Skip to main content

Nvidia’s new GPUs are already running into problems

Nvidia Blackwell chips.
Notebook Check / Nvidia

Nvidia’s latest Blackwell GPUs are running into problems in the data center, reports The Information. According to the report, Nvidia’s customers are worried about how well the AI accelerators will hold up, as overheating issues have caused delays in server racks being deployed for AI training.

The Blackwell architecture is at the heart of both Nvidia’s next-gen AI accelerators and its upcoming RTX 50-series graphics cards. In the data center, the architecture was previously delayed due to “design flaws,” pushing the deployment of the B100 and B200 GPUs back. That’s despite big orders with AI players like Meta, Microsoft, and Google.

Recommended Videos

According to the report, the big problem in the data center stems from cramming 72 of the AI accelerators together in a server rack, which has led to overheating problems. Reuters reports that Nvidia has asked suppliers to redesign the server racks “several times” in order to get around the overheating issues.

Get your weekly teardown of the tech behind PC gaming
Check your inbox!

Blackwell is a signficant step for Nvidia. It’s at the heart of the next generation of GPUs, which could earn spots among the best graphics cards. Blackwell is also a point for Nvidia to cement its lead ahead of AMD. Team Red has already deployed its MI300X AI accelerator in data centers, and it’s currently rolling out its MI325X accelerator as it prepares next-gen AI chips.

Nvidia claims Blackwell is able to train large language models at 25 times lower cost and energy consumption compared to its last-gen Hopper architecture, or that it’s able to train these models up to 30 times faster. That kind of speed-up has a big impact on heat, which is already an issue that data centers need to deal with when it comes to AI accelerators.

It could have implications for RTX 50-series GPUs, too. Although we know cards like the RTX 4090 are incredibly efficient when it comes to gaming, Nvidia’s previous flagship still ran into issues with high power consumption and melting power connectors. The latest speculation is that a card like the RTX 5090 could push power requirements further, up to 600 watts. Corsair also confirmed that Nvidia’s next-gen graphics cards will stick with the 12V-2×6 connector that’s been at the center of melting issues on the RTX 4090.

Gamers won’t be cramming 72 RTX 5090s inside a PC, but the scale of overheating issues is different between a data center and a desktop PC. If the Blackwell architecture is running into these issues in the data center, it could spell trouble for Nvidia’s desktop range.

For now, all we can do is wait. Nvidia is expected to reveal its RTX 50-series GPUs in January at CES 2025. Recent reports suggest Nvidia is winding down production of its RTX 40-series cards, most likely clearing the way for next-gen options.

Jacob Roach
Lead Reporter, PC Hardware
Jacob Roach is the lead reporter for PC hardware at Digital Trends. In addition to covering the latest PC components, from…
Windows 11 Recall finally arrives, but with one new problem
Recall screenshot.

The Windows 11 Recall feature has been troublesome since its announcement. Now that the feature is available in a testing capacity, it is still causing users some issues.

Tech reviewers testing the feature have observed that Recall will now fail to save the snapshots that allow the function to work. CNBC noted that it may take “several minutes” for a snapshot to save, which may leave delays in the AI processing. Tom Warren of The Verge, noted on Bluesky that snapshots were not saving at all in his experience.

Read more
You can now buy eight RTX 5090s for over $50,000
Comino Grando server.

It's true that we're all waiting for the release of Nvidia's best graphics cards, but ... not like this. The RTX 5090 made a surprise appearance in a GPU server made by Comino, a company that deals with data center design and construction. Although still unannounced by Nvidia, the GPU can now be preordered directly from the manufacturer, but not by itself -- and at a whopping price of around $5,000 per GPU.

Just to clarify: No, Nvidia's RTX 50-series is still not here, and as of yet, Nvidia hasn't even confirmed that there will be an RTX 5090 at all. However, Comino opened up preorders for its next-gen Grando Server equipped with either six or eight RTX 5090 GPUs. This is worrying news for gamers because the RTX 5090 was always meant to be a consumer card. Still, the reality is that the RTX 4090 remained in high demand throughout its tenure because of its AI capabilities, and the RTX 5090 is only going to be even better in that regard.

Read more
Nvidia may have found a new way to bypass GPU export restrictions
The RTX 4090 graphics card sitting on a table with a dark green background.

Chinese gamers are expected to have limited access to Nvidia's best graphics card due to strict export restrictions. However, Nvidia may have found a way around it without cutting down its flagship GPU. A new leak suggests that the RTX 5090D will have the same hardware specifications as the worldwide version, and the solution lies in firmware adjustments.

As a quick refresher, the "D" in RTX 5090D stands for "Dragon." It marks GPUs made by Nvidia to bypass the export restrictions imposed by the U.S. on China, which limit the sale of high-performance graphics processors. We first saw it appear in the RTX 4090D as a remedy for the fact that the base RTX 4090 exceeds the performance thresholds set by these regulations.

Read more