## NVIDIA Data Center Deep Learning Product Performance

## Reproducible Performance

Learn how to lower your cost per token and maximize AI models with [The IT Leader’s Guide to AI Inference and Performance](https://www.nvidia.com/en-us/solutions/ai/inference/balancing-cost-latency-and-performance-ebook/).

* * *

## View Performance Data For:

Latest NVIDIA Data Center Products

 ![Training networks to convergence allows AI deployment in real-world applications](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/deeplearning/training-to-convergence-630x354.jpg)

### Training to Convergence  

Deploying AI in real-world applications requires training networks to convergence at a specified accuracy. This is the best methodology to test whether AI systems are ready to be deployed in the field to deliver meaningful results.

[Learn More](https://developer.nvidia.com/deep-learning-performance-training-inference/training)

 ![AI inference lets customers quickly deploy AI models into real-world production](https://d29g4g2dyqv443.cloudfront.net/sites/default/files/akamai/deeplearning/ai-inference-630x354.jpg)

### AI Inference  

Real-world inferencing demands high throughput and low latencies with maximum efficiency across use cases. An industry-leading solution lets customers quickly deploy AI models into real-world production with the highest performance from data center to edge.

[Learn More](https://developer.nvidia.com/deep-learning-performance-training-inference/ai-inference)

 ![High-Performance Computing (HPC) Acceleration](https://developer.download.nvidia.com/images/hpc-t500-devzone-630x354.jpg)

### High-Performance Computing (HPC) Acceleration  

Modern HPC data centers are crucial for solving key scientific and engineering challenges. NVIDIA Data Center GPUs transform data centers, delivering breakthrough performance with reduced networking overhead, resulting in 5X–10X cost savings.

[Learn More](https://developer.nvidia.com/nvidia-hpc-application-performance-v100-t4)

* * *

#### NVIDIA Blackwell Ultra Delivers up to 50x Better Performance and 35x Lower Cost for Agentic AI  

Built to accelerate the next generation of agentic AI, NVIDIA Blackwell Ultra delivers breakthrough inference performance with dramatically lower cost. Cloud providers such as Microsoft, CoreWeave, and Oracle Cloud Infrastructure are deploying NVIDIA GB300 NVL72 systems at scale for low-latency and long-context use cases, such as agentic coding and coding assistants.  
  
This is enabled by deep co-design across NVIDIA Blackwell, NVLink™, and NVLink Switch for scale-out; NVFP4 for low-precision accuracy; and NVIDIA Dynamo and TensorRT™ LLM for speed and flexibility—as well as development with community frameworks SGLang, vLLM, and more.

[Explore technical results](https://developer.nvidia.com/deep-learning-performance-training-inference/ai-inference)

![Data center illustration showing multi-modal AI tokens for image, audio, visual and more as part of the NVIDIA “Think SMART” framework.](https://developer.download.nvidia.com/images/dgx-press-gb300-1920x1080.jpg)

