Earlier this week the Radeon Pro 5600M with HBM2 was released, on Apple platforms solely though, which as it turns out is based on NAVI12 with two stacks of HBM2 memory. The first set of benchmarks have surfaced.
So yes, a bit unexpected, but Navi12 had been released slash announced for MAC. As discussed before, this is a multi-chip module based on the 7nm GPU die and two 16 Gbit (4 GB per IC) HBM2 memory stacks. The maximum shader count of the GPU die on the "Navi 14" is not shared, but for this product we see 40 compute units amounting, and that's 2,560 shader processors, 160 TMUs, and likely 64 ROPs (unconfirmed). The base clock of the card runs at 1035 MHz (peak but remember, mobile) with the HBM2 memory purring at 1.54 Gbps (2048-bit bus), so that's 394 GB/s of memory bandwidth. The card would consume just 50 Watt. You're looking at a to be released PC mainstream GPU here folks.
The Radeon Pro 5600M solution was roughly 50 percent faster than the Radeon Pro 5500M and 13 percent behind the Radeon Vega 48 (Apple SKU). Check the Max Tech video for more details, below some screengrabs of the benchmarks.