Nvidia unveils its Hopper data centre architecture
Nvidia's GPU Technology Conference (GTC) is underway. During CEO Jensen Huang’s keynote, details of Nvidia’s next generation Hopper architecture were revealed. Though it's an AI and data centre focused GPU, it gives us a few hints of what we can expect from Nvidia’s gaming-oriented Ada Lovelace GPU architecture, which is due for release later in 2022.
The H100 is a major step forward over the current flagship A100. The full GPU contains 80 billion transistors or 26 billion more that the A100. It’s built on a custom TSMC 4nm process. It supports up to 80GB of HBM 3 memory delivering up to 3 TB/s of bandwidth.
The H100 supports PCIe 5.0 and NVLink for connecting multiple GPUs together. It can deliver 2,000 TFLOPS of FP16 and 1,000 TFLOPS of TF32 performance, triple that of the A100. Hopper introduces a new instruction set called DPX. It’s designed to accelerate performance in fields as varied as disease diagnosis, quantum simulation, graph analytics and routing optimizations.
The full H100 GPU includes 18432 CUDA cores and 576 Tensor cores. That compares to the A100 with 8192 and 512 respectively, though for now not all of the cores are unlocked, presumably to maximise yields. The core clocks are also not finalised. Despite being fabricated on such an advanced node, the SXM version of the H100 comes with a TDP of 700W. That’s right, seven. hundred. watts.
Steam Deck review: Our verdictSteam Deck availability: How to get one Steam Deck battery life: The real battery lifeHow loud is the Steam Deck? Say what?The emulation dream machine: The ultimate emulatorThe best budget gaming PC: Price point hero
The H100 is set to be a monster of a card, but is it relevant to PC gamers? The answer is sort-of. H100 is all about compute
Read more on pcgamer.com