## Deep Learning Product Performance Resources



| title | featured | x_formats | document_url | technologies | document_date | short_summary | document_title | learning_level | x_content_types |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Run Key Genomics and Protein Folding Workloads Faster with NVIDIA RTX PRO 4500 Blackwell | false | blog | https://developer.nvidia.com/blog/run-key-genomics-and-protein-folding-workloads-faster-with-nvidia-rtx-pro-4500-blackwell/ | BioNeMo, Blackwell, NVIDIA Parabricks, RTX GPU | 2026-05-26T13:00:00.000Z | Accelerate genomics and protein folding workloads on workstation-class RTX PRO 4500 GPUs. | Run Key Genomics and Protein Folding Workloads Faster with NVIDIA RTX PRO 4500 Blackwell | Technical - Intermediate | Explainer |
| Achieving Peak System and Workload Efficiency on NVIDIA GB200 NVL72 with Slurm Block Scheduling | false | blog | https://developer.nvidia.com/blog/achieving-peak-system-and-workload-efficiency-on-nvidia-gb200-nvl72-with-slurm-block-scheduling/ | Blackwell, GB200 | 2026-05-07T18:20:14.000Z | Configure Slurm block scheduling to extract peak efficiency from GB200 NVL72 racks. | Achieving Peak System and Workload Efficiency on NVIDIA GB200 NVL72 with Slurm Block Scheduling | Technical - Intermediate | How-to |
| Transform Video Into Instantly Searchable, Actionable Intelligence with AI Agents and Skills | false | blog | https://developer.nvidia.com/blog/transform-video-into-instantly-searchable-actionable-intelligence-with-ai-agents-and-skills/ | Metropolis | 2026-05-13T15:00:00.000Z | Build searchable, real-time video intelligence agents using the AI-Q Blueprint and VLMs. | Transform Video Into Instantly Searchable, Actionable Intelligence with AI Agents and Skills | Technical - Intermediate | Explainer |
| Unlock Exascale Performance on NVIDIA GB200 NVL72 with Slurm Topology-Aware Job Scheduling | false | blog | https://developer.nvidia.com/blog/unlock-exascale-performance-on-nvidia-gb200-nvl72-with-slurm-topology-aware-job-scheduling/ | Blackwell, GB200 | 2026-05-21T14:32:56.000Z | Configure Slurm topology-aware scheduling for exascale jobs on GB200 NVL72 clusters. | Unlock Exascale Performance on NVIDIA GB200 NVL72 with Slurm Topology-Aware Job Scheduling | Technical - Intermediate | How-to |
| Accelerated X-Ray Analysis for Nanoscale Imaging (XANI) of Novel Materials | false | blog | https://developer.nvidia.com/blog/accelerated-x-ray-analysis-for-nanoscale-imaging-xani-of-novel-materials/ | Blackwell, CUDA Toolkit, cuPyNumeric, GB200 | 2026-05-13T13:39:20.000Z | Accelerate nanoscale X-ray imaging analysis using cuPyNumeric on GB200 NVL72. | Accelerated X-Ray Analysis for Nanoscale Imaging (XANI) of Novel Materials | Technical - Advanced | Explainer |
| Streaming Tokens and Tools: Multi-Turn Agentic Harness Support in NVIDIA Dynamo | false | blog | https://developer.nvidia.com/blog/streaming-tokens-and-tools-multi-turn-agentic-harness-support-in-nvidia-dynamo/ | Blackwell, Dynamo | 2026-05-08T12:59:16.000Z | Stream tokens and tools across multi-turn agentic workflows using NVIDIA Dynamo. | Streaming Tokens and Tools: Multi-Turn Agentic Harness Support in NVIDIA Dynamo | Technical - Advanced | Explainer |
| Building for the Rising Complexity of Agentic Systems with Extreme Co-Design | false | blog | https://developer.nvidia.com/blog/building-for-the-rising-complexity-of-agentic-systems-with-extreme-co-design/ | Blackwell, Dynamo | 2026-05-05T12:52:15.000Z | Explore extreme co-design strategies for the rising complexity of agentic AI systems. | Building for the Rising Complexity of Agentic Systems with Extreme Co-Design | Technical - Beginner | Explainer |
| Speed Up Unreal Engine NNE Inference with NVIDIA TensorRT for RTX Runtime | false | blog | https://developer.nvidia.com/blog/speed-up-unreal-engine-nne-inference-with-nvidia-tensorrt-for-rtx-runtime/ | RTX GPU, TensorRT | 2026-04-30T14:00:00.000Z | Speed up Unreal Engine Neural Network Engine inference using TensorRT for RTX runtime. | Speed Up Unreal Engine NNE Inference with NVIDIA TensorRT for RTX Runtime | Technical - Intermediate | How-to |
