AMD Zen 2 Microarchitecture Analysis: Ryzen 3000 and EPYC Rome

Floating Point

The key highlight improvement for floating point performance is full AVX2 support. AMD has increased the execution unit width from 128-bit to 256-bit, allowing for single-cycle AVX2 calculations, rather than cracking the calculation into two instructions and two cycles. This is enhanced by giving 256-bit loads and stores, so the FMA units can be continuously fed. AMD states that due to its energy aware scheduling, there is no predefined frequency drop when using AVX2 instructions (however frequency may be reduced dependent on temperature and voltage requirements, but that’s automatic regardless of instructions used)

In the floating point unit, the queues accept up to four micro-ops per cycle from the dispatch unit which feed into a 160-entry physical register file. This moves into four execution units, which can be fed with 256b data in the load and store mechanism.

Other tweaks have been made to the FMA units than beyond doubling the size – AMD states that they have increased raw performance in memory allocations, for repetitive physics calculations, and certain audio processing techniques.

Another key update is decreasing the FP multiplication latency from 4 cycles to 3 cycles. That is quite a significant improvement. AMD has stated that it is keeping a lot of the detail under wraps, as it wants to present it at Hot Chips is August. We’ll be running a full instruction analysis for our reviews on July 7^th.