More details regarding AMD's next-generation Vega GPU have come online showing that we will be looking at a GPU with around 24 TFLOPs of half-precision compute performance.

Coming from AMD's internal server roadmap, the slide suggest that the Vega 10 GPU, which will use the new GFX9 architecture made on 14nm FinFET manufacturing process, will pack 64 Compute Units and a total of 24 TFLOPs of single-precision compute performance.

As detailed earlier, Vega 10 based graphics cards will come with 16GB of HBM2 memory with 512Gbps of memory bandwidth and a TDP of around 225W.

The same slide also suggest that while the single-GPU graphics card will come in Q1 2017, there will also be a dual-GPU version with two Vega 10 GPUs coming in the second quarter of 2017, with a TDP of around 300W.

The slide also reveals a bit more information regarding the Vega 20 GPU, a GPU based on the same GFX9 architecture but made on a 7nm manufacturing process. It will also have 64 Compute Units but pack 32GB of HBM2 memory with 1TB/s memory bandwidth and significantly lower TBP of around 150W, thanks to the new manufacturing process. It will also support PCI-Express 4.0 interface.

According to further details coming from Videocardz.com, AMD plans to replace the Polaris 10 GPU with Vega 11 GPU sometime next year and this one should be also based on 14nm FinFET manufacturing process. The new Navi 10 and Navi 11 GPUs are currently planned for 2019, which is a slight delay from the earlier seen roadmap.

Hopefully, we will hear more information around CES 2017 show which starts early next year.



Source: Videocardz.com.