| Model Quantization: Post-Training Quantization Using NVIDIA Model Optimizer | false | blog | https://developer.nvidia.com/blog/model-quantization-post-training-quantization-using-nvidia-model-optimizer/ | RTX GPU, TensorRT | 2026-05-07T18:18:06.000Z | Apply post-training quantization on consumer RTX GPUs using NVIDIA Model Optimizer. | Model Quantization: Post-Training Quantization Using NVIDIA Model Optimizer | Technical - Intermediate | Tutorial |
| How to Build In-Vehicle AI Agents with NVIDIA: From Cloud to Car | false | blog | https://developer.nvidia.com/blog/how-to-build-in-vehicle-ai-agents-with-nvidia-from-cloud-to-car/ | DRIVE, Nemotron, TensorRT-LLM | 2026-05-05T13:00:00.000Z | Build agentic, multimodal in-vehicle assistants using DRIVE Thor, Nemotron, and TensorRT-LLM. | How to Build In-Vehicle AI Agents with NVIDIA: From Cloud to Car | Technical - Intermediate | How-to |
| How to Eliminate Pipeline Friction in AI Model Serving | false | blog | https://developer.nvidia.com/blog/how-to-eliminate-pipeline-friction-in-ai-model-serving/ | Dynamo, TensorRT | 2026-05-12T15:00:00.000Z | Remove model export, conversion, and serving friction using Triton, ONNX, and TensorRT. | How to Eliminate Pipeline Friction in AI Model Serving | Technical - Intermediate | How-to |
| Scaling the AI-Ready Data Center with NVIDIA RTX PRO 4500 Blackwell Server Edition and NVIDIA vGPU 20 | false | blog | https://developer.nvidia.com/blog/scaling-the-ai-ready-data-center-with-nvidia-rtx-pro-4500-blackwell-server-edition-and-nvidia-vgpu-20/ | Blackwell, RTX GPU | 2026-04-22T17:30:00.000Z | Scale enterprise AI data centers with RTX PRO 4500 Blackwell Server Edition and vGPU 20. | Scaling the AI-Ready Data Center with NVIDIA RTX PRO 4500 Blackwell Server Edition and NVIDIA vGPU 20 | Technical - Intermediate | Explainer |
| Scaling Token Factory Revenue and AI Efficiency by Maximizing Performance per Watt | false | blog | https://developer.nvidia.com/blog/scaling-token-factory-revenue-and-ai-efficiency-by-maximizing-performance-per-watt/ | Blackwell, Omniverse | 2026-03-25T08:00:00.000Z | Maximize AI factory revenue per watt across Blackwell, Hopper, and Vera Rubin systems. | Scaling Token Factory Revenue and AI Efficiency by Maximizing Performance per Watt | Technical - Intermediate | Explainer |
| Scaling Biomolecular Modeling Using Context Parallelism in NVIDIA BioNeMo | false | blog | https://developer.nvidia.com/blog/scaling-biomolecular-modeling-using-context-parallelism-in-nvidia-bionemo/ | BioNeMo, TensorRT | 2026-04-28T16:00:00.000Z | Scale biomolecular models using BioNeMo context parallelism to break single-GPU memory limits. | Scaling Biomolecular Modeling Using Context Parallelism in NVIDIA BioNeMo | Technical - Advanced | Explainer |
| Accelerating Vision AI Pipelines with Batch Mode VC-6 and NVIDIA Nsight | false | blog | https://developer.nvidia.com/blog/accelerating-vision-ai-pipelines-with-batch-mode-vc-6-and-nvidia-nsight/ | Blackwell, Nsight Compute | 2026-04-02T17:00:00.000Z | Profile and accelerate vision AI pipelines using batch-mode VC-6 codec and NVIDIA Nsight. | Accelerating Vision AI Pipelines with Batch Mode VC-6 and NVIDIA Nsight | Technical - Intermediate | How-to |
| NVIDIA IGX Thor Powers Industrial, Medical, and Robotics Edge AI Applications | false | blog | https://developer.nvidia.com/blog/nvidia-igx-thor-powers-industrial-medical-and-robotics-edge-ai-applications/ | Blackwell, Jetson | 2026-03-23T17:24:17.000Z | Deploy industrial, medical, and robotics edge AI applications on NVIDIA IGX Thor. | NVIDIA IGX Thor Powers Industrial, Medical, and Robotics Edge AI Applications | Technical - Beginner | Overview |
| Build with DeepSeek V4 Using NVIDIA Blackwell and GPU-Accelerated Endpoints | false | blog | https://developer.nvidia.com/blog/build-with-deepseek-v4-using-nvidia-blackwell-and-gpu-accelerated-endpoints/ | Blackwell | 2026-04-24T20:29:56.000Z | Build applications with DeepSeek-V4-Pro and V4-Flash using GPU-accelerated endpoints. | Build with DeepSeek V4 Using NVIDIA Blackwell and GPU-Accelerated Endpoints | Technical - Intermediate | How-to |
