PantheonSim / GPUs / NVIDIA GeForce RTX 3090

Simulated NVIDIA GeForce RTX 3090: a GPU profile for CUDA and ROCm testing

PantheonSim, the GPU simulator for CUDA and ROCm, simulates the NVIDIA GeForce RTX 3090 (Ampere, 8.6) as the profile nvidia/rtx3090. A 24 GB Ampere GeForce flagship. 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/rtx3090`
Vendor and architecture	NVIDIA, Ampere
Compute capability	`8.6`
Device memory	24.0 GiB, GDDR
Streaming multiprocessors	82
Warp size	32
Max threads per multiprocessor	1536
Shared memory per block (opt-in maximum)	99 KiB
L2 cache	6 MiB
Memory bus width	384-bit
Power limit (profile)	350 W
Maximum clocks (profile)	1695 MHz shader, 9750 MHz memory
ECC memory	no
PCIe link reported	Gen4 x16
Tensor or matrix cores	yes
Verified (measured on a physical card)	not yet measured

What this GPU is for

A 24 GB Ampere GeForce flagship.

Ampere covers two compute capabilities in these profiles: 8.0 for the GA100 data-center parts and 8.6 for the GA10x parts. The simulator runs the SASS a binary carries for the architecture and falls back to PTX when it carries none.

Among the 37 NVIDIA profiles it ranks 25th by device memory (24.0 GiB) and 24th by SMs (82), 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_86 my_app.cu -o my_app
vgpu run --gpu nvidia/rtx3090 ./my_app
vgpu shell -y --gpu nvidia/rtx3090 --count 4
vgpu info --gpu nvidia/rtx3090

This profile is compute capability 8.6. 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. Link the CUDA runtime shared (-cudart shared) so the simulator can stand in for it.

To use it every time, put gpu: nvidia/rtx3090 in a pantheonsim.yaml settings file, or in a GitHub Actions job give the PantheonSim action gpu: nvidia/rtx3090 (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 GeForce RTX 3090, as written in the file:

Source note for nvidia/rtx3090

Public values; verified: false, nothing here has been measured on a card. Where each value comes from: - compute capability 8.6: NVIDIA's CUDA GPUs table (developer.nvidia.com/cuda-gpus). - SMs 82: NVIDIA Ampere GA102 GPU Architecture whitepaper (v2), Appendix A, Table 9 (RTX 3090 FE: 82 SMs, 10496 CUDA Cores at 128 an SM, 328 Tensor Cores). - memory 24 GB GDDR6X, 19.5 Gbps, 384-bit, 936 GB/s: the same table, which gives the frame buffer as 24576 MB. NVIDIA's "GB" for graphics memory is binary (see rtx2080ti.yaml), so vram_bytes is 24576 MiB; a real card's nvidia-smi total is a little smaller, so this is an upper bound. - boost clock 1695 MHz (Founders Edition) and 350 W Total Graphics Power: the same table. - PCIe Gen4: the whitepaper's "PCIe Gen 4" section (a x16 slot, 64 GB/s); x16 is the slot the card is built for. - PCI device 0x2204, "NVIDIA GeForce RTX 3090", from the name table in NVIDIA's open GPU kernel modules (g_nv_name_released.h). - Per-SM limits are compute capability 8.6's, the same as nvidia/a10 and nvidia/rtx3080ti, which were read from physical cards. - CUDA attributes (the `cuda:` section and l2_cache_bytes): L2 6144 KB and the 384-bit bus: GA102 whitepaper; persisting L2 copied from nvidia/a10.

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

Other Ampere profiles: