AMD Announces New Vega High Performance Discrete GPU For Mobility With HBM2 Memory
AMD's CES announcements also include the Vega discrete GPU for mobility. This one is particularly special because just like its desktop counterpart, it will retain HBM2 memory and is probably the first GPU in the mobility space to do so. The timeline isn't clear at this point but we expect to see it with OEMs soon enough. The low power design of the chip should make it ideal for the thermally constrained environments of mobility.
AMD announces new Vega mobile Radeon graphics with HBM2 memory for ultra-thin platforms
The Vega mobility GPU lineup is a direct response to NVIDIA's existing discrete GPUs for mobility lineup. Not a lot of details were revealed at this point but we know for a fact now that AMD will be striking hard in this space. Coupled with Zen processors these could go on to become insanely popular with OEMs and AMD-only laptops. The AMD dGPUs for mobility have historically tended to be cheaper than their NVIDIA counterparts and this has led to wide OEM adoption.
The use of HBM2 will give the product quite an edge since it will allow it to tap into large amounts of bandwidth, something GDDR5 cannot do on this platform because of thermal constraints and overclocking limitations. The low power requirements of the memory should also work in AMD's advantage since that make it an ideal candidate for ultra thin platforms.
All in all, Vega mobile GPU should be a significant rainmaker for the company, because where it might loose out to GeForce high end products in raw performance numbers, it can easily make up for it by volume in OEM spaces. The Z-height metric is also important here because it would allow the GPU to compete in ultrabook variants where the MX150 from NVIDIA is usually the go-to choice for manufacturers that want a powerful dGPU without it overpowering the delicate thermal envelop.
In the marketing material, the company is also targeting the VR platform which requires a minimum horsepower of 3.5 TFLOPs. Because of this, we can safely deduce the minimum performance level of the chip, which would make it roughly 1.75x faster than the MX150 if true. NVIDIA would almost certainly release a competitor (another Max-Q design?) but the one thing they will not be able to match is the advantage of the HBM2 memory which will always leave more TDP room for AMD OEMs to play with.
Keep in mind we have already seen the Vega architecture in AMD mobility APUs before, but those use the standard DDR memory. This will be the first instance where HBM2 is used in such a thing format and will undoubtedly lead to some great results. HBM2 has been an expensive thing to source so far, but with SKHynix entering volume production last year, it looks like cost has finally come down to a point where AMD can offer HBM2 even on its mobility side of things.