| How to Build, Run, and Scale High-Quality Creator Workflows in ComfyUI | false | blog | https://developer.nvidia.com/blog/how-to-build-run-and-scale-high-quality-creator-workflows-in-comfyui/ | Blackwell, RTX GPU | 2026-04-30T13:16:04.000Z | Construct production-grade ComfyUI generative workflows that scale across RTX hardware. | How to Build, Run, and Scale High-Quality Creator Workflows in ComfyUI | Technical - Beginner | Tutorial |
| Achieving Single-Digit Microsecond Latency Inference for Capital Markets | false | blog | https://developer.nvidia.com/blog/achieving-single-digit-microsecond-latency-inference-for-capital-markets/ | Blackwell | 2026-04-02T13:00:00.000Z | Achieve single-digit microsecond inference latency on Blackwell for algorithmic trading. | Achieving Single-Digit Microsecond Latency Inference for Capital Markets | Technical - Advanced | Explainer |
| Scaling Autonomous AI Agents and Workloads with NVIDIA DGX Spark | false | blog | https://developer.nvidia.com/blog/scaling-autonomous-ai-agents-and-workloads-with-nvidia-dgx-spark/ | Isaac Lab, TensorRT-LLM | 2026-03-16T17:30:00.000Z | Run autonomous AI agents and long-context workloads on NVIDIA DGX Spark. | Scaling Autonomous AI Agents and Workloads with NVIDIA DGX Spark | Technical - Intermediate | Explainer |
| Deploying Disaggregated LLM Inference Workloads on Kubernetes | false | blog | https://developer.nvidia.com/blog/deploying-disaggregated-llm-inference-workloads-on-kubernetes/ | Blackwell, Dynamo | 2026-03-23T04:01:00.000Z | Deploy disaggregated LLM inference workloads on Kubernetes with NVIDIA Dynamo. | Deploying Disaggregated LLM Inference Workloads on Kubernetes | Technical - Advanced | How-to |
| Introducing NVIDIA BlueField-4-Powered CMX Context Memory Storage Platform for the Next Frontier of AI | false | blog | https://developer.nvidia.com/blog/introducing-nvidia-bluefield-4-powered-inference-context-memory-storage-platform-for-the-next-frontier-of-ai/ | Blackwell, Dynamo | 2026-03-16T17:30:00.000Z | Discover the BlueField-4 CMX platform for scaling agentic AI context memory storage. | Introducing NVIDIA BlueField-4-Powered CMX Context Memory Storage Platform for the Next Frontier of AI | Technical - Beginner | Overview |
| Enhancing Distributed Inference Performance with the NVIDIA Inference Transfer Library | false | blog | https://developer.nvidia.com/blog/enhancing-distributed-inference-performance-with-the-nvidia-inference-transfer-library/ | Dynamo, NVIDIA Inference Xfer Library (NIXL) | 2026-03-09T14:00:00.000Z | Speed up distributed inference data transfers using the NVIDIA Inference Transfer Library. | Enhancing Distributed Inference Performance with the NVIDIA Inference Transfer Library | Technical - Advanced | Explainer |
| NVIDIA Vera Rubin POD: Seven Chips, Five Rack-Scale Systems, One AI Supercomputer | false | blog | https://developer.nvidia.com/blog/nvidia-vera-rubin-pod-seven-chips-five-rack-scale-systems-one-ai-supercomputer/ | Blackwell, Vera Rubin | 2026-03-16T13:05:58.000Z | Understand the Vera Rubin POD architecture: seven chips, five rack-scale systems, one supercomputer. | NVIDIA Vera Rubin POD: Seven Chips, Five Rack-Scale Systems, One AI Supercomputer | Technical - Beginner | Overview |
| NVIDIA RTX Innovations Are Powering the Next Era of Game Development | false | blog | https://developer.nvidia.com/blog/nvidia-rtx-innovations-are-powering-the-next-era-of-game-development/ | Blackwell, CloudXR, DLSS, Nsight Graphics, RTX Kit | 2026-03-10T12:30:00.000Z | Discover NVIDIA RTX ray tracing and neural rendering innovations shaping next-generation games. | NVIDIA RTX Innovations Are Powering the Next Era of Game Development | Technical - Beginner | Overview |
| Validate Kubernetes for GPU Infrastructure with Layered, Reproducible Recipes | false | blog | https://developer.nvidia.com/blog/validate-kubernetes-for-gpu-infrastructure-with-layered-reproducible-recipes/ | Blackwell, Dynamo | 2026-03-12T13:30:00.000Z | Validate Kubernetes GPU infrastructure using layered, reproducible recipes. | Validate Kubernetes for GPU Infrastructure with Layered, Reproducible Recipes | Technical - Intermediate | How-to |
| Build AI-Ready Knowledge Systems Using 5 Essential Multimodal RAG Capabilities | false | blog | https://developer.nvidia.com/blog/build-ai-ready-knowledge-systems-using-5-essential-multimodal-rag-capabilities/ | Nemotron | 2026-02-17T15:00:00.000Z | Build AI-ready knowledge systems using five essential multimodal RAG capabilities. | Build AI-Ready Knowledge Systems Using 5 Essential Multimodal RAG Capabilities | Technical - Intermediate | Explainer |
| Automating Inference Optimizations with NVIDIA TensorRT LLM AutoDeploy | false | blog | https://developer.nvidia.com/blog/automating-inference-optimizations-with-nvidia-tensorrt-llm-autodeploy/ | Blackwell, TensorRT, TensorRT-LLM | 2026-02-09T15:30:00.000Z | Automate LLM inference optimizations and deployment using TensorRT LLM AutoDeploy. | Automating Inference Optimizations with NVIDIA TensorRT LLM AutoDeploy | Technical - Intermediate | Tutorial |
| Making Softmax More Efficient with NVIDIA Blackwell Ultra | false | blog | https://developer.nvidia.com/blog/making-softmax-more-efficient-with-nvidia-blackwell-ultra/ | Blackwell, GB200 | 2026-02-25T14:00:00.000Z | Explore softmax kernel optimizations for MLA and GQA attention on Blackwell Ultra. | Making Softmax More Efficient with NVIDIA Blackwell Ultra | Technical - Advanced | Explainer |
| How NVIDIA Dynamo 1.0 Powers Multi-Node Inference at Production Scale | false | blog | https://developer.nvidia.com/blog/nvidia-dynamo-1-production-ready/ | Blackwell, Dynamo | 2026-03-16T17:30:00.000Z | See how Dynamo 1.0 enables production-scale multi-node inference for reasoning models. | How NVIDIA Dynamo 1.0 Powers Multi-Node Inference at Production Scale | Technical - Beginner | News |
| Removing the Guesswork from Disaggregated Serving | false | blog | https://developer.nvidia.com/blog/removing-the-guesswork-from-disaggregated-serving/ | Blackwell, Dynamo | 2026-03-09T13:00:00.000Z | Remove guesswork from disaggregated LLM serving using Dynamo configuration tuning. | Removing the Guesswork from Disaggregated Serving | Technical - Advanced | Explainer |
| Accelerating LLM and VLM Inference for Automotive and Robotics with NVIDIA TensorRT Edge-LLM | false | blog | https://developer.nvidia.com/blog/accelerating-llm-and-vlm-inference-for-automotive-and-robotics-with-nvidia-tensorrt-edge-llm/ | DRIVE, JetPack SDK, Jetson, TensorRT-LLM | 2026-01-08T14:28:49.000Z | Run LLM and VLM inference at the edge for automotive and robotics with TensorRT Edge-LLM. | Accelerating LLM and VLM Inference for Automotive and Robotics with NVIDIA TensorRT Edge-LLM | Technical - Intermediate | Explainer |
| Smart Multi-Node Scheduling for Fast and Efficient LLM Inference with NVIDIA Run:ai and NVIDIA Dynamo | false | blog | https://developer.nvidia.com/blog/smart-multi-node-scheduling-for-fast-and-efficient-llm-inference-with-nvidia-runai-and-nvidia-dynamo/ | Blackwell, Dynamo | 2025-09-29T12:00:00.000Z | Schedule fast, efficient multi-node LLM inference using NVIDIA Run:ai and NVIDIA Dynamo. | Smart Multi-Node Scheduling for Fast and Efficient LLM Inference with NVIDIA Run:ai and NVIDIA Dynamo | Technical - Advanced | Explainer |
| Scaling Large MoE Models with Wide Expert Parallelism on NVL72 Rack Scale Systems | false | blog | https://developer.nvidia.com/blog/scaling-large-moe-models-with-wide-expert-parallelism-on-nvl72-rack-scale-systems/ | Blackwell, Dynamo, GB200 | 2025-10-20T13:00:00.000Z | Scale large MoE models using wide expert parallelism on NVL72 rack-scale systems. | Scaling Large MoE Models with Wide Expert Parallelism on NVL72 Rack Scale Systems | Technical - Advanced | Explainer |
| Accelerating Long-Context Inference with Skip Softmax in NVIDIA TensorRT LLM | false | blog | https://developer.nvidia.com/blog/accelerating-long-context-inference-with-skip-softmax-in-nvidia-tensorrt-llm/ | Blackwell, GB200, TensorRT-LLM | 2025-12-16T18:00:00.000Z | Accelerate long-context LLM inference using Skip Softmax optimizations in TensorRT LLM. | Accelerating Long-Context Inference with Skip Softmax in NVIDIA TensorRT LLM | Technical - Advanced | Explainer |
