News Nvidia Blackwell GPUs allegedly delayed due to design flaws — launch expected to be pushed back by three months or more

[Admin]

Nvidia is reportedly pushing back the initial deliveries of the B100 and B200 AI Superchips by at least three months to address design flaws.

Nvidia Blackwell GPUs allegedly delayed due to design flaws — launch expected to be pushed back by three months or more : Read more

 

[BearRaid]

Can we start calling the datacenter/AI/server parts by any other name than "GPU"? It gets real confusing, and it's not as if they actually are used to process graphics in any relevant amount.

 

[Alvar "Miles" Udell]

AI Accelerator would seem to fit well.

 

[jlake3]

It feels like that designation might be somewhat model dependent?

The top Turing-based datacenter card did have graphics support and could be used as a cloud GPU, the top Ampere card does not appear to have DX or Vulkan support and is a pure accelerator, and the top Ada-based datacenter card currently listed on TechPowerUp not only has graphics support but is actually noted as being what's running GeForce Now.

GB100/GB200 I think is going to be like GA100 and have no DX/Vulkan support, and thus something like "compute accelerator" may be more accurate, but it's not as simple as "datacenter products aren't graphics cards."

 

[thisisaname]

Nice story to short Nvidia stock?

 

[tamalero]

Nice story to short Nvidia stock?

Someone inform Warren Buffet.. whose company already sold 50% of their apple stake XD

 

[Amdlova]

Nvidia is waiting the TSMC put the new price on market and RAM supplier's make the adjust... After all that they will see what extortion prices they will make! Just buy a graphics now cry later

 

[watzupken]

Delay or not, buyers will be willing to wait. Assuming there are wasted batches of Blackwell chips, I don't believe it will impact Nvidia much since they are charging their customers a very steep margin to begin with. Like I don't believe the cost of R&D and production is half of what Nvidia is charging their customers.

 

[nogames]

Good news, the world will not run out of power this year

 

[DS426]

Isn't word on the street that AMD's Zen 5 launch delay due to model numbers being wrong on boxes and the laser-etching on the lids, e.g. Ryzen 9 7600X? In other words, we're not talking about a chip design defect but a problem in the end stages of finishing and manufacturing, so although it was probably worth mentioning in this article, we're not talking like-for-like here.

 

[CelicaGT]

Maybe it's just the lighting, but why does Jensen's head look massive in this picture?

 

[Taslios]

Can we start calling the datacenter/AI/server parts by any other name than "GPU"? It gets real confusing, and it's not as if they actually are used to process graphics in any relevant amount.

that is one thing about CDNA/RDNA MIxxx/Radeon Instinct/Radeon it is far less confusing as to which product they are talking about when dealing with AMD...

 

[tamalero]

Maybe it's just the lighting, but why does Jensen's head look massive in this picture?

It's his ego trying to leave his skull

 

[ManDaddio]

Well, NVIDIA has to make sure quality control is top notch. These components have to pass a vigorous verification process.

Their name is everything now.

And people talking about the delay killing sales? I doubt that. NVIDIA has a great hardware/software ecosystem now.

If it was so easy creating these things then there would be competition all over the place. Slapping something together that does similar things doesn't create the ecosystem. It just creates a piece of hardware.
NVIDIA just has an extreme headstart on multiple levels.

It will take years for others to catch up.
Even Elon Musk (the supposed AI master) is ordering these things, last I heard.

 

[JRStern]

Can we start calling the datacenter/AI/server parts by any other name than "GPU"? It gets real confusing, and it's not as if they actually are used to process graphics in any relevant amount.

GPT -> GPU, "generative".

 

[StevenTao]

Does anyone know if the defects in production are the result of the interposer architecture?