RTX 5060 Ti Benchmark — Pi0.5 (SM120/Blackwell)
I'm submitting Pi0.5 benchmark results for the NVIDIA GeForce RTX 5060 Ti (Blackwell SM120).This submission includes the quickstart benchmark metrics, full system environment info, and raw nvidia-smi output. I've also added supplementary LIBERO spatial benchmark results as well as performance for your reference.
Quickstart Benchmark
Command:
python examples/quickstart.py \
--checkpoint /path/to/pi05libero \
--config pi05 \
--framework torch \
--hardware rtx_sm120 \
--benchmark 20 \
--warmup 200Results (Pi0.5, 2 views, FP8, wall-clock model.predict()):
| Metric |
Value |
| P50 |
41.4 ms |
| Mean |
41.4 ms |
| Min |
40.9 ms |
| Max |
43.2 ms |
| Throughput |
~24 Hz |
System Info
| Component |
Version |
| GPU |
NVIDIA GeForce RTX 5060 Ti |
| Compute Capability |
SM 12.0 (Blackwell) |
| VRAM |
16311 MiB (16 GB) |
| NVIDIA Driver |
591.86 |
| CUDA (nvidia-smi) |
13.1 |
| PyTorch |
2.11.0+cu128 |
| torch CUDA |
12.8 |
| Python |
3.12.3 |
| GCC |
13.3.0 |
nvidia-smi output:
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 590.57 Driver Version: 591.86 CUDA Version: 13.1 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5060 Ti On | 00000000:01:00.0 On | N/A |
| 0% 44C P3 23W / 180W | 12263MiB / 16311MiB | 27% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
LIBERO Spatial Benchmark (Bonus)
Ran examples/thor/eval_libero.py --task_suite libero_spatial --num_trials 50:
| Task |
Description |
Success Rate |
Avg Episode Time |
| 0 |
bowl between plate and ramekin → plate |
50/50 (100%) |
11.2s |
| 3 |
bowl on cookie box → plate |
49/50 (98%) |
11.5s |
| 4 |
bowl in top drawer of cabinet → plate |
49/50 (98%) |
16.6s |
| 5 |
bowl on ramekin → plate |
47/50 (94%) |
13.4s |
| 7 |
bowl on stove → plate |
50/50 (100%) |
16.2s |
| 8 |
bowl next to plate → plate |
50/50 (100%) |
12.7s |
| 9 |
bowl on wooden cabinet → plate |
49/50 (98%) |
16.1s |
| Overall |
(7 tasks) |
344/350 (98.3%) |
|
Performance
| GPU |
Pi0.5 Latency (2v, FP8) |
Throughput |
| RTX 5060 Ti (SM120, 16GB) |
41.4 ms |
24 Hz |
RTX 5060 Ti Benchmark — Pi0.5 (SM120/Blackwell)
I'm submitting Pi0.5 benchmark results for the NVIDIA GeForce RTX 5060 Ti (Blackwell SM120).This submission includes the quickstart benchmark metrics, full system environment info, and raw nvidia-smi output. I've also added supplementary LIBERO spatial benchmark results as well as performance for your reference.
Quickstart Benchmark
Command:
python examples/quickstart.py \ --checkpoint /path/to/pi05libero \ --config pi05 \ --framework torch \ --hardware rtx_sm120 \ --benchmark 20 \ --warmup 200Results (Pi0.5, 2 views, FP8, wall-clock
model.predict()):System Info
nvidia-smi output:
LIBERO Spatial Benchmark (Bonus)
Ran
examples/thor/eval_libero.py --task_suite libero_spatial --num_trials 50:Performance