NIO is driving efficiency in autonomous driving workloads. By integrating HAMi (a CNCF Sandbox project), NIO has transformed its large-scale cloud infrastructure, achieving a 4x increase in GPU utilization for CI pipelines and reducing simulation time by 30%. NIO's team manages complex tasks including: * Model training & simulation * Online inference * GPU performance optimization Rather than a one-size-fits-all approach, they adopted a hybrid strategy combining NVIDIA MIG, time-slicing, and HAMi to balance isolation with resource sharing. Read the full case study to see how they scaled to 800+ GPUs: https://lnkd.in/gDWbT5Wr #CNCF #CloudNative #Kubernetes #AutonomousDriving #GPU #OpenSource #NIO
Cloud Native Computing Foundation (CNCF)’s Post
More Relevant Posts
-
4x more throughput. Billions of virtual miles. 🚗 GM rebuilt its entire AV simulation pipeline on Google Cloud G4 VMs — and the results speak for themselves ⚡ NVIDIA Blackwell GPUs let ML teams stress-test edge cases no test track could safely replicate 📉 Slow iteration cycles were killing momentum toward eyes-off autonomy — elastic cloud compute fixed that 🎯 Target locked: Cadillac ESCALADE IQ ships with eyes-off autonomous driving by 2028 Full breakdown 👇 https://lnkd.in/eQa2CfTp 🐯 AI of the Tiger 🐯
To view or add a comment, sign in
-
-
⚙️ High-Performance AI Infrastructure for Autonomous Driving Autonomous driving pushes the limits of AI compute, demanding dense, reliable, and scalable GPU systems. SafeAD is leveraging 12 GIGABYTE GPU servers, each configured with 8 Nvidia GPUs, to power large-scale training of AI models for autonomous driving applications. The training data is stored on 8 GIGABYTE storage servers containing 8 NVMe SSDs each that provide a total storage capacity of 983.04 TB. Our GPU-dense server platforms are designed to support demanding deep learning workloads, enabling faster model development and stable operation at scale. We’re proud to support SafeAD in building the infrastructure behind next-generation autonomous driving AI. #AIInfrastructure #HPC #GPUComputing #AutonomousDriving #DeepLearning #GIGABYTE
To view or add a comment, sign in
-
-
🚀 Testing Google’s New Gemma 4 on Real Hardware We ran hands-on tests with Gemma from Google Cloud — not in theory, but on real devices: ⚡ NGX Spark ASUS ⚡ RTX 3090 NVIDIA Watch here 👇 https://lnkd.in/gf3jUnAd ⸻ 💡 What we’re seeing AI is moving fast toward: • Real-time local inference • Lower latency systems • Full control over deployment This is exactly the direction we’re building at NavTalk real-time digital humans powered by low-latency infrastructure, voice, and streaming pipelines 🔥 Key insight 👉 Models are improving 👉 But infrastructure is the real differentiator “Avatars are not the product. Infrastructure is.” #NavTalk #RealtimeAI #AIInfrastructure #Gemma #Google #NVIDIA #ASUS #EdgeAI #DigitalHumans
Gemma 4 Performance Showdown on Real Devices: Jetson Orin Nano vs RTX 3090 vs NVIDIA DGX Spark
https://www.youtube.com/
To view or add a comment, sign in
-
🚀 Real Device AI Testing – This Is Where Things Get Interesting Last night, I ran some hands-on tests with Gemma (new release from Google Cloud) — not in theory, but on real hardware: ⚡ NGX Spark ASUS ⚡ RTX 3090 NVIDIA And the results are worth a look 👇 🎥 https://lnkd.in/gwCq9ds6 T ⸻ 💡 Why this matters We’re moving fast toward a world where: • AI runs locally, in real-time • Latency drops to near zero • Full control stays with the developer or enterprise This is exactly where platforms like NavTalk and NavBot AI are heading — combining AI models + real-time voice + digital humans on powerful local infrastructure. ⸻ 🔥 Key takeaway It’s no longer just about the model. 👉 It’s about how and where you run it Local GPUs + optimised stacks = ⚡ Faster 🔒 More secure 📡 Fully controllable ⸻ Big shoutout to the ecosystem pushing this forward: #NVIDIA #Gemma #Google #ASUS #AI #EdgeAI #RealtimeAI #DigitalHumans #NavTalk #NavBot ⸻ If you’re building in this space or exploring self-hosted AI infrastructure, this is the direction 👀
Gemma 4 Performance Showdown on Real Devices: Jetson Orin Nano vs RTX 3090 vs NVIDIA DGX Spark
https://www.youtube.com/
To view or add a comment, sign in
-
🙌 Congrats Google DeepMind, Google AI for Developers on the release of your Gemma 4 models!🎉 The new multimodal and multilingual models are built for fast, efficient, and secure AI across devices – and optimized to run locally on NVIDIA RTX, RTX PRO, DGX Spark, and Jetson. You can now also run demanding Gemma 4 inference workloads efficiently on Cloud Run, leveraging the power of G4 instance based on NVIDIA RTX PRO 6000 (Blackwell) GPUs. 👉 Prototype the 31B model and start experimenting for free on https://bit.ly/4cw3iWf 🔗Check out the details to get started in our Technical Blog: https://bit.ly/4mfhlno
To view or add a comment, sign in
-
-
Have you tested Nvidia Kimodo? I did the wait in que twice. Used my 5 mins. Love the idea, right in line with my take on the vision/robotics domain. Saw cool stuff but not enough to understand the business cases. I am building an adapter to ingest thier .bvh output for the Willow Dynamics Cloud Oracle. Anyone have a take? Is this thing viable yet for any real use cases? #nvidia #kimodo #nvidiakimodo #computervision #robotics #ros #nvidiaisaac #jetson #mediapipe #ultralytics #opencv
To view or add a comment, sign in
-
We’re scaling up our AI training infrastructure 🚀 Training autonomous driving AI at scale requires serious compute power. At SafeAD, we’ve expanded our in-house AI infrastructure to 12 high-performance GIGABYTE servers, each equipped with 8 Nvidia GPUs. This gives us a total of 96 GPUs dedicated to training and validating our autonomous driving models. 8 GIGABYTE storage servers, each equipped with 8 NVMe SSDs, provide a total storage capacity of 1 petabyte. This setup enables faster iteration cycles, larger and more complex models, and greater control over our AI development pipeline which is a critical step as we continue to scale our technology. We’re excited to work with GIGABYTE as our server infrastructure partner, providing the performance and reliability needed for real-world autonomous driving AI. #AutonomousDriving #AITraining #DeepLearning #GPUComputing #AIInfrastructure #SafeAD #GIGABYTE
To view or add a comment, sign in
-
-
Nebius AI Cloud 3.5 Deploys Serverless Compute with Blackwell GPUs 📌 Nebius AI Cloud 3.5 just dropped a game-changer: serverless compute powered by NVIDIA’s ultra-fast Blackwell GPUs - no cluster config, no infra headaches. Developers can now launch DevPods, Jobs, and Endpoints instantly for prototyping, training, or inference - while the RTX PRO 6000’s 96GB memory fuels spatial AI, robotics, and drug discovery. It’s not just faster - it’s smarter, scaling with your workloads, not the other way around. 🔗 Read more: https://lnkd.in/dZjB9Ycn #Nebiusai #Blackwellgpu #Serverlesscompute #Aetherplatform #Nvidia
To view or add a comment, sign in
-
A New Engine for Intelligence: The Rubin Platform Introducing the audience to pioneering American astronomer Vera Rubin, after whom NVIDIA named its next-generation computing platform, Huang announced that the NVIDIA Rubin platform, the successor to the record‑breaking NVIDIA Blackwell architecture and the company’s first extreme-codesigned, six‑chip AI platform, is now in full production. https://lnkd.in/dU3N7s38
To view or add a comment, sign in
-
Gemma 4 Inference on Raspberry Pi CM5 Stuck by Hardware Limits 📌 Running Google’s Gemma 4 LLMs locally on Raspberry Pi CM5 is hardware-limited, with memory and latency bottlenecks making real-time voice interaction nearly impossible without cloud offloading. Developers face tough trade-offs: even the smallest model (E2B) needs 5GB RAM, which gets eaten up by OS and robotics stacks. For true edge AI, dedicated accelerators like NVIDIA Jetson or AMD GPUs are essential - not just for speed, but for seamless, conversational robot behavior. 🔗 Read more: https://lnkd.in/dwCD_D8w #Gemma4 #Raspberrypicm5 #Multimodal #Localinference #Hardwarelimits
To view or add a comment, sign in