At this point, NVIDIA’s upcoming Ampere graphics architecture isn’t even remotely a surprise. However, questions still remained about how similar or different the consumer-facing Ampere architecture would be compared to the enterprise-oriented GA100 that Team Green unveiled today.
After the NVIDIA CEO, Jensen Huang formally confirmed the existence of GeForce Ampere Gaming GPU for consumers (think the GeForce RTX 3080 Ti, GeForce RTX 3080, and GeForce RTX 3070), we got some juicy details on how consumer Ampere GPUs would stack up to the big daddy 8192 CUDA core GA100 workstation behemoth. When a journalist asked NVIDIA about consumer Ampere GPUs, the reply was that “there’s great overlap in the architecture, but not in the configuration.”
This tells us a few things. For starters, sadly, it’s not likely that we’ll see an 8192 CUDA core Ampere GPU for consumers. Even the GeForce RTX 3080 Ti will most likely be significantly smaller than the GA100. (On the other hand, with 7nm yields improving over time, we wouldn’t rule out NVIDIA pulling a “SUPER” on us on that front.) However, unlike with Volta, which was a completely different architecture mostly limited to the consumer space, consumer and enterprise Ampere will be functionally the same. We still haven’t heard much from NVIDIA about the consumer chips but TL:DR, they’re a lot like the GA100, but smaller.