PantheonSim / GPUs / NVIDIA Jetson AGX Thor

Simulated NVIDIA Jetson AGX Thor: a GPU profile for CUDA and ROCm testing

PantheonSim, the GPU simulator for CUDA and ROCm, simulates the NVIDIA Jetson AGX Thor (Blackwell, 11.0) as the profile nvidia/thor. The Blackwell GPU in the Jetson AGX Thor module. It is part of a system on a chip and shares 128 GiB of memory with the CPU. Its PCI id in the profile is a placeholder. 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/thor`
Vendor and architecture	NVIDIA, Blackwell
Compute capability	`11.0`
Device memory	128.0 GiB
Streaming multiprocessors	20
Warp size	32
Max threads per multiprocessor	2048
Shared memory per block (opt-in maximum)	227 KiB
Memory bus width	256-bit
Power limit (profile)	130 W
Maximum clocks (profile)	1570 MHz shader, 4266 MHz memory
Tensor or matrix cores	yes
Verified (measured on a physical card)	not yet measured

What this GPU is for

The Blackwell GPU in the Jetson AGX Thor module. It is part of a system on a chip and shares 128 GiB of memory with the CPU. Its PCI id in the profile is a placeholder.

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 7th by device memory (128.0 GiB) and 37th by SMs (20), 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. The profile file itself calls some of its values placeholders. 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_75 my_app.cu -o my_app
vgpu run --gpu nvidia/thor ./my_app
vgpu shell -y --gpu nvidia/thor --count 4
vgpu info --gpu nvidia/thor

This profile is compute capability 11.0. 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/thor in a pantheonsim.yaml settings file, or in a GitHub Actions job give the PantheonSim action gpu: nvidia/thor (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 Jetson AGX Thor, as written in the file:

Source note for nvidia/thor

compute capability 11.0, sm_110). Public values; verified: false, nothing here has been measured on a card. The PCI device id below (0x7F11) is a PLACEHOLDER, not NVIDIA's. NVIDIA has not published an id for Thor's GPU: NVIDIA's open GPU kernel modules name table (g_nv_name_released.h) has no Thor or Tegra entry. 0x7F11 is in no entry of that table, so it collides with no NVIDIA part, and is below 0x8000 because CUDA reports (device << 16 | vendor) as a signed int. Replace it when NVIDIA publishes one, as the MI300X and MI350X placeholders are. Rechecked 2026-10-09 against main (ca2d03e, 2026-10-07): still no Thor or Tegra entry. Where each value comes from: - compute capability 11.0 (sm_110): the CUDA C++ Programming Guide's table has an 11.0 column, and the CUDA toolkit release notes (cuBLAS 13.0) add "sm_110 GPUs for arm64-sbsa" and "TF32 GEMMs on Thor GPUs". NVIDIA's CUDA GPUs table lists the Jetson T5000 and T4000 under 11.0 but has no Thor row. - CUDA cores 2560, 10 TPCs, GPU max frequency 1.57 GHz, 128 GB 256-bit LPDDR5X at 273 GB/s, 40 W to 130 W configurable: NVIDIA's Jetson Thor page (T5000). SMs 20 are DERIVED: 2560 / 128 CUDA cores an SM (NVIDIA's RTX Blackwell whitepaper), which is also 10 TPCs at two SMs each; NVIDIA states no SM count. Unverified. - Memory: vram_bytes is 128 GiB, the whole unified pool shared with the CPU, an upper bound (NVIDIA's "GB" is binary here as elsewhere). The memory clock is what 273 GB/s on 256 bits gives, 8.53 Gbps, half of it 4266 MHz, the figure NVIDIA's DGX Spark page gives for the same LPDDR5X. - Per-SM shared memory: the programming guide's 11.0 column, 228 KB an SM and 227 KB a block (sm_100's; the block figure is inferred, not read from the guide), 64K 32-bit registers an SM. - COPIED from nvidia/b200 (sm_100) because the guide's resident-block and thread rows could not be tied to the 11.0 column: 2048 threads and 32 blocks an SM. power_limit_w is the top of the configurable range, 130 W. - Not recorded: ECC, a memory type (LPDDR5X is neither "hbm" nor "gddr"), PCIe and NVLink links (Thor is an SoC), a temperature limit. - CUDA attributes (the `cuda:` section): the 256-bit bus is NVIDIA's; the L2 is unpublished and left unset.

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

Other Blackwell profiles: