The AMD Radeon VII will not have support for double precision floating point (FP64)

Robert Sole 15 January 2019

0 1.210 1 minute of reading

To segment the market, AMD has decided to remove the support for double precision floating point (FP64) in the new AMD Radeon VII.

The upcoming AMD Radeon VII graphics cards are generating a lot of buzz. Many rumors of various kinds are circulating about this card. The latest is that it apparently won't have double-precision floating-point (FP64). Sasa Marinkovic, AMD's Director of Product Marketing, confirmed to Techgage that FP64 will be disabled. This graphics card uses the Vega 20 silicon based on TSMC's 7nm lithography.

AMD Radeon VII without floating point double precision?

This new graphics has the same characteristics as the Radeon Instinct MI50. They both have 3840 Stream Processors, 240 TMUs, 64 ROPs, and 16GB of HBM2 memory. Radeon VII is directly intended for gaming, so it will have higher frequencies. To prevent the Instinct MI50 from suffering a drop in sales AMD has had the idea of eliminating FP64 in the gaming version.

Instinct MI50 can offer up to 6.7 TFLOPS on FP64. Suppressing this function in the gaming graphics, supposes a gigantic gap between both graphics. Even so, this graph is capable of offering 862 GFLOPS in FP64, well above the Titan RTX that offers 509 GFLOPS and the RTX 2080 Ti that gives 420 GFLOPS.

It wouldn't be the only feature disabled on the Radeon VII. It seems that AMD modified the PCIe interface to use PCIe 3.0 instead of PCIe 4.0 like the Instinct MI50 does. It seems that PCIe 4.0 is used in the MI50 only for artificial intelligence, which needs a lot of bandwidth. This will be the other big difference with the professional model.

It should be noted that this will not affect video game fans at all. It is a restriction on the part of AMD not to lose sales of the Instinct MI50, which is much more expensive.