PantheonSim / GPUs / NVIDIA GB200

Simulated NVIDIA GB200: a GPU profile for CUDA and ROCm testing

PantheonSim, the GPU simulator for CUDA and ROCm, simulates the NVIDIA GB200 (Blackwell, 10.0) as the profile nvidia/gb200. One Blackwell GPU of the Grace Blackwell Superchip, which pairs two Blackwell GPUs with a Grace CPU. Its profile is built from public documents and not yet measured against a physical card.

The profile at a glance

These are the numbers a program can ask this simulated GPU, and the values nvidia-smi show for it. They come from the profile file in the simulator's repository, pinned to one commit of it.

Property	Value
Profile id	`nvidia/gb200`
Vendor and architecture	NVIDIA, Blackwell
Compute capability	`10.0`
Device memory	186.0 GiB, HBM
Streaming multiprocessors	148
Warp size	32
Max threads per multiprocessor	2048
Shared memory per block (opt-in maximum)	227 KiB
Memory bus width	8192-bit
Power limit (profile)	1000 W
Maximum clocks (profile)	1965 MHz shader, 4000 MHz memory
ECC memory	yes
NVLink reported	NVLink5 x18
Tensor or matrix cores	yes
Verified (measured on a physical card)	not yet measured

What this GPU is for

One Blackwell GPU of the Grace Blackwell Superchip, which pairs two Blackwell GPUs with a Grace CPU.

Blackwell is split across several compute capabilities in these profiles: 10.0 for the B200 and GB200, 10.3 for the B300, 11.0 for Jetson Thor, 12.0 for the GeForce RTX 50 series and the RTX PRO 6000 cards, and 12.1 for the GB10. Only the RTX PRO 6000 Server Edition was read from a physical card.

Among the 37 NVIDIA profiles it ranks 3rd by device memory (186.0 GiB) and 6th by SMs (148), counting the largest as first.

Is this profile verified?

No. This profile is marked not verified: it is built from public documents only, and nothing in it was measured on a physical card. Some figures are derived or copied from the nearest profile where the vendor publishes none, so treat them as approximate, and memory sizes as upper bounds. Programs still run on it, and it still answers every call as the architecture's rules require, but a value such as a clock or a limit may differ from a real card's.

Simulate this GPU

Build the simulator (see the docs) or open the playground, then:

nvcc -cudart shared -arch=compute_100 my_app.cu -o my_app
vgpu run --gpu nvidia/gb200 ./my_app
vgpu shell -y --gpu nvidia/gb200 --count 4
vgpu info --gpu nvidia/gb200

This profile is compute capability 10.0. A kernel built for compute_75 runs on it and on every other NVIDIA profile; a kernel built for a newer architecture is refused by a profile that is older than it. Blackwell targets need CUDA 12.8 or newer. Link the CUDA runtime shared (-cudart shared) so the simulator can stand in for it.

To use it every time, put gpu: nvidia/gb200 in a pantheonsim.yaml settings file, or in a GitHub Actions job give the PantheonSim action gpu: nvidia/gb200 (see CUDA and HIP tests in GitHub Actions). The simulator checks behaviour and has no timing model, so a run on a physical GPU before a release still counts. The CUDA guide explains the rest.

Where the numbers come from

Each profile file begins with a note on its sources. This is the note for NVIDIA GB200, as written in the file:

Source note for nvidia/gb200

Blackwell Superchip, sm_100). Public values; verified: false, nothing here has been measured on a card. This differs from nvidia/b200 where NVIDIA publishes a per-GPU figure that does: - memory: the GB200 NVL72 page gives the superchip (two Blackwell GPUs) "372 GB HBM3E" and 16 TB/s, so one GPU has 186 GB and 8 TB/s (derived, halving the superchip's figures). NVIDIA's "GB" for HBM is binary (a B200's 180 GB reports 183359 MiB), so vram_bytes is 190464 MiB; an upper bound. - PCI device 0x2941, "NVIDIA GB200", from the name table in NVIDIA's open GPU kernel modules (g_nv_name_released.h); the B200 is 0x2901. - NVLink: 3.6 TB/s for the superchip's two GPUs, 1.8 TB/s each, which is NVLink 5's 18 links (the B200 profile's count). Everything else is COPIED from nvidia/b200 because NVIDIA's page gives no separate figure: compute capability 10.0 (CUDA GPUs table lists GB200 as 10.0), the SM count (148, itself unverified there), per-SM limits, clocks, and the 1000 W power limit (the page lists no TDP), the 90 C limit and the memory sensor. The Grace link is NVLink-C2C, so the PCIe link is left out rather than taken from the B200. - CUDA attributes (the `cuda:` section): the bus width is the B200's; the L2 is unpublished and left unset.

The GPU index lists every profile, and this one is not yet measured.

Other Blackwell profiles: