Support for Nvidia 50 series

[Vertax42]

How to use Isaaclab on the Nvidia 50 series graphics card platform? I recently bought a 5080，download the latest nvidia driver 570.86.16 for ubuntu 20.04 and updated the isaaclab environment to torch 2.7.0, all training will be stuck before printing the model structure, when will isaaclab officially support blackwell architecture?

[rthaker01]

Hi @Vertax42 - Thank you for reaching out. Could you pls try out this line

pip install --upgrade --pre torch --index-url https://download.pytorch.org/whl/nightly/cu128

[Vertax42]

Thank you so much, it works on torch 2.7.0, just ignore the incompatible errors and it work fine!

[singh-jayant]

I have another issue. On my 5070ti, Every visual geometry rendered in the viewport appears VERY noisy. But the graphics look normal when I switch to Interactive Path Tracing.

[zhimou]

same here, and I can run any RL example without "--headless"

[renhong-zhang]

I have another issue. On my 5070ti, Every visual geometry rendered in the viewport appears VERY noisy. But the graphics look normal when I switch to Interactive Path Tracing.

Same here. Have you solved this problem?

[FrankYu09]

same here, and I can run any RL example without "--headless"

Met the same issue. I my case, this also means I cannot run distributed training on multiple gpus. Have you found a workaround?

[HugoXANA]

thanks for @Vertax42 raising the question and the anser provided by @rthaker01 . Im also using Isaac lab with RTX5080,and I followed the answer

Thank you so much, it works on torch 2.7.0, just ignore the incompatible errors and it work fine!

but found error when excuting

./isaaclab.sh -p scripts/demos/quadrupeds.py

the error was:

Traceback (most recent call last):
  File "/home/my_jammy/IsaacLab/source/isaaclab/isaaclab/assets/asset_base.py", line 94, in <lambda>
    lambda event, obj=weakref.proxy(self): obj._initialize_callback(event),
  File "/home/my_jammy/IsaacLab/source/isaaclab/isaaclab/assets/asset_base.py", line 265, in _initialize_callback
    self._initialize_impl()
  File "/home/my_jammy/IsaacLab/source/isaaclab/isaaclab/assets/articulation/articulation.py", line 1176, in _initialize_impl
    self._data = ArticulationData(self.root_physx_view, self.device)
  File "/home/my_jammy/IsaacLab/source/isaaclab/isaaclab/assets/articulation/articulation_data.py", line 56, in __init__
    gravity_dir = math_utils.normalize(gravity_dir.unsqueeze(0)).squeeze(0)
RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)

nvrtc compilation failed: 

#define NAN __int_as_float(0x7fffffff)
#define POS_INFINITY __int_as_float(0x7f800000)
#define NEG_INFINITY __int_as_float(0xff800000)


template<typename T>
__device__ T maximum(T a, T b) {
  return isnan(a) ? a : (a > b ? a : b);
}

template<typename T>
__device__ T minimum(T a, T b) {
  return isnan(a) ? a : (a < b ? a : b);
}

extern "C" __global__
void fused_clamp_unsqueeze_div(float* tx_1, float* tv_, double veps_1, float* aten_div) {
{
if ((long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)<3ll ? 1 : 0) {
    float tv__1 = __ldg(tv_ + 0ll);
    float v = __ldg(tx_1 + (long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x));
    aten_div[(long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)] = v / (tv__1<(float)(veps_1) ? (float)(veps_1) : tv__1);
  }}
}

and it repeated for several times,and it eventually caused the error:

Traceback (most recent call last):
  File "/home/my_jammy/IsaacLab/scripts/demos/quadrupeds.py", line 191, in <module>
    main()
  File "/home/my_jammy/IsaacLab/scripts/demos/quadrupeds.py", line 186, in main
    run_simulator(sim, scene_entities, scene_origins)
  File "/home/my_jammy/IsaacLab/scripts/demos/quadrupeds.py", line 143, in run_simulator
    root_state = robot.data.default_root_state.clone()
  File "/home/my_jammy/IsaacLab/source/isaaclab/isaaclab/assets/articulation/articulation.py", line 107, in data
    return self._data
AttributeError: 'Articulation' object has no attribute '_data'. Did you mean: 'data'?

I wonder if anyone has ever encountered with this problem and how to solve it? Thanks a lot.