usually the shader count scaling in single GPU setups is very linear and only limited by lack of resources elsewhere. In SLI setups, there are far more limiting factors and your chart shows that.

You say again: "The sli performance in the past scaled perfectly and now a chip with double the cores scales at best to 50%."
Buddy what?! That's exactly not what we're seeing and what I've said above and exactly NOT what your charts have shown. The core increase from 4090 to 5090 creates an almost linear scaling; its about 30% more cores and you get even more performance in return, in part due to better clocking and faster vram.

SLI performance in the past NEVER scaled perfectly. Even the best SLI performance games are not achieving linear scaling. Which is obvious, because there's more communication between GPUs and they might have to wait for each other. This costs performance. You gave the example of 680 to 690, there is almost no game that gives you +100% perf with the second GPU.

So circling back to the article... sure, we don't disagree on the low- effort gen Blackwell seems to be. But even so, what's there scales fine.

 

[Godrilla]

Euh... I think there's a language thing going on, but scaling in terms of SLI or in terms of GPU size is the same thing; it is how much return you see on the hardware investment: does the additional amount of shaders you have now (also goes for SLI, doubled shader count) scale linearly, or better, or worse than that? Usually the shader count scaling in single GPU setups is very linear and only limited by lack of resources elsewhere. In SLI setups, there are far more limiting factors and your chart shows that.

You say again: "The sli performance in the past scaled perfectly and now a chip with double the cores scales at best to 50%."
Buddy what?! That's exactly not what we're seeing and what I've said above and exactly NOT what your charts have shown. The core increase from 4090 to 5090 creates an almost linear scaling; its about 30% more cores and you get even more performance in return, in part due to better clocking and faster vram.

SLI performance in the past NEVER scaled perfectly. Even the best SLI performance games are not achieving linear scaling. Which is obvious, because there's more communication between GPUs and they might have to wait for each other. This costs performance. You gave the example of 680 to 690, there is almost no game that gives you +100% perf with the second GPU.

5080 vs 5090 has double the cores, 4080 vs 4090 has 67% more cores. Hence why I said let focus on same generation. The scaling is not linear. The 690 vs 680 same generation had almost 100% scaling. Questions?
Update my Organically modified Lovazzza coffee gives me super powers.

 

[Vayra86]

5080 vs 5090 has double the cores, 4080 vs 4090 has 67% more cores. Hence why I said let focus on same generation. The scaling is not linear. The 690 vs 680 same generation had almost 100% scaling. Questions?

Ahhhh core scaling. Not performance. Gotcha. But how relevant is that? If you get 100% more cores and in the better half of all games you only see 150% performance... that's pretty shit, and that's what SLI was doing for us.

Also you have to keep in mind that when it comes to scaling on a 5090, we're likely not seeing the full performance of the GPU yet, especially if you're looking at an aggregated bench with lots of games limiting this GPU, while barely limiting a 5080.

5080 vs 5090 has double the cores, 4080 vs 4090 has 67% more cores. Hence why I said let focus on same generation. The scaling is not linear. The 690 vs 680 same generation had almost 100% scaling. Questions?
Update my Organically modified Lovazzza coffee gives me super powers.

Enjoy the coffee. Its beer time over here

 

[Degreco]

Nvidia is silent because it's not a problem. Paper launch to the rescue: Only a handful of private customers own the thing as we speak.

They shipped most of their 5090s to creators, reviewers and scalpers. So Nvidia can fix this issue silently and comfortably in the next weeks. Reputation saved.
Until then the 5090/D is simply not available.

 

[Godrilla]

Nvidia is silent because it's not a problem. Paper launch to the rescue: Only a handful of private customers own the thing as we speak.

They shipped most of their 5090s to creators, reviewers and scalpers. So Nvidia can fix this issue silently and comfortably in the next weeks. Reputation saved.
Until then the 5090/D is simply not available.

View attachment 383327

AMD was like we aren't competing in the high-end . Nvidia is like what that means they are all in in ai. Let's make a smoke screen to the public that we are still competing in the top end.

 

[brutalame]

Look like the problem come from the lastest drivers : https://www.nvidia.com/en-us/geforc...3/556589/driver-57216-causing-massive-issues/

 

[Scattergrunt]

I'm super late to this crapshoot of a thread but yea, honestly, I had suspected the RTX 5090 (and especially the 5080) were rushed launches and this kinda further adds to that idea. Though honestly, I'm sure NVIDIA will fix it. Doesn't make it excusable, but its kind of lucky that we had a paper launch in the first place. I can imagine a world where the RTX 5000 series didn't start out as a paper launch and alot more people had these brick.. Oh boy, that would be a story.

 

[Random_User]

Spoiler: Some thoughts...

P.S.: Also, the RTX 5000 PCB design is seems like the lazy-ass engineering, because it resembles the enterprize GPUs of SXM socket.
The only amazing thing is indeed the cooler. The engineers did amazing job.

Don't get me wrong. But what is the chance, of these GPUs did not come from the data-centers/servers, which been pulled out, due to heat issues earlier?/SPOILER]