HI, Thus, Thanks
Hi, Do you put the device in recovery mode before running the steps? Please use lsusb and refer to below developer guide to check. docs.nvidia.com lsusb - Quick Start — NVIDIA Jetson Linux Developer Guide 1 documentation This topic will help you get started quickly using NVIDIA® Jetson™ Linux with an NVIDIA Jetson developer kit. Thanks
Thanks for your reply. Then if I use torch.mm api to test FLOPS, how can I specify the acc data type?
We are hosting a school management web application on a Jetson AGX Thor Developer Kit as an on-premise appliance for a single campus. It is a conventional three-tier server workload, not an inference workload, which is why we are unsure how much of the usual Jetson tuning guidance applies to us. API: ASP.NET Core 8 (linux-arm64) in Docker ENVIRONMENT Device : Jetson AGX Thor Developer Kit (T5000)
Yes, it’s true even though both RTX Pro 6000 and RTX 5090 in fact use the same chip (GB202).
Hi, Do you put the device in recovery mode before running sdk manager? Please use the lsusb to check whether log show. docs.nvidia.com lsusb - Quick Start — NVIDIA Jetson Linux Developer Guide 1 documentation This topic will help you get started quickly using NVIDIA® Jetson™ Linux with an NVIDIA Jetson developer kit. Thanks
Hi, Please provide the serial console log: RidgeRun Developer Wiki – 29 Sep 25 NVIDIA Jetson AGX Thor - Accessing the Board via Serial Console Explore in-depth developer resources, setup guides, and technical documentation for NVIDIA Jetson AGX Thor platforms, optimized by RidgeRun for advanced AI and robotics, particularly humanoid robots. Thanks
Already saw, that repo author updated its decision this night with your patch: Activity · taoofshawn/spark-recipes · GitHub Checking it now. Thank you for this great patch!
Hi, Please try to use below commands $ sudo ./flash.sh jetson-agx-orin-devkit internal Thanks
Nanodeoclus: This is not the case on workstation (RTX Pro) Is that true even if they have the same CC? For example, both RTX Pro 6000 Blackwell and RTX 5090 are CC 12.
My recipe fixes exactly this for the Intel AutoRound quant and my tool eval bench scores are way better, but you might not have used the z.ai recommended temperature for agentic work (most people here use dsv4f recommended temp, not sure why): GLM 5.3 Flash Intel AutoQuant W4A16 TP2 MTP3 - Concurrent Agentic Use DGX Spark / GB10 Projects awesome thank you! going to test this out
Hi, Please refer to the related topics JON Jetpack 5.1.5 SDK Manager Flash Failure -- HELP PLS Jetson Orin Nano doc on your normal device to get the board config. And use below commands under the ~/nvidia/nvidia_sdk/JetPack_5.1.5_Linux $sudo SKIP_EEPROM_CHECK=1 BOARDID= BOARDSKU= ./tools/kernel_flash/l4t_initrd_flash.sh --external-device nvme0n1p1 \ -c tools/kernel_flash/fla… Thanks
Hi, Do you have any error log when run the commands as developer guide suggest? Thanks
I’ve been trying to use it since an hour ago and it doesn’t answer. I got some ECONNRESET errors, then one successful response and now 429.
Hi, Sorry, we don’t support this. Thanks
Do you have IMX477, please? try this: gst-launch-1.0 nvarguscamerasrc sensor-id=0 aeregion="2000 1100 2356 1456 1.0" ! 'video/x-raw(memory:NVMM), width=3840, height=2160, framerate=21/1, format=(string)NV12' ! fakesink Argus log also attached. left 2000 - crash left 2032 - crash left 2064 - ok argus.log (7.2 KB)
Some other factors you have to be careful about, when interpreting FLOPS numbers: clock: base vs. boost clock unit: FFMA operations or floating-point operations. 1 FFMA instruction or operation counts as 2 FLOP / floating-point operations (multiply and add) matrix-format: sparse vs. dense calculated on which hardware compoinent: conventional arithmetic units vs. TensorCores vs. both together
Believe me, I know exactly what you mean. I used to get very angry with Claude on a daily basis. I had my fair share of ranty swearing sessions at various Qwen models. I don’t tend to get angry at ds4flash at all as it mostly just works - follows my workflows, tells me if it can’t do something, and I don’t see the hallucination that a lot of people say is there with the model. Perhaps that’s the w
Hi, Thank you very much for the explanation. Do you happen to know when IGX-SW 2.1 (r39.x based) is expected to be released? Even a rough timeframe would be helpful.
