# NeuralMesh for Autonomous Vehicles

NeuralMesh for autonomous vehicles removes storage bottlenecks from sensor ingest through training and simulation, keeping GPUs fed and supporting safety compliance.

**Type:** Solution Brief

## Key Benefits of NeuralMesh

**Reproduce Any Training Run on Demand**.

- Version every dataset from ingest to production.
- Trace lineage to sensor frame.
- Support regulatory defense.

**Checkpoint Training Runs Without Stalls or Re-Runs**.

- Absorb checkpoint write storage constantly.
- Complete training cleanly.
- Protect expensive GPU cycles.

**Accelerate Simulation and Synthetic Data**.

- Cover gaps that real on-road driving misses.
- Replay synthetic sensor data.
- Generate rare edge cases fast.

**Maximize GPU Saturation Across AV Workloads**.

- Feed complex training clusters nonstop.
- Sustain 90%+ GPU utilization.
- Eliminate idle GPU cycles.

**Unify Storage across Edge, On-Premises, and Cloud**.

- Operate on all data in a single namespace.
- Share sensor & training data.
- Eliminate cross-tier copies.

**Handle Billions of Small Sensor Files**.

- Promote archive data to hot tier on demand.
- Scale past metadata limits caused by lots of small files.

**Integrate Directly with NVIDIA Platforms**.

- Validate training runs on DGX SuperPOD.
- Use GPUDirect Storage.
- Leverage NVMe-to-GPU I/O.

**Absorb Fleet Ingest Without Backlogs**.

- Clear pre-processing backlogs automatically.
- Ingest petabyte fleet data.
- Route data to training lake.

```json
{
  "_key": "rpqa-stk5v8n3jya",
  "_type": "statsBlock",
  "markDefs": null,
  "stats": [
    {
      "_key": "st0mwl38j6aj",
      "_type": "statItem",
      "label": "GPUs in a single NeuralMesh cluster",
      "value": ">200,000"
    },
    {
      "_key": "st1fpekpx7kl",
      "_type": "statItem",
      "label": "More bandwidth compared to legacy NAS",
      "value": "10x"
    },
    {
      "_key": "st2jawj0hen6",
      "_type": "statItem",
      "label": "Single-epoch time reduced from 14 days to 4 hours",
      "value": "14d→4h"
    }
  ]
}
```

```json
{
  "_key": "e1f2af61655d",
  "_type": "statsBlock",
  "markDefs": null,
  "stats": [
    {
      "_key": "cf9b3a142916",
      "_type": "statItem",
      "label": "GPU utilization, up from ~30%",
      "value": ">95%"
    },
    {
      "_key": "d7b28459cab0",
      "_type": "statItem",
      "label": "of data supported at 16TB/s all day, every day.",
      "value": "600+ PB"
    }
  ]
}
```

## Autonomous Vehicle Use Cases

**Sensor Data Capture and Ingest**.  
NeuralMesh eliminates pre-processing backlogs by ingesting PB-sized fleet data from LiDAR, radar, and camera sensors directly into the training data lakes.

**AV Model Training and Validation**.  
NeuralMesh delivers concurrent, high-bandwidth access across distributed workloads with frequent checkpointing, eliminating I/O waits and stalls.

**Data Curation and Active Learning**.  
NeuralMesh resolves LOSF challenges and enables instant promotion of archive data back into the active training set for faster iteration.

**Simulation and Synthetic Data Generation**.  
NeuralMesh feeds simulation workloads at the throughput required to produce, store, and replay synthetic sensor data for rare-scenario edge cases.

```json
{
  "_key": "17a0e1cd7aaf",
  "_type": "customerQuoteBlock",
  "markDefs": null,
  "quote": {
    "anonymizedAttribution": null,
    "anonymous": false,
    "authorName": "Oren Ben Ibghei",
    "companyLogo": {
      "_type": "image",
      "asset": {
        "_id": "image-b7ef86e0687399d2588a64a978d6f4ada7c7f87a-300x90-svg",
        "metadata": {
          "dimensions": {
            "_type": "sanity.imageDimensions",
            "aspectRatio": 3.3333333333333335,
            "height": 90,
            "width": 300
          },
          "lqip": "data:image/png;base64,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"
        },
        "url": "https://cdn.sanity.io/images/ult5g8gw/production/b7ef86e0687399d2588a64a978d6f4ada7c7f87a-300x90.svg"
      },
      "crop": null,
      "hotspot": null
    },
    "companyLogoMono": {
      "_type": "image",
      "asset": {
        "_id": "image-819c44eb44fce3651561956490ebc726f71c1e2c-300x90-svg",
        "metadata": {
          "dimensions": {
            "_type": "sanity.imageDimensions",
            "aspectRatio": 3.3333333333333335,
            "height": 90,
            "width": 300
          },
          "lqip": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAAGCAYAAADDl76dAAAACXBIWXMAAAsTAAALEