⋮    ⋮  

AMD Teases Vega GPU Architecture Announcement For 2017 – Powerful Radeon RX 500 Series Cards With Next-Gen Architecture, Coupled With HBM2


AMD has officially teased their next-generation Vega GPU architecture announcement through a cryptic image on Twitter.

AMD "RTG" Radeon RX Teases Vega GPU Architecture Announcement For 2017 - New Year, New Architecture

The teaser was posted by AMD's official Radeon RX twitter page which gives us a hint about what the teaser is all about. With the arrival of 2017, AMD is going to bring forward a new GPU architecture for consumers. This new architecture would be known as Vega and will be part of the latest Radeon RX series family which will be using HBM2 memory to power high-performance, enthusiast-grade graphics cards.

The teaser itself shows a person facing towards a bring red light. We literally don't know what it could mean as this image could be part of a trailer for the upcoming Vega based cards or something better. What AMD does make clear by posting it on their Radeon RX page is that this is the first official teaser for the Vega GPUs that will be arriving early, next year.

AMD Vega GPU Specifications - Those that are Confirmed and Rumored

The AMD Vega GPU will be embedded inside a new generation of Radeon RX series cards. There isn't much room to fill inside the Radeon RX 400 series so my best guess would be a new family for Vega. It certainly makes more marketing sense since a full fledged Radeon RX 500 series lineup sounds much better. The new family will feature several cards based on the Vega GPU, rumor has it we will be looking at both HBM2 and GDDR5/X variants.

AMD Vega GPU For AI Acceleration Features 16 GB HBM2 Memory Running at 512 GB/s - The Instinct MI25 Card

Coming to the specifications, we have already seen two variants of the Vega GPU demoed by AMD. First one is the Instinct MI25 Accelerator which made highlights at AMD's Tech Summit 2016. The AMD Vega 10 GPU has a peak compute performance of 12.5 TFLOPs in single precision mode. It also has twice the packed math or mixed precision compute performance that is rated at 25 TFLOPs. We can assume that this chip has a dual precision performance of 6.25 TFLOPs if it features a 2:1 ratio of single- to double-precision throughput like its competitor.

AMD packs a few additional technologies such as NCU which is being assumed as “Next Compute Unit”. AMD also packs in a high band width cache and controller. The whole card package is rated at less than 300W which is lower than the P100.

The specific product has a memory bandwidth of 512 GB/s on an HBM2 interface. This would mean we are either looking at a 2048-bit memory bus (2 HBM2 stacks) at 1000 MHz or a 4096-bit memory bus (4 HBM2 stacks) at 500 MHz. The capacity of the card is 16 GB and would be utilizing the second generation high-bandwidth memory interface.

AMD Vega 10 and NVIDIA Pascal GP100 GPU Specs Comparison:

GPU ArchitectureNVIDIA PascalAMD Vega
Product MarketTesla (P100.2)Instinct MI25
GPU Process16nm FinFET14nm FinFET
Flagship ChipGP100 GPUVega 10 GPU
GPU DesignSMP (Streaming Multiprocessor Pascal)NCU (Next Compute Unit)
Maximum Transistors15.3 BillionTBD
Maximum Die Size610mm2500-540mm2
Maximum Cores3840 CUDA Cores4096 Stream Processors
FP16 Compute~24.0 TFLOPs25.0 TFLOPs
FP32 Compute~12.0 TFLOPs12.5 TFLOPs
FP64 Compute~6.00 TFLOPs0.75 TFLOPs
Maximum VRAM16 GB HBM216 GB HBM2 (High Bandwidth Cache and Controller)
Maximum Bandwidth720 GB/s512 GB/s
Maximum TDP300W
Launch YearQ2 20161H 2017

AMD Vega GPU For Consumers Features 8 GB of HBM2 Memory Running at 512 GB/s - The Secret 687F:C1 Card

The consumer variant of the Vega GPU was demoed in Doom and put up some fiery performance against an GeForce GTX 1080 graphics card. The card was seen outpacing the NVIDIA solution but we don't have exact metrics at hand. Regardless, it's nice to see Vega GPU being faster than an enthusiast solution based on the NVIDIA's latest Pascal GPU architecture.

Later on, we confirmed that this was the exact GPU that showed up in AOTS database earlier this month.This variant packs 8 GB of HBM2 and is more focused at desktop PCs rather than HPC market. Performance of this card was on par with GeForce GTX 1080 but drivers have not been optimized for it yet so we can expect even faster performance from the final variant.

AMD 687F:C1 DirectX 11 / DX12 Tests (1080P/1440P):

There are four benchmarks that were conducted with DirectX 11 API mode. These consist of Crazy, Standard and Low settings. The only score that has actually been compared to other GPUs was the Standard 1080P DX 12 test. This shows the card scoring 8400 points which is in the range of what GeForce GTX 1080 scores. The score was listed at 131th position before it was deleted. The fact that it was deleted could indicate that it was not supposed to be shown to public.

The benchmarks also shows the card performing next to a high-end GTX 1080. Currently, there's no GPU from AMD that comes this close in the Polaris Radeon 400 series lineup. If that's not enough, AMD also demoed the Vega GPU at New Horizon event and showed the card running Star Wars Battlefront at 4K with greater than 60 FPS which was higher than the refresh rate of the monitor displaying the game.


AMD Vega Lineup

Graphics CardRadeon R9 Fury XRadeon RX 480Radeon RX Vega Frontier EditionRadeon Vega ProRadeon RX Vega (Gaming)Radeon RX Vega Pro Duo
GPUFiji XTPolaris 10Vega 10Vega 10Vega 102x Vega 10
Process Node28nm14nm FinFETFinFETFinFETFinFETFinFET
Stream Processors40962304409635844096 (?)Up to 8192
Performance8.6 TFLOPS
8.6 (FP16) TFLOPS
5.8 (FP16) TFLOPS
~25 (FP16) TFLOPS
22 (FP16) TFLOPS
>25 (FP16) TFLOPS
Memory Bus4096-bit256-bit2048-bit2048-bit2048-bit4096-bit
Launch20152016June 2017June 2017July 2017TBA

This card really looks to be top-notch and built for enthusiasts and we can't wait to learn more about it. AMD will obviously be giving us more details of the Vega GPU at CES 2017 and the new RYZEN CPUs which are also coming out in Q1 2017.