Whether it is Hopper or Ampere is a little irrelevant at this stage, as it is just a name. What matters more is that every now and then you see a new GPU in some sort of benchmark that is unknown, but is tied to readable specs.
This round our local internet detective rogame stumbled into two new GPU entries in Geekbench, and the specs don't lie, man, one GPU shows 108 CUs (Compute units). Multiply that with 64 shaders per CU (historically what NVIDIA is using), and you'll get to 6,912 shader processors. It probably is a pro-series part as it clocks in at 1.01 GHz and is showing 46.8 GB of memory available. The score: 141,654 points.
But wait dave there's more. An even larger GPU with 118 CUs was spotted as well, that's 7552 shader processors! To compare a bit, the currently fastest desktop product, RTX 2080 Ti has 4352 shader processors. It scores 184,096 in OpenCL where a 2080 Ti scores roughly 129,000 points.
We do know that NVIDIA's next-gen products would be 7nm based, almost doubling space for their transistor budget, so yeah this one is possible. The btw last entry runs 1.1 GHz and is listed at 23.8 GB. The memory numbers are weird we agree. Also, the VRAM sizes obviously indicate something Quadro or Tesla. NVIDIA has been talking a bit like AMD lately, using multiple chip modules per package.
Updated: yet another entry has been spotted. This round it is listed at 124 CUs, which is 7,936 shader cores. The GPU card is clocked at 1.1 GHz and is showing off 32 GB of HBM2e, a score of 222,377 points in the Geekbench benchmark has been noted. BTW everything works in bits and bytes in the world of GPUs, so 124 CUs so perhaps we can round it up to 128 CUs for a fully enabled processor? We'd be passing 8K Shader processors there.