Nvidia’s biggest customers for AI chips, including Microsoft, Amazon, Google and Meta, have reportedly cut their orders for the new Blackwell series of GPUs due to overheating issues. However, that doesn’t worry us for Nvidia’s new RTX 50 family of gaming GPUs, despite them also being based on the Blackwell architecture.
The Information (paywall, via Reuters) claims that overheating and other related “glitches” have caused customers to delay Blackwell orders or switch to Nvidia’s earlier and presumably less problematic “Hopper” generation of AI-optimized GPUs.
Back in October, Nvidia’s head honcho Jensen Huang has admitted that the new Blackwell series of AI GPUs had “design flaws” which was leading to delays in shipping AI racks to customers.
“The design flaw caused the yield to be low. It was 100% Nvidia’s fault,” Huang said. Now, those comments didn’t directly refer to any overheating, instead talking about yields, which typically entails the number of functional chips that can be harvested from a larger wafer containing many GPU dies.
So, those flaws could be completely separate from whatever is causing the alleged overheating. Or they could be related on some level.
Indeed, overheating Blackwell GPU stories have been circulating since last November, with The Information reporting that Nvidia liquid-cooled racks containing 72 Blackwell GPUs had been redesigned several times in an attempt to solve the issue.
Whatever, what we can say for sure is that there’s no reason to assume that any issues with Nvidia’s AI GPUs will translate into problems for its new RTX 50 family of gaming chips just announced at CES.
Best CPU for gaming: The top chips from Intel and AMD.
Best gaming motherboard: The right boards.
Best graphics card: Your perfect pixel-pusher awaits.
Best SSD for gaming: Get into the game ahead of the rest.
Yes, those too are based on the Blackwell architecture and indeed built on the same TSMC N4 silicon. However, in terms of actual layout and also the number and balance of functional units, the gaming chips are totally different.
Admittedly, it’s possible there’s a problem, say, with the design of Blackwell’s Tensor cores that might also map to the gaming GPUs. But the odds of that are probably pretty slim. Moreover, the workloads and software that runs on a gaming chip is completely different to training or inferencing an AI model.
It’s also worth noting that it’s the same source, namely The Information, pushing this overheating narrative again. Some wider confirmation would add weight to the story. Still, by Nvidia’s own account, there have been problems with Blackwell.
We’ll know soon enough just how fast and indeed how hot Nvidia’s new RTX 50 gaming GPUs really. Watch this space for our full reviews.