AMD is set to launch its next-gen Vega graphics cards in May, with an event just before Computex - just as the Polaris-based Radeon RX 400 series cards were unveiled in Macau, just before Computex 2016. If you haven't seen my '5 reasons why AMD's next-gen Vega is going to kick ass' article, you should - and then come back here to enjoy the rest.
We're hearing that AMD will launch its enthusiast SKU of the Vega 10-based GPU, powered by 8GB of HBM2 memory. 8GB might not sound like much, but if you want to read how it will be impressive - I have teased a little about High Bandwidth Cache (HBC) - which is a large part of how AMD is going to dominate the GPU game in 2017, and beyond.
As for pricing, I think we're going to be looking at around $799-$899 for the Vega 10 graphics card with 8GB of HBM2, which should slot right next to my expected $899 pricing on the GTX 1080 Ti from NVIDIA. We should expect reference cards from AMD that are deliciously sexy, and very powerful - offering 4K 60FPS performance from a single GPU.
Vega 10 Specifications
This is where AMD should fight against NVIDIA in the GTX 1070, GTX 1080 and upcoming GTX 1080 Ti.
GPU: 14nm Vega 10 (64 NCUs)
Performance: 12 TFLOPs of single precision performance (750 GFLOPS of double precision)
GPU clock speeds: 1465MHz on reference, 1600MHz+ on AIB partner cards
RAM: 8-16GB of HBM2 (512GB/sec bandwidth)
PCIe: PCIe 3.0 x16
TDP: 225W
Release: May/June 2016
Dual Vega 10 Specifications
Reduced clock speeds, but dual Vega 10 GPUs - 4-stack HBM2, so 1TB/sec memory bandwidth (16-32GB)
GPU: 14nm Vega 10 (64 NCUs) x 2 (128 NCUs total)
Performance: 24 TFLOPs (12 TFLOPs per GPU) of single precision performance (1.5 TFLOPs of double precision)
GPU clock speeds: 1200MHz+ reference only
RAM: 16-32GB of HBM2 (1024GB/sec bandwidth)
PCIe: PCIe 3.0 x16
TDP: 300W
Release: Late 2017
Vega 20 Specifications
Professional/workstation card, the Titan X of AMD's Vega line up in a way. Shaders are set up differently on Vega 20, so we should see some exciting things with the shift to 7nm and the compute performance. If we do hit 20TFLOPs+ of single precision, we're in for a world of performance on the workstation side of things before Navi hits in late-2018, early 2019.
GPU: 7nm Vega 20 (64 NCUs)
Performance: Interesting - 35-40TFLOPs half precision (20TFLOPs+ of single precision)
GPU clock speeds: Faster clock speeds, nothing concrete
RAM: 16-32GB of HBM2 (1024GB/sec bandwidth)
PCIe: PCIe 4.0 x16 host connection
TDP: 150-300W
Release: First half 2018
Navi 10 Estimated Specifications
Navi will be arriving in a single- and dual-GPU card with Navi 10 and Navi 10 x2, with the dual-GPU variant set to be an absolute monster.
GPU: 7nm Navi 10 (128 NCUs)
Performance: 50TFLOPs+ half precision, 30TFLOPs single precision
GPU clock speeds: Unknown
RAM: 16-128GB of "Nexgen Memory"
PCIe: PCIe 4.0 x16 host connection
TDP: 150-300W
Release: 2019+
Dual Navi 10 Estimated Specifications
This is just nuts, maybe some of this will hit - with my estimates, it'll be interesting to come back to this in 2 years time.
GPU: 7nm Navi 10 (128 NCUs) x 2 (256 NCUs total)
Performance: 100TFLOPs+ half precision, 60TFLOPs single precision
GPU clock speeds: Unknown
RAM: 32-256GB of "Nexgen Memory"
PCIe: PCIe 4.0 x16 host connection
TDP: 150-300W
Release: 2019+