Welcome to the next generation of GPUs. During his keynote address at the GTC 2022 on Tuesday morning, Nvidia CEO Jensen Huang officially unveiled “Hopper,” the graphics architecture that powers the company’s next generation of products, and it looks like an absolute monster—at least in the form of a data center.
Hopper’s debut is shaped by the data center-centric Nvidia H100, an obnoxious chip that uses 80 million transistors, the first use of next-generation HBM3 memory, and an advance in the company’s fast NVLink interconnection technology. Hopper is “our generation’s biggest leap ever,” Huang proudly declared.
The introduction of new builds into the GTC is equal to the Nvidia cycle; Ampere, Volta, and Pascal also rose to the spotlight there. And as it’s been revealed in the past, while the H100 is highly specialized for data center uses, with hardware laden with tense cores designed to optimize deep learning tasks, that doesn’t mean that humble PC gamers can’t conclude Some Information from Huber’s AI-based Detection.
Here are three main things Nvidia’s Hopper architecture means for the next generation of GeForce lineup.
1. Hopper jumps to TSMC 4nm
Nvidia RTX 20-series “Turing” GPUs are built at TSMC, a factory that also makes chips for AMD, Apple, Intel, and nearly all of the high-tech companies. But the Ampere GPUs inside the GeForce RTX 30 series switched to Samsung’s 8nm process instead. They gave a ferocious performance, but they also drew massive amounts of strength.
For the Hopper – or at least the H100 – Nvidia returns to TSMC. This chip is built using a custom version of the TSMC 4N process, which is a more power-friendly version of the TSMC N5 process. AMD’s RDNA 2 Radeon RX 6000-series graphics cards have pulled the power-efficiency crown from Nvidia, and it looks like the next generation of these competitors will all be manufactured by TSMC. That’s not a guarantee – the Ampere-based A100 was also built at TSMC, unlike its RTX 30-series siblings – but the current wave of rumors is all there is to it. The game started.
nvidia
2. PCIe Gen 5 (and NVLink 4)
AMD’s Radeon RX 5000 series cards were the first in PCIe 4.0, but Nvidia overtook them to reach the center of the PCIe 5.0 pole. The H100 is the first graphics processor built with faster interconnect technology, and since Intel’s 12th-generation processors recently brought PCIe 5.0 support to mainstream desktops, there’s every reason to believe Hopper will include it when the architecture appears in their Next-Gen GeForce GPUs.
Hopper also marks the emergence of the fourth generation of NVLink, Nvidia’s high-bandwidth interconnection. The RTX 30 series assigns multiple GPU configurations to the flagship GeForce RTX 3090 alone. Assuming Nvidia continues to offer multi-GPU support in its flagship GeForce cards, you’ll notice a significant speed boost with NVLink 4. The H100’s NVLink 4 implementation provides 900GB/s of bandwidth, Nvidia says, compared to the 600GB/s that NVLink 4 provides. Provided by Ampere – based on the A100 and 3rd generation NVLink offerings.
The H100 is incredibly optimized for data center tasks, and Nvidia (oddly) has not released any information about CUDA core numbers, clock speeds, or GPU block sizes — all of which will provide much deeper clues to what to expect from Hopper-powered GeForce cards. But the company’s specifications Act Reveal shows that Nvidia isn’t afraid to evolve with Hopper.
The groundbreaking Hopper graphics processor inside the H100 main filament is stunning 80 billion Transistors — Big Leap after 54.2 billion transistors in A100-based amps, approx four times Several transistors like 21.1 billion lurk inside the Volta-based V100.
nvidia
Pushing all those transistors to peak performance requires a Much It’s strength though. Nvidia says the H100 draws up to 700 watts, which is 300 watts more than the fastest A100 chip. Holy Molly. this, no necessarily It means that Hopper-based GeForce GPUs will be gas consuming as well, but current rumors suggest that next-generation Nvidia graphics cards will consume a lot of power. The H100’s sighting of plenty of wattage lends credence to that claim, although we’ll have to wait for the official reveal of the Hopper-based GeForce graphics card to be sure.
We still want to see it
Almost everything. GTC reveals previous Nvidia builds that provided much more detail about base count and clock speed, giving a much deeper insight into what to expect from their consumer-focused GeForce cousins. The H100 is set to hit the streets in the third quarter, and we expect to hear more about the next generation of GeForce GPUs bringing Hopper to gamers sometime then too. Stay tuned!