NVIDIA Grace (ARM Neoverse V2) CPU
The SLeeLa NVIDIA Grace (ARM Neoverse V2) profile models the processor as a 64-bit server CPU (ARM Neoverse) with a RISC general-purpose register file and an architectural, timing-oriented view of its execution.
Identity
| Field | Value | Evidence |
|---|---|---|
| Family | NVIDIA Grace (ARM Neoverse V2) | documented |
| Data width | 64-bit | documented |
| External address width | 48-bit | documented |
| Address space | 256 TiB | derived from 48-bit addressing |
| Programmer registers | AArch64 register model; many Neoverse-V2 cores with a coherent mesh and large shared cache | documented |
| Rated clock (model) | about 3400 MHz | documented/representative |
| Cache (model) | L1 + L2 + L3 | documented/architecture model |
| Integrated DMA | No | documented/architecture model |
| MMU/TLB | Yes | documented/architecture model |
The profile is a functional model: enough register, memory and timing structure to run the CPU and host SLeeLa/C/C++ workloads, not a cycle-accurate silicon recreation.
SLeeLa modeling
The profile composes the shared CPU objects:
- SLNvidiaGraceCPU (extends SLCPURuntime)
- SLClock
- SLRegisterFile
- SLALU
- SLInstructionDecoder
- SLIOBus
- SLIOTiming
- SLInterruptController
Cache, DMA and MMU objects are instantiated only where this part actually integrates them (see CACHE.md, DMA.md).
Variant rule
NVIDIA Grace datacenter CPU built on ARM Neoverse V2 cores. Do not silently merge variant-specific I/O, cache, addressing or opcode behavior into this base model.
See SPECIFICATION.md for the detailed model.
Sources
- https://www.nvidia.com/en-us/data-center/grace-cpu/
- https://en.wikipedia.org/wiki/ARM_Neoverse