XIO

xWire

Posts from every feed QUEST has found across the web — blogs, news, podcasts, channels and more.

NVIDIA Developer Forums - Latest posts

Latest posts

Assemble URDF models

Hi @remi.tortue – sorry this went unanswered for so long. The error you hit comes from a rename, and the API has changed since 4.5.0, so here is the current way to do this on Isaac Sim 6.2.0. Why the import failed. The extension omni.isaac.robot_assembler was renamed to isaacsim.robot_setup.assembler (see the 4.5 extension renaming table). On 6.2.0, enable_extension("omni.isaac.robot_assembler") does not enable anything and import omni.isaac.robot_assembler raises ModuleNotFoundError, which…

Jetson Sponsorship Request – Odonata: AI drone for bark beetle detection on Orin NX 16GB (Jugend forscht 2027, Germany)

Thanks for your interest in NVIDIA Jetson! Unfortunately, we’re unable to fulfill hardware sponsorship requests at this time — the volume of requests we receive makes it impossible to support them individually, and we do not have a grant program to direct you to at this time. We’d encourage you to explore purchasing Jetson through our authorized distributors, or check out the NVIDIA Jetson store for current options. We hope you find a path to get your project off the ground, and we’d love to…

Applying random material to objects

Hi @ph.silva1 – sorry this went unanswered for so long. I re-tested this on Isaac Sim 6.2.0 (Replicator omni.replicator.core 1.13.38). I could not reproduce “the material is applied to the Xform but not the meshes” with a simple Xform > Mesh asset, but I found two ways to end up with no visible material change that may match what you saw. I do not have your screw USD, so I cannot confirm which one applies. Randomizing the Xform works for a normal asset. rep.get.prims(semantics=...) returns the…

Is there any tutorials for construct new program using Isaac sim's python interpreter fro new learners?

Hi @2230580090 – sorry this went unanswered for so long. Here is how to set this up on Isaac Sim 6.2.0. 1. Your script can live in any folder. You do not need to put it inside the Isaac Sim install. Run it with the Isaac Sim Python wrapper and pass the path to your file: cd ~/my_project ~/isaacsim/isaacsim-6.2.0/python.sh hello.py # hello.py from isaacsim import SimulationApp simulation_app = SimulationApp({"headless": False}) # must be created before other Isaac Sim imports import…

NVIDIA Build API access unavailable after Developer Program welcome

Hello, I received the NVIDIA Developer Program welcome email, and my NVIDIA Developer profile shows that my program membership is connected. However, Try NVIDIA NIM APIs still shows “API Access Unavailable” and asks me to contact support to verify my account access. Could you please check whether my account needs manual verification and help enable Build API access? Thank you.

Importing .pcd file into Isaac Sim

Hi @sv6659 and @chenyog – an update based on Isaac Sim 6.2.0, since the earlier replies only linked the Pointclouds docs and the follow-up question was never answered. What Isaac Sim imports. The point cloud importer (omni.kit.pointclouds) loads .e57 and .pts files and turns them into a USD Points prim. There is no native .pcd importer, and opening a .pcd directly fails with “Unsupported file”. NVIDIA has no current plans to add other point cloud formats. Importing a .pcd (for example from…

60 tok/s GLM-5.3 Flash on a single DGX Spark

I gave it a try for the fun of it. I applied PR#4 Runtime knobs (ctx/util/seqs/async/APC/drafter block) + mamba prefix-hit seed fix (vllm#55600), README on the single-Spark memory budget by pondzikk · Pull Request #4 · gitcommit90/glm-5.3-one-spark · GitHub And some of the tweaks in Issue#4 ( One-Spark findings: 0.90 starves the host (Xid 31/hangs), prefix-cache hits crash KDA (upstream vLLM bug + 1-line fix), and how 262k/524k fits at 0.80 · Issue #3 · gitcommit90/glm-5.3-one-spark · GitHub )…

NVIDIA Nemotron Integration and Branding Guidance for an Independent Student Project

Hello NVIDIA Developer Community, My name is Mario Rosario, and I am an undergraduate Computer Science student at John Jay College of Criminal Justice, part of the City University of New York (CUNY). I am independently developing an experimental software project called IDOS Observatory, which explores AI-assisted conversations and structured decision-making using fictional scenarios. I am currently experimenting with NVIDIA Nemotron 3.5 Lightning through Vercel AI Gateway. NVIDIA Customer Care…

Advanced camera properties

Hi @justinpg and @1178065793 – closing out this older topic with an update, since this is now possible in Isaac Sim 6.2.0. Exposure time is now a per-camera setting. Isaac Sim 6.2.0 ships camera exposure attributes that the RTX renderer reads per camera, so you no longer need to change global tone-mapping settings: exposure:time, exposure:iso, exposure:fStop, exposure:responsivity (and the exposure EV offset) on the camera prim Auto-exposure on the camera via omni:rtx:autoExposure:enabled – set…

Request for DLSS 5 early access for Astra (Vulkan, ray-traced, multiplayer)

