PantheonSim / GPUs / NVIDIA GeForce RTX 2080 Ti

Simulated NVIDIA GeForce RTX 2080 Ti: a GPU profile for CUDA and ROCm testing

PantheonSim, the GPU simulator for CUDA and ROCm, simulates the NVIDIA GeForce RTX 2080 Ti (Turing, 7.5) as the profile nvidia/rtx2080ti. An 11 GB high-end Turing GeForce card. 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/rtx2080ti`
Vendor and architecture	NVIDIA, Turing
Compute capability	`7.5`
Device memory	11.0 GiB, GDDR
Streaming multiprocessors	68
Warp size	32
Max threads per multiprocessor	1024
Shared memory per block (opt-in maximum)	64 KiB
L2 cache	5.5 MiB
Memory bus width	352-bit
Power limit (profile)	250 W
Maximum clocks (profile)	1545 MHz shader, 7000 MHz memory
ECC memory	no
PCIe link reported	Gen3 x16
Tensor or matrix cores	yes
Verified (measured on a physical card)	not yet measured

What this GPU is for

An 11 GB high-end Turing GeForce card.

Turing is compute capability 7.5, the oldest architecture among the NVIDIA profiles. A kernel built for compute_75 runs on every NVIDIA profile here, which is why it is the usual target for a test that has to run on all of them.

Among the 37 NVIDIA profiles it ranks 36th by device memory (11.0 GiB) and 29th by SMs (68), 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_75 my_app.cu -o my_app
vgpu run --gpu nvidia/rtx2080ti ./my_app
vgpu shell -y --gpu nvidia/rtx2080ti --count 4
vgpu info --gpu nvidia/rtx2080ti

This profile is compute capability 7.5. 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/rtx2080ti in a pantheonsim.yaml settings file, or in a GitHub Actions job give the PantheonSim action gpu: nvidia/rtx2080ti (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 2080 Ti, as written in the file:

Source note for nvidia/rtx2080ti

Public values; verified: false, nothing here has been measured on a card. Where each value comes from: - compute capability 7.5: NVIDIA's CUDA GPUs table (developer.nvidia.com/cuda-gpus). - SMs 68: NVIDIA Turing GPU Architecture whitepaper (WP-09183-001), Table 1 (68 SMs, 4352 CUDA cores at 64 an SM, 544 Tensor Cores); the Ada whitepaper's Table 1 repeats it. - memory 11 GB GDDR6, 14 Gbps, 352-bit, 616 GB/s: the same Turing table, which gives the frame buffer as 11264 MB. NVIDIA's "GB" for graphics memory is binary (the whitepapers' MB figures are 1024 x the GB), so vram_bytes is 11264 MiB; a real card's nvidia-smi total is a little smaller, so this is an upper bound. - boost clock 1545 MHz and board power 250 W: the Turing whitepaper's reference-card figures (the Founders Edition is 1635 MHz and 260 W). - PCIe Gen3: the Ada whitepaper's Table 2 "PCI Express Interface" row; x16 is the slot the card is built for, not a figure NVIDIA states there. - PCI device 0x1E04, "NVIDIA GeForce RTX 2080 Ti", from the name table in NVIDIA's open GPU kernel modules (g_nv_name_released.h). The table also lists 0x1E07 for the same name (a later revision); 0x1E04 is the first. - Per-SM limits are compute capability 7.5's: the same as nvidia/t4, which was read from a physical card (registers 64K, 1024 threads, 16 blocks, 64 KB shared memory), and as the CUDA C++ Programming Guide's table gives. - CUDA attributes (the `cuda:` section and l2_cache_bytes): L2 5632 KB and the 352-bit bus: Turing whitepaper Table 1; no persisting L2 before 8.0 (as nvidia/t4).

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

Other Turing profiles:

Tesla T4