| Optimizing Communication for Mixture-of-Experts Training with Hybrid Expert Parallel | false | blog | https://developer.nvidia.com/blog/optimizing-communication-for-mixture-of-experts-training-with-hybrid-expert-parallel/ | Blackwell | 2026-02-02T15:43:08.000Z | Optimize all-to-all communication for hyperscale MoE training using Hybrid Expert Parallel. | Optimizing Communication for Mixture-of-Experts Training with Hybrid Expert Parallel | Technical - Advanced | Explainer |
| Adaptive Inference in NVIDIA TensorRT for RTX Enables Automatic Optimization | false | blog | https://developer.nvidia.com/blog/adaptive-inference-in-nvidia-tensorrt-for-rtx-enables-automatic-optimization/ | RTX GPU, TensorRT | 2026-01-26T18:00:00.000Z | Enable automatic optimization across consumer GPUs using adaptive inference in TensorRT for RTX. | Adaptive Inference in NVIDIA TensorRT for RTX Enables Automatic Optimization | Technical - Intermediate | Explainer |
| Scaling NVFP4 Inference for FLUX.2 on NVIDIA Blackwell Data Center GPUs | false | blog | https://developer.nvidia.com/blog/scaling-nvfp4-inference-for-flux-2-on-nvidia-blackwell-data-center-gpus/ | Blackwell, GB200, TensorRT-LLM | 2026-01-22T16:21:07.000Z | Scale FLUX.2 image generation using NVFP4 inference on Blackwell data center GPUs. | Scaling NVFP4 Inference for FLUX.2 on NVIDIA Blackwell Data Center GPUs | Technical - Intermediate | Explainer |
| Inside the NVIDIA Vera Rubin Platform: Six New Chips, One AI Supercomputer | false | blog | https://developer.nvidia.com/blog/inside-the-nvidia-rubin-platform-six-new-chips-one-ai-supercomputer/ | Dynamo, Vera Rubin | 2026-01-05T19:20:12.000Z | Explore the six chips inside the Vera Rubin platform that form one AI supercomputer. | Inside the NVIDIA Vera Rubin Platform: Six New Chips, One AI Supercomputer | Technical - Beginner | Overview |
| NVIDIA Blackwell Leads on SemiAnalysis InferenceMAX v1 Benchmarks | false | blog | https://developer.nvidia.com/blog/nvidia-blackwell-leads-on-new-semianalysis-inferencemax-benchmarks/ | Blackwell, Dynamo, GB200, TensorRT-LLM | 2025-10-13T14:33:19.000Z | Review Blackwell inference performance results on SemiAnalysis InferenceMAX v1 benchmarks. | NVIDIA Blackwell Leads on SemiAnalysis InferenceMAX v1 Benchmarks | Technical - Beginner | News |
| Streamline Complex AI Inference on Kubernetes with NVIDIA Grove | false | blog | https://developer.nvidia.com/blog/streamline-complex-ai-inference-on-kubernetes-with-nvidia-grove/ | Blackwell, Dynamo | 2025-11-10T11:00:00.000Z | Streamline complex multi-component AI inference on Kubernetes using NVIDIA Grove. | Streamline Complex AI Inference on Kubernetes with NVIDIA Grove | Technical - Intermediate | Explainer |
| NVIDIA Accelerates OpenAI gpt-oss Models Delivering 1.5 M TPS Inference on NVIDIA GB200 NVL72 | false | blog | https://developer.nvidia.com/blog/delivering-1-5-m-tps-inference-on-nvidia-gb200-nvl72-nvidia-accelerates-openai-gpt-oss-models-from-cloud-to-edge/ | Blackwell, Dynamo, GB200 | 2025-08-05T14:10:00.000Z | See how NVIDIA accelerates OpenAI gpt-oss models to 1.5M TPS on GB200 NVL72. | NVIDIA Accelerates OpenAI gpt-oss Models Delivering 1.5 M TPS Inference on NVIDIA GB200 NVL72 | Technical - Beginner | News |
| Dynamo 0.4 Delivers 4x Faster Performance, SLO-Based Autoscaling, and Real-Time Observability | false | blog | https://developer.nvidia.com/blog/dynamo-0-4-delivers-4x-faster-performance-slo-based-autoscaling-and-real-time-observability/ | Blackwell, Dynamo | 2025-08-13T12:30:00.000Z | Discover Dynamo 0.4 features including 4× faster performance and SLO-based autoscaling. | Dynamo 0.4 Delivers 4x Faster Performance, SLO-Based Autoscaling, and Real-Time Observability | Technical - Beginner | News |
| How the NVIDIA Vera Rubin Platform is Solving Agentic AI’s Scale-Up Problem | true | blog | https://developer.nvidia.com/blog/how-the-nvidia-vera-rubin-platform-is-solving-agentic-ais-scale-up-problem/ | Dynamo, Vera Rubin | 2026-05-14T16:24:35.000Z | Explore how Vera Rubin scale-up architecture handles non-deterministic agentic inference workloads. | How the NVIDIA Vera Rubin Platform is Solving Agentic AI’s Scale-Up Problem | Technical - Intermediate | Explainer |
