AMD’s Next Gen Vega 10, Vega 11 And 7nm Vega 20 GPUs Detailed In Specs Leak

AMD

The real battle for GPU supremacy will take place when AMD ships its Vega GPUs starting sometime next year. Yes, AMD only recently announced Polaris, but that's not the architecture that's supposed to compete with NVIDIA's Pascal GPUs, at least not at the high end. That's where Vega comes into play, and a new leak shines light on some specific details about AMD's next big GPU architecture.

Previous leaks suggested Vega was inbound for a release sometime in the first half of 2017, and probably earlier than later. Those supposed roadmaps also showed Vega using second generation high bandwidth memory (HBM2). According to the latest bit of inside information from a "very good source" who spoke with Videocardz—the same source who confirmed mobile Polaris specs weeks before it was released—that's still true, with Vega 10 slated to arrive in the first quarter of 2017.

The unnamed source says Vega 10 will arrive with 64 Compute Units, 24 TFLOPs of 16-bit computing power, and a 225 TDP. Assuming AMD sticks with 64 compute cores per compute unit, as is the case with GCN, we can extrapolate from that 4,096 cores for a fully functioning Vega 10 GPU.

AMD GPU Roadmap
Information about Vega from a new leak aligns with an old leaked roadmap

Vega 10 will be built on a 14nm FinFET manufacturing process. It will sport 16GB of HBM2 memory, serving up 512GB/s of memory bandwidth. That's a bit lower than we were expecting—HBM2 supports up to 256GB/s of memory bandwidth per stack, and if using 4GB stacks, that would work out to 1TB/s. It seems like a missed opportunity to us if AMD holds back here.

Finally in regards to Vega 10, the source indicated a dual-GPU variant with a 300W TDP slated for the second quarter of next year.

Perhaps around the same time, Vega 11 will arrive to replace Polaris 10 in the mainstream market. Specs aren't known yet, though the assumption is that Vega 11 will also be built on a 14nm FinFET manufacturing process. It also would presumably run HBM2.

Looking further down the road, Vega 20 will show up to the scene sometime in 2018, barring any problems by GlobalFoundries with manufacturing 7nm FinFET chips. Vega 20 is said to have 64 Compute Units and 32GB of HBM2 with 1TB/s of memory bandwidth. It will also boast PCI-Express 4.0 support and have a TDP of 150W.