CAN AMD RESPOND?
Intel isn’t just sticking an AMD graphics chip into the same package as its CPUs. Intel also now plans to compete directly in the high-performance GPU market with its own in-house designs. The endgame seems pretty clear—namely a unified high-performance solution that combines full CPU and graphics capabilities, and pushes all of Intel’s major competitors in the performance PC market to the sidelines.
But hang on: There’s another company that has both high-performance x86 CPU and graphics technology. That’s AMD, and in fact, it’s the only company that currently has both in-house. For Intel, high-performance in-house graphics is merely an aspiration at the moment. If Intel’s big new idea with its CPUGPU package is so clever, surely AMD can simply do its own version long before Intel can catch up on the graphics side?
On paper, that’s true. In practice, AMD struggles to sell its products at the high end, even when competitive. AMD could very well produce an outstanding CPU-GPU package and struggle to get OEM PC builders to adopt it. It just doesn’t have the marketing clout to fight off Intel’s aggressive tactics.
Oh, and if you are wondering, AMD’s new CPU-GPU chips, or APUs, codenamed Raven Ridge, are not competitors for the Intel package, with performance AMD graphics and HBM2 memory inside. Currently, Raven Ridge is available in two Ryzen APU models: the Ryzen 7 2700U and Ryzen 5 2500U, both aimed at laptop PCs. Yes, Raven Ridge does combine AMD’s fantastic new Ryzen CPU with its latest Vega graphics. In fact, it does so on the same slice of silicon, not just in the same package. But it’s that tight integration that reveals the real target for Raven Ridge: Intel’s existing laptop processors with integrated graphics. The top Ryzen 7 2700U version of Raven Ridge offers 10 AMD graphics compute units, which works out at 640 unified shaders. That compares with 1,536 shaders in the Intel-AMD product, albeit of a slightly different spec.
Raven Ridge is also a conventional APU in that the GPU part of the chip must share access to memory with the CPU via a standard memory controller. With current PC system memory performance, that pretty much makes it impossible to support high-resolution 3D graphics. CPU memory controllers and DDR4 RAM don’t offer nearly enough bandwidth for the GPU alone, let alone when sharing resources with the CPU.