| Introducing NVIDIA Jetson Thor, the Ultimate Platform for Physical AI | false | blog | https://developer.nvidia.com/blog/introducing-nvidia-jetson-thor-the-ultimate-platform-for-physical-ai/ | Blackwell, Cosmos, Jetson, Metropolis, NVIDIA Holoscan, NVIDIA Isaac GROOT | 2025-08-25T14:57:00.000Z | Discover Jetson Thor, the new edge platform for generalist robots and physical AI. | Introducing NVIDIA Jetson Thor, the Ultimate Platform for Physical AI | Technical - Beginner | Overview |
| Optimizing LLMs for Performance and Accuracy with Post-Training Quantization | false | blog | https://developer.nvidia.com/blog/optimizing-llms-for-performance-and-accuracy-with-post-training-quantization/ | Blackwell, TensorRT | 2025-08-01T18:27:23.000Z | Optimize LLM latency, throughput, and memory using post-training quantization techniques. | Optimizing LLMs for Performance and Accuracy with Post-Training Quantization | Technical - Intermediate | Explainer |
| An Introduction to Speculative Decoding for Reducing Latency in AI Inference | false | blog | https://developer.nvidia.com/blog/an-introduction-to-speculative-decoding-for-reducing-latency-in-ai-inference/ | TensorRT, TensorRT-LLM | 2025-09-17T15:09:12.000Z | Get introduced to speculative decoding techniques that reduce LLM inference latency. | An Introduction to Speculative Decoding for Reducing Latency in AI Inference | Technical - Beginner | Explainer |
| Deploy High-Performance AI Models in Windows Applications on NVIDIA RTX AI PCs | false | blog | https://developer.nvidia.com/blog/deploy-ai-models-faster-with-windows-ml-on-rtx-pcs/ | RTX GPU, TensorRT | 2025-09-23T16:20:46.000Z | Deploy high-performance AI models in Windows applications on NVIDIA RTX AI PCs using Windows ML. | Deploy High-Performance AI Models in Windows Applications on NVIDIA RTX AI PCs | Technical - Intermediate | How-to |
| Full-Stack Optimizations for Agentic Inference with NVIDIA Dynamo | true | blog | https://developer.nvidia.com/blog/full-stack-optimizations-for-agentic-inference-with-nvidia-dynamo/ | Blackwell, Dynamo | 2026-04-17T19:52:47.000Z | Optimize agentic inference end-to-end using NVIDIA Dynamo across the full software stack. | Full-Stack Optimizations for Agentic Inference with NVIDIA Dynamo | Technical - Advanced | Explainer |
| How to Reduce KV Cache Bottlenecks with NVIDIA Dynamo | false | blog | https://developer.nvidia.com/blog/how-to-reduce-kv-cache-bottlenecks-with-nvidia-dynamo/ | Blackwell, Dynamo | 2025-09-18T13:30:00.000Z | Reduce KV cache bottlenecks in LLM inference using NVIDIA Dynamo cache management. | How to Reduce KV Cache Bottlenecks with NVIDIA Dynamo | Technical - Intermediate | How-to |
| Deploying Your Omniverse Kit Apps at Scale | false | blog | https://developer.nvidia.com/blog/deploying-your-omniverse-kit-apps-at-scale/ | Blackwell, Omniverse, RTX GPU | 2025-08-20T10:00:00.000Z | Deploy Omniverse Kit-based 3D applications at scale across NVIDIA infrastructure. | Deploying Your Omniverse Kit Apps at Scale | Technical - Intermediate | How-to |
| NVIDIA Platform Delivers Lowest Token Cost Enabled by Extreme Co-Design | true | blog | https://developer.nvidia.com/blog/nvidia-platform-delivers-lowest-token-cost-enabled-by-extreme-co-design/ | Blackwell, Dynamo, TensorRT-LLM, Vera Rubin | 2026-04-01T12:00:48.000Z | Understand how hardware-software co-design across the NVIDIA platform delivers the lowest token cost. | NVIDIA Platform Delivers Lowest Token Cost Enabled by Extreme Co-Design | Technical - Beginner | Explainer |
| 3 Ways NVFP4 Accelerates AI Training and Inference | true | blog | https://developer.nvidia.com/blog/3-ways-nvfp4-accelerates-ai-training-and-inference/ | Blackwell | 2026-02-06T13:00:00.000Z | Examine how NVFP4 accelerates AI training and inference on Blackwell hardware. | 3 Ways NVFP4 Accelerates AI Training and Inference | Technical - Beginner | Explainer |
| Top 5 AI Model Optimization Techniques for Faster, Smarter Inference | true | blog | https://developer.nvidia.com/blog/top-5-ai-model-optimization-techniques-for-faster-smarter-inference/ | TensorRT | 2025-12-09T15:00:00.000Z | Survey the top five model optimization techniques for faster, more efficient inference. | Top 5 AI Model Optimization Techniques for Faster, Smarter Inference | Technical - Beginner | Overview |