Hello, I’m the developer of Astra (https://astra.game), a multiplayer space game set on full-scale planets in a shared universe, and I would like to request early access to DLSS 5. Astra runs on a custom Vulkan renderer on Windows and Linux. It uses hardware ray tracing for global illumination and reflections, and it already integrates DLSS Super Resolution, Ray Reconstruction, Frame Generation and Reflex. Players regularly move from planetary surfaces to orbit and back without loading screens,…

DeepSeek-V4.1-Flash-1017 (EXL3 2.9bpw) on 2x DGX Spark with TensorFold: 57 tok/s text, 92 tok/s code, 2.1-2.3x the vLLM kit

On my Mac M5 max 64G, get 150tg, TEB 3.0.1 91 Checkpoint: TensorFold/Qwen3.8-27B-MLX-4bit — MLX 4-bit group-64, 14.98 GiB Engine: tensorfold-native 1.1.6 · metal · M5 Max , model id qwen3.8-27b , ctx 65,536 (maximum 180k) vision on, drafter z-lab/Qwen3.8-27B-DFlash2

Manual Account Verification Request – Kazakhstan (+7)

Hello NVIDIA Support Team, I am attempting to complete the registration process for NVIDIA NIM on build.nvidia.com to obtain an API key for testing and educational purposes. Unfortunately, the registration fails for my region because the country code for Kazakhstan (+7) is missing from the SMS verification dropdown list / not accepted by the system. Could you please manually verify my developer account? My details: Registered email: pushkareva1997@inbox.ru Country: Kazakhstan Phone number: +7…

Native DirectX 12 on Linux

Run it without your patches on vanilla 615.78.08 Several patches touch exactly this area: 0032 changes partial-unmap VA bookkeeping and the dmaFreeMapping error path. 0036 makes DMA map/unmap take the API lock for read instead of write, so these calls can run concurrently in new ways. 0033 and 0035 change mapping_reuse. 0041 and 0042 change uncached/WC sysmem, which is where eviction goes when VRAM is full.

Resulting labels of function nppiLabelMarkersUF_8u32u_C1R_Ctx seperate connected component?

For anyone who finds this thread: NPP 11.8 to 13.2 still split connected regions into several labels. NVIDIA has fixed this under bug 6963973, scheduled for CUDA 13.6. Reproducer and details: nppiLabelMarkersUF splits connected regions, NPP 11.8 to 13.2 and NPP+ 0.10 (3 × 3 repro; the cuda-samples outputs are affected too)

So for Dual Spark what is the Choice now: GLM5.3 Flash or Deepseek v4 Flash

Okay. This post is my rant about DeepSeek V4.1 1017 experience after I used it for couple weeks. Here is my update after just 2 weeks of struggling with DeepSeek V4.1 1017 (NVFP4). Is it fast? yes fast, can it code? Yes it can code good. Can it do some frontier work? Yes it can do some. But, due to the nature of work I am working on, I am returning back to GLM5.3-Flash-TP4 setup even though its much much slower (due to overthinking), at least it tries not to bull*it me. My experience with…

[615.78.08] "can't update VA space" when VRAM is full: user space driver reuses stale VAs

Related reports The same signature (0 == (pMapNode->gpuMask & gpuMask), can't update VA space, NV_ERR_INVALID_ARGUMENT) is reported in NVIDIA/open-gpu-kernel-modules#1364 with 610.57.04 and 595.84 on GA102, AD107 and TU104, for Deadlock with native Vulkan (no DXVK) and Borderlands 4 via vkd3d-proton, also with VRAM at about 96 %. So it is not specific to vkd3d-proton, this game or 615. This is not internal bug 5762513 (Bug Report: Wayland-only: dmaAllocMapping_GM107 VA mapping failures (does…

Blackwell Pro 6000 MHz degrading

This is really shitty, am i to go to my boss, hey that other very expensive gpu also broke, no problem we will get RMA but i also need an additional 4000 cash to buy a replacement that might break again in 6 months… I was thinking. The issue sounds really like a hardware issue. I have a 5090 as well, no issues, there have been many more 5090 sold, this is basically the same die, same board only more memory + same cooler. So than chip degradation seems unlikely because i have not really seen…

[615.78.08] "can't update VA space" when VRAM is full: user space driver reuses stale VAs

Follow-up to my earlier report in the 615 release feedback thread (615 release feedback & discussion - #231 by hurikhan77), now with a trace that shows when the problem starts and that it is not a race. System RTX 3080 Ti (GA102, 12 GB, VBIOS 94.02.71.40.4C), Gentoo, kernel 6.18.59, KDE Plasma on Wayland, open kernel modules with default module options. Marvel’s Guardians of the Galaxy (Steam 1088850), D3D12 via vkd3d-proton (Proton Experimental 11.0-20261027b, vkd3d-proton cae2e003b), no…

7.3 kernel and Nvidia eGPU

Anybody success with latest kernel and nvidia egpu? I really CANT make my aorus 5060 ti working with latest kernel via thunderbolt. FIXED by workaround. New kernel seems to be more agresive about rebar, and i need to add hpmemsize=32GB or hpmmioprefsize=32G as kernel flag to even get my nvidia egpu to work. Still nvidia-smi show 512mb rebar but even in tb4 AAA games works well. After all i am on 7.3.1 kernel and playing stalker 2, and seems this is not about nvidia driver but hard changes in…