TECH TALK What We Can Expect Next from Nvidia
WHAT DOES NVIDIA have planned for its next round of graphics cards? The Turing architecture is coming up on its second birthday, and after the Super refresh, everyone is looking to 7nm and the Ampere architecture. At least, I think it’s called Ampere— Nvidia is being tight-lipped, as usual. To be clear, anyone claiming to have specifications for the RTX 3080 right now is making stuff up. But even if we don’t know the exact specs, looking at the underlying tech—TSMC’s and Samsung’s 7nm process technologies—can provide a good sense of what Nvidia is likely to do.
Nvidia GPUs build on each preceding generation, adding features and improvements. Turing is like Pascal, with new ray-tracing and Tensor hardware, concurrent FP and INT execution, an improved NVENC block for video encoding, and some other tweaks. Pascal is like Maxwell, with tweaks to the architecture and some new features; Pascal also marked a transition from 28nm lithography to 16nm FinFET. Maxwell built on two generations of Kepler GPUs, with better memory compression and some VR-specific features that didn’t catch on.
This next transition is going to be the most like the Maxwell to Pascal move, because it’s not just about tuning the architecture. The next GeForce series will shrink the process from 12nm to 7nm, paving the way for big improvements. I also expect Nvidia to be less ambitious on pricing, as the RTX 20-series didn’t sell as many units as anticipated. The lack of interest from cryptocurrency miners probably had a lot to do with that.
Nvidia isn’t likely to do a straight die shrink; instead, it will pack more features and hardware into Ampere GPUs: more RT cores for ray tracing, more CUDA cores for general graphics, and more Tensor cores for machine learning. There will also probably be new tech to improve performance, as Nvidia does that every generation— stuff like variable rate shading.
Looking to AMD’s 7nm parts gives a good idea of how the 7nm process scales. The Vega 20 in the Radeon VII is about 30 percent smaller than Vega 10, while packing in 6 percent more transistors. Even better, the Navi 10 GPU is almost half as large as Vega 10, and the 5700 XT easily outperforms the RX Vega 64, while using 225W instead of 295W. Some of that is architecture, but lots comes from TSMC’s 7nm lithography.
Nvidia can already beat AMD’s RX 5700 XT, and its Ampere RTX 3080 GPUs will be even faster. It’s easy to imagine Nvidia putting 20–25 percent more cores in each level of GPU hardware. The 7nm process should allow Nvidia to clock those cores 20–25 percent higher, too, so Nvidia could have GPUs consistently running at over 2.0GHz. Even with the added cores and features, die sizes should be smaller than Turing equivalents.
We could be looking at a generational improvement in performance of 40-50 percent or more for RTX 3080, with a chip that’s still 15 percent smaller than the RTX 2080. What’s more, we might not have to wait too long.
TSMC has been making 7nm chips since 2018, when the Apple A12 launched. AMD started selling 7nm in late 2018, and has been selling lots of 7nm CPUs and GPUs since July 2019. The time is ripe for Nvidia to launch its first 7nm parts, and past experience suggests that spring 2020 would be ideal.
If you still have a GTX 10-series or similar, or even a GTX 900, you may finally find a reason to upgrade. And wouldn’t it be perfect if Nvidia launched RTX 30-series parts, with far better raytracing performance, in time for
Cyberpunk2077? Maybe we’ll have
to wait for Vampire:Bloodlines2, but Cyberpunk is the bigger target.
Jarred Walton has been a PC and gaming enthusiast for over 30 years.
Nvidia GPUs build on each preceding generation, adding features and improvements.