The Open-Source compilers GCC and LLVM landed the support of the nvidia Olympus nuclei for the Vera processor of Nvidia which is part of their new generation Rubin microarchitecture which succeeds Blackwell.
Last week, at the NVIDIA GTC conference when he spoke of Vera, they noted that they used “personalized” Olympus nuclei twice as fast as the Grace processor using Neoversse-V2 nuclei. With Vera, there will be 88 of these personalized arm nuclei named Olympus.
Now, with the patches for LLVM / Clang and GCC, we have a little more knowledge of Nvidia Olympus. THE GCC patch is merged for the next stable version of the GCC 15.1 and there is the LLVM patch For the LLVM / Clang 21 release 21 later in the year. The two patches carry a similar message:
“This correction adds the support of the Nvidia Olympus nucleus.
This does not add any special adjustment decision, and these can come later. “”
We therefore lack special insight, but the compiler fixes note that Olympus with nvidia nuclei “n” uses the armv9.2-a Isa and is similar to neoversse-v3.
Additional ISA Arm capacities in addition to ARMV9.2-A include SVE2_Bitperm, RNG, LS64, Memtag, Profile, Faminmax, FP8DOT2, LUT, SVE2_Aes, SVE2_SHA3, SVE2_SM4.
With the LLVM patch for the moment at least its use of the NEOVERSSE-V2 planning model. Olympus uses two separate CPU room numbers from 0x10 and 0x010.
These compiler fixes confirm the ARMV9.2-A Isa processors for NVIDIA Vera and some of the sustained extensions, but not much more information, with apparently very similar to the NEVERSSE-V3 which is also based on ARMV9.2-A without these fixes to shed light on these “personalized” Olympus cores.