50 series GPUs nvrtc compilation failed problems in none-headless mode

[Vertax42]

When I use isaac lab to verify the reinforcement learning model, I got the following error after running the play script without adding the headless command:
Traceback (most recent call last):
File "/home/vertax/IsaacLab/scripts/reinforcement_learning/rsl_rl/play.py", line 159, in
main()
File "/home/vertax/IsaacLab/scripts/reinforcement_learning/rsl_rl/play.py", line 141, in main
obs, _, _, _ = env.step(actions)
File "/home/vertax/IsaacLab/source/isaaclab_rl/isaaclab_rl/rsl_rl/vecenv_wrapper.py", line 164, in step
obs_dict, rew, terminated, truncated, extras = self.env.step(actions)
File "/home/vertax/Downloads/[email protected]/exts/omni.isaac.ml_archive/pip_prebundle/gymnasium/wrappers/order_enforcing.py", line 56, in step
return self.env.step(action)
File "/home/vertax/IsaacLab/source/isaaclab/isaaclab/envs/manager_based_rl_env.py", line 206, in step
self.reward_buf = self.reward_manager.compute(dt=self.step_dt)
File "/home/vertax/IsaacLab/source/isaaclab/isaaclab/managers/reward_manager.py", line 148, in compute
value = term_cfg.func(self._env, **term_cfg.params) * term_cfg.weight * dt
File "/home/vertax/IsaacLab/source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/reach/mdp/rewards.py", line 49, in position_command_error_tanh
des_pos_w, _ = combine_frame_transforms(asset.data.root_state_w[:, :3], asset.data.root_state_w[:, 3:7], des_pos_b)
**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
device T maximum(T a, T b) {
return isnan(a) ? a : (a > b ? a : b);
}

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

extern "C" global
void fused_mul_add_add(float* tv_, float* tvec_7, float* tv__, float* tt_1, float* aten_add_1) {
{
if ((long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)<96ll ? 1 : 0) {
float v = __ldg(tvec_7 + ((long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)) % 3ll + 7ll * (((long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)) / 3ll));
float v_1 = ldg(tv + 13ll * (((long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)) / 3ll));
float v_2 = __ldg(tt_1 + (long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x));
float v_3 = _ldg(tv + (long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x));
aten_add_1[(long long)(threadIdx.x) + 512ll * (long long)(blockIdx.x)] = (v + v_1 * v_2) + v_3;
}}
}
I guess it's because the new architecture of the 50 series graphics card cannot be correctly recognized. How can I solve this problem?

[kellyguo11]

For 50 series GPU, please update to the latest nightly PyTorch build as mentioned in the docs - https://isaac-sim.github.io/IsaacLab/main/source/setup/installation/pip_installation.html#installation

[Vertax42]

Hi, I had already update my torch to 2.7.0+cu128 20250308dev so the training process can be down, now the error is occurring when I run play.py without adding —headless

[kellyguo11]

ah I see, we will probably have to try to repro this on our end, we haven't done extensive testing on the 50 series GPUs yet.

[moribots]

I am also having the same issue on a 5080, and have tried the same resolution steps

[SunnyChen-nx]

I also meet this problem. Has this issue been resolved?