Nvidia's new GPU is so powerful that its interconnect is almost the size of an RTX 3090 chip

Nvidia’s new GPU is so powerful that its interconnect is almost the size of an RTX 3090 chip

Nvidia’s new GPU is so powerful that its interconnect is almost the size of an RTX 3090 chip

Nvidia’s new Hopper server GPU architecture has spawned a massive chip. I’m not even talking about the 80 billion transistor H100 graphics processor itself. No, the new NVSwitch embedded in the DGX 100 server system reportedly has around 25 billion transistors – which is only slightly less than the number inside the GA102 GPU inside the RTX 3090.

The third-generation NVLink switch chip of the Hopper generation is a monster in itself. It is also manufactured at TSMC 4N and has 25 billion transistors. The switch provides 64 NVLink ports. The predecessor had 2 billion transistors, was fabricated on TSMC 12 nm and offered 18 ports. 25, 2022

see more

The NVSwitch chip is an interconnect that allows multiple instances of NVLink (a direct chip-to-chip connection) to communicate and enables many Nvidia graphics cards to work together. This means they can pool resources to complete complex calculations faster than a single GPU or even two GPUs by themselves.

Think SLI, but with a lot of chips connected, and, you know, works.

Built on the same TSMC 4N node as the H100 chip itself, this 3rd-generation version can directly connect up to 8 GPUs with a chip-to-chip bandwidth of 900GB/s and a total bandwidth of 7.2TB/s.

The reason I’m really quoting these numbers is because they’re too big, not because they have any impact on us or PC gaming. Hopper architecture is an architecture designed purely for server environments, and NVSwitch interconnects are even more so.

When SLI and CrossFire truly became a thing, discrete GPUs didn’t play well together, and true multi-GPU processing would only exist in a more compute-centric environment. But even there, you need very high-speed links to take full advantage of these high-speed processors, which is presumably why Nvidia has so many damn 4nm transistors in NVSwitch.

(Image credit: Nvidia)

For reference, the largest chip in the GeForce space, the GA102 GPU, has 28.3 billion transistors, billions more than NVSwitch. On the AMD side, Navi 21 GPUs running the RX 6900 XT, 6800 XT and RX 6800 are close to 26.8 billion. But it’s still far more than the 17.4 billion transistors of the GA104 chip that powers the RTX 3070 Ti, and even some RTX 3060 cards.

So yes, a big chip.

But we’ll likely see more big chips in next-gen graphics cards, and the red team will bring back the idea of ​​multi-GPU gaming. Not necessarily the same as CrossFire, but with a chiplet design more like its Ryzen processors, you’ll be on the right track.

The key to multi-GPU gaming is to have the OS see it all as one chip, and it looks like AMD has figured out an internal GPU interconnect that can do that and made its top-of-the-line Radeon RX 7000 series cards capable of running the multi-chip design. And it will likely exceed the expected monolithic designs of Nvidia’s future Lovelace-based RTX 4000-series cards.

Discover more articles in our categories Gaming & News et encore Anime.

Thanks for visiting we hope our article Nvidia’s new GPU is so powerful that its interconnect is almost the size of an RTX 3090 chip

, we invite you to share the article on Facebook, pinterest and whatsapp with the hashtags ☑️ #Nvidias #GPU #powerful #interconnect #size #RTX #chip ☑️!

Bart Thompson
Bart is's List Writer . He is from Houston, Texas, and is currently pursuing a bachelor's degree in creative writing, majoring in non-fiction writing. He likes to play The Elder Scrolls Online and learn everything about The Elder Scrolls series.