Take a look at the roadmap detailing upcoming AMD VEGA and NAVI architectures. We don’t have all info regarding the two architectures but have enough to provide you a closer look at what you can expect from AMD during the next three years, regarding professional GPU market.
Radeon Pro S9 Nano
Radeon Pro S9 Nano is R9 Nano made for servers, just under a different moniker. The Pro S9 Nano should sport the same specs as the R9 Nano, and should come out during the next few weeks.
VEGA 11
Details surrounding VEGA 11 architecture are still under a cloud of darkness. One thing is sure; VEGA 11 will replace Polaris 10 architecture during the next year, at least on the professional market. Although AMD will offer new cards based on Polaris 10 architecture, the professional market will see VEGA 11 cards replacing Polaris 10 in the Radeon Pro graphic card series.
Differences between VEGA and PASCAL
Firstly, Vega 10 sports two stacks of HBM2 memory, meaning it will have up to 16 GB of vRAM, double the amount of Polaris 10, and more than Pascal offers. Dual VEGA 10 solutions will have 4 stacks of HMB2, meaning (at least in theory) whopping 32 GB of vRAM, but we doubt that mass consumer cards will come with this amount of memory; four stacks of HBM2 memory will rise bandwidth from 512GB/s to 1TB/s.
Vega 10 will offer 750 GFLOPs of dual-precision computing power (1/16 of SP); VEGA 20 should incorporate a divider of 1/2 SP. Because of this, GPUs offering a solid dual-precision performance will be here during the next year. Vega 10 will probably compete with Pascal in half and single-precision computing.
VEGA11 | VEGA10 | VEGA10x2 | VEGA20 | GP100 TESLA P100 |
|
Available | 2017 | 1H 2017 | 2H 2017 | 2H 2018 ‘ | Already Available |
Architecture | 14nm GFX9 | 14nm GFX9 | 14nm GFX9 | 7nm GFX9 | 16nm Pascal |
Compute Clusters | ? | 64 | 2x 64 | 64 | 56 |
Memory | ? | 16GB 2 stacks HBM2 | 4 stacks HBM2 | 16/32 GB 4 stacks HBM2 | 12/16GB 3/4 stacks HBM2 |
Bandwidth | ? | 512 GB/s | 1 TB/s | 1 TB/s | 540/720 GB/s |
TBP/TDP | ? | 225W | 300W | 150-300W | 250W |
Interface | PCI-E 3.0 | PCI-E 3.0 | PCI-E 3.0 | PCI-E 4.0/xGMI | PCI-E 3.0/NVLink |
Double-precision (Floating Point Performance) | ? | 750 GF (1/16 SP) | ? | 1/2 SP | 4.7 TF |
Single-precision (Floating Point Performance) | ? | 12 TF | ? | ? | 9.3 TF |
Half-precision (Floating Point Performance) | ? | 24 TF | ? | ? | 18.7 TF |
AMD is also constructing an alternative to NVLink (a high-bandwidth, energy-efficient interconnect that enables ultra-fast communication between the CPU and GPU, and between GPUs), called xGMI. It should be incorporated into Naples Architecture, based on AMD Zen and Vega 20, and should see that light of day during the second half of 2018.
For the end, take a look at AMDs Server GPU roadmap below:
2016 | 2016 | 2017 | 2018 | 2019 | |
HPC | HAWAII | HAWAII | HAWAII | VEGA20 | VEGA20 |
DEEPLEARNING | FIJI | POLARIS10 | VEGA10 VEGA10x2 | VEGA10 VEGA10x2 | NAVI10 NAVI10x2 |
INFERENCE | Radeon Pro S9 Nano | POLARIS10/11 | VEGA11 | VEGA11 | NAVI11 |
Source: Videocardz.com