Platform: Jetson T5000 (p3834-0008) on a custom carrier, TE1070M silicon, L4T 38.4.0 and L4T 39.2.1 178d66cb448f0a0c8d80-b8f4737c39f). What we are trying to do Our carrier routes PCIe-C5 x4 (UPHY lanes 8-11) to an AMD Versal Gen 2 FPGA qdma_x4 endpoint. C5 is the only PCIe path to that FPGA, and x4 is a production — we cannot trade width for a workaround. We do not use MGBE at : what we want is a
Hello NVIDIA team, I registered an NVIDIA account but cannot generate an API key because the Account email: [email protected] Phone number: +86 17388245785 Goal: generate API key Screenshot attached. Could you please manually verify my account so I can
Hi, I reran the test using the original nvme_read_write.c attached in the NVIDIA forum thread. The source code was compiled and executed without any modifications. Platform: NVIDIA Jetson Orin NX 8GB Jetson Linux R36.5.2 Kernel 5.15.199-tegra SSD: Samsung 9100 PRO 1TB PCIe link status: Gen4, 16 GT/s, x4 Test preparation: sudo dd if=/dev/urandom of=/mnt/testfile1 bs=64M count=960 Test command: sudo
Здравствуйте! build.nvidia.com, но не могу проверить номер телефона, так как Россия (+7) Email: [[email protected]]
$ uuidgen > bootloader/l4t-rootfs-uuid.txt_ext dcaaf728-800a-41ec-b672-db1902fff835 安装好系统后，进系统执行： $ cat /boot/extlinux/extlinux.conf | grep UUID dcaaf728-800a-41ec-b672-db1902fff835 rw rootwait rootfstype $ sudo blkid | grep APP dcaaf728-800a-41ec-b672-db1902fff835” 看上去是好起来了，对吗？
Hello NVIDIA Developer Community, I’ve been developing an prototype of a semiconductor manufacturing digital twin using OpenUSD and NVIDIA Omniverse technologies. The 17-level prototype explores: Telemetry → OpenUSD → Live Synchronization → OEE/KPI Analytics → What-if Simulation → Decision Engine → Decision Feedback Technology stack includes OpenUSD, NVIDIA usd-exchange, MQTT, Python and SQLite, w
I guess it for you to tell - try, it does not bite. It’s faster and now on par with quality
Is your repo the winner (I dont mind if you say so). I have been using Mia one, but been getting a bit loopy at higher contexts, so been firing up Mia’s GLM to get through to solutions.
The pins are named UART4_TX and UART4_RX in the Jetson Thor pin description. However, UART4 is not listed in the Design Guide, and I cannot find any UART function for these pins in the pinmux spreadsheet. Can these pins be configured as UART4 and used as a standard UART interface, or are they reserved for internal use only?
the.elite.pro.0224: /* Create gstreamer elements */ /* Create Pipeline element that will form a connection of other elements */ pipeline = gst_pipeline_new ("dstest-sr-pipeline"); // Create the NTP clock // Parameters: name, remote_address, remote_port, base_time GstClock *clock = gst_ntp_clock_new("ntp-clock", "time1.google.com", 123, 0); // Set the pipeline to use the NTP clock gst_pipeline_use_
On consumer GPUs (GeForce RTX) the peak performance not only depends on the A/B types (input matrices) but also on the C/D type (output/accumulation matrix), i.e. whether you are using FP16 or FP32 accumulation. This is not the case on workstation (RTX Pro) or datacenter GPUs, where peak throughput is the same for FP16 and FP32 accumulation. See appendices in this document for detailed GeForce RTX
stu.miller: ds4flash 0731 beats everything. My dual spark cluster up for almost 6 weeks flawlessly. 24:7 hard work. Not a single issue. Same here, ds4 flash 0731 is so good for my workflows. It accomplishes whatever I throw at it (using for SWE). And I’ve had almost zero sessions that annoyed me (you know, the ones where you start cussing at the agent - I know it’s not just a me-thing haha). But t
That job was a fine group effort! Thanks everyone - we have a winner now! Also, VLLM team added image rehoming from the tool calls to user part - no proxy needed anymore, harnesses can send messages in tools - it works, no error, correct vision. Mechanism (found in the image source): vllm/tokenizers/deepseek_v4_encoding.py ships merge_tool_messages() — the vLLM team ported DeepSeek’s own reference
Custom board; this Ethernet port connects directly to the Jetson; JP5.1.4.
appear NIM is not answering from 4AM ( from france ) so i think you are not the only one
According to the table below, Jetson Thor has four UART ports. UART3 (H62, K60) is marked as a debug UART. Is this port dedicated to debugging only, or can it be repurposed and used as a standard UART interface in a custom design?
This is with default sampling temp 1.0 ╭────────────────────────────────────────── 🏆 Benchmark Complete ──────────────────────────────────────────╮ │ │ │ Model: deepseek-ai/DeepSeek-V4-Flash-Vision-Exp │ │ Score: 93 / 100 │ │ Rating: ★★★★★ Excellent │ │ Benchmark: tool-eval-bench v2.6.1.dev57+g9ab686613 │ │ Engine: vLLM 0.28.1rc1.dev475+g6fbb00b18.d20260907 │ │ Max context: 1,048,576 tokens │ │ │
Hello NVIDIA Support Team, I am attempting to generate an API key on build.nvidia.com to access the NVIDIA NIM foundation models. However, I am completely blocked at the phone verification step. The Issue: The country code for Pakistan (+92) is missing from the automated SMS country selection dropdown menu, making it impossible for me to enter my number or receive an OTP. Registered Account Email:
I was able to find a workaround with Claude Opus 5 after few hours of trial and error: Update: still reproduces on 7.0.0-31 / 580.173.02, and a working workaround Following up with newer versions and a more precise trace. Same GPU (RTX 4060 Ti 16GB, AD106). Environment Ubuntu 24.04, kernel 7.0.0-31-generic (HWE) Driver 580.173.02 — reproduced identically with both the proprietary and the open kern
Hello, We develop realtime sensor applications for Jetson Orin Nano and thus require a realtime kernel for its predictable timings. On Monday September 7th the realtime pacakges went missing from the r39.2 main repo (Unable to locate package). While the common repo got an update for the 39.2.1 version on September 4th, the realtime repo looks to be left behind. When can we expect the realtime kern
Sorry for overlooking the fact that you are using Orin NX and cannot upgrade to the latest DeepStream version. Bindings for NvDsRoiMeta are already supported in the pyds for DS-8.0; you can simply port this functionality to DS-7.1. git clone --branch v1.2.0 --depth 1 \ https://github.com/NVIDIA-AI-IOT/deepstream_python_apps.git cd deepstream_python_apps git apply /path/to/roi_classifier_tensor_met