From his now famous kitchen, Jensen Huang recently showed the world what he was cooking up for the future of Nvidia. Despite the name, sadly gaming graphics is no longer what the Graphics Technology Conference (GTC) is about. Instead, there was a lot of focus on AI, data centres, ‘accelerated computing’ and ‘intelligent networking’, with graphics served as a side dish.
However, in discussion of Nvidia’s next-gen server chips, Jensen detailed a roadmap of CPU, GPU and DPU (essentially network processors) architectures, which showed a two-year cadence of each, with alternating years of CPU and GPUs. Last September, Nvidia’s Ampere GPUs launched, and this year, its Arm v9 architecture hits the streets.
That means the next generation of GPUs is likely to land around the third quarter of 2022, which is good, since no one can buy a graphics card this year anyway. It’s probable that Nvidia will drop a mid-cycle GPU refresh with speed bumps – as with the RTX 2000 Super series, but if and when that happens will be down to when it can actually make them.
“The next generation of GPUs is likely to land in the third quarter of 2022”
My expectation is that Nvidia will jump from using the current 8nm process for its current RTX 3000-series GPUs to a 7nm process for its mid-cycle refresh, then next year’s ‘RTX 4000-series’ GPUs will be built on a 5nm node. The differences between eight, seven and five may not sound like a lot, but they represent huge leaps in transistor density, going from ~45 million per millimetre (8nm) to ~95 million (7nm) and then to ~175 million (5nm).
It’s unlikely Nvidia will use this opportunity to lower the enormous power use of its high-end GPUs, but rather use it to pack in more RT, Tensor and shader cores. Alternatively, Nvidia could opt to spend some of its transistor budget on other tech, such as big caches, echoing this design choice at AMD (Infinity Cache) and Intel (Rambo Cache).
“This would be a huge gaming boost for Chromebooks…”
We need to watch Nvidia’s CPUs too. In a low-key announcement, it stated that it was working in partnership with MediaTek to bring its RTX 3000 graphics to Chromebooks.
MediaTek is currently the largest non-x86 Chromebook chipmaker, and given the current litigation between Nvidia and Qualcomm over the former’s potential Arm acquisition, this partnership makes sense. Nvidia shared a slide indicating that its GPUs would work alongside MediaTek’s chips in a CPU-GPU paring, as with a traditional laptop, rather than putting RTX tech into a single chip.
This would be a huge gaming boost for Chromebooks, and it could also represent the first steps Nvidia is taking to bring its RTX tech into Android smartphones in future. Chromebooks have been a success in education and office environments, because they’re often cheaper and have longer battery life than their Windows counterparts, so unlocking the gaming market will make Microsoft, Intel and AMD feel the heat.
Nvidia has outlined that it will make its own server processors on a two-year cadence, but it could also open the door to its own high-performance PC/laptop CPU based on a common core that competes with Intel Core and AMD Zen. It would be seriously awesome to see a third option on the market, but I doubt it will happen. If GTC reflects Nvidia’s ambitions, then AI and data centres are its main focus now, rather than PCs and gaming.