The GPU inside the VII is called Vega 20, which is a die-shrunk version of the Vega 10 in the Vega 64.
The new chip has 128 ROPs to the old chip’s 64, doubling the number of rendered rasterized pixels it can produce. However, it does fall behind its predecessor in one spec: it has only 60 compute units (3,840 stream processors) compared to 64 (4,096 stream processors).
With the clock-speed boost more than making up for the loss of CUs, the new chip delivers single-precision floating-point performance of 13.8 teraFLOPS compared to 12.7 from its predecessor. AMD is claiming that it more or less matches the RTX 2080 in a number of games, with , , and all called out as titles where the VII ties or beats the RTX 2080 at 4K with maximum settings.
However, the VII doesn’t have the RTX 2080’s raytracing support and is likely more power-hungry, as it’s expected to draw 295-300W. Still, it means that the card is good enough for single GPU 4K gaming.
This isn’t the first Vega 20 product to hit the market. The Radeon Instinct MI50 and MI60 compute cards already use Vega 20. Those cards, which don’t have any display outputs, are aimed at data centers as well as for machine learning, rendering, and other compute-intensive tasks. The MI50 has the same 16GB/60 CU configuration and runs at close to the same 1,800MHz, so it looks as if the Radeon VII is a consumerized version of the MI50.
The new card will become available on February 7 and will cost $699. AMD will be selling a version of the card directly, and there should be third-party versions coming, too. AMD’s version will ship with , , and .