[Download the raw results data (JSON)](https://developer.nvidia.com/search-data/deep_learning_performance.json)



## NVIDIA Data Center Deep Learning Product Performance FAQs

NVIDIA inference cost per million tokens has improved dramatically across generations: NVIDIA Blackwell Ultra (GB300 NVL72) delivers up to 50x higher throughput per megawatt and up to 35x lower cost per token than NVIDIA Hopper for low-latency agentic workloads, through hardware–software codesign, according to [SemiAnalysis InferenceX benchmarks](https://inferencex.semianalysis.com/) (Q1 2026). Software optimization drives continuous improvement—GB200 token output improved 4x in three months, resulting in a proportional decrease in token cost.

In [MLPerf Inference v6.0](https://mlcommons.org/benchmarks/) (April 2026), systems powered by NVIDIA Blackwell Ultra GPUs (GB300 NVL72) delivered the highest throughput across the widest range of models and scenarios. On DeepSeek-R1, GB300 NVL72 delivered 2.5 million tokens per second—up to 2.7x higher token throughput compared to GB300 NVL72 debut submissions just six months prior, as a result of NVIDIA TensorRT™-LLM software updates.

NVIDIA Blackwell B200 achieves up to 60,000 tokens per second per GPU on GPT-OSS-120B with the latest TensorRT-LLM stack, according to [SemiAnalysis InferenceX benchmarks](https://inferencex.semianalysis.com/)as of April 2026—representing a roughly 4x throughput improvement over H200 with TensorRT-LLM. This level of throughput allows NVIDIA Blackwell B200 to achieve $0.02 per million tokens on the same model using TensorRT-LLM.

NVIDIA's TensorRT-LLM and Dynamo software stack delivers continuous inference cost improvements without hardware changes. NVIDIA Blackwell B200 cost per million tokens dropped from $0.11 at launch to $0.02 on GPT-OSS-120B within two months, according to [SemiAnalysis InferenceX benchmarks](https://inferencex.semianalysis.com/) as of April 2026—a 5x improvement from software alone. Each TensorRT-LLM release typically delivers throughput gains through kernel fusion, quantization improvements, and scheduling optimizations.

Explore software containers, models, Jupyter notebooks, and documentation.

[NVIDIA NGC Catalog  
](https://catalog.ngc.nvidia.com/collections?filters=&orderBy=weightPopularDESC&query=&page=&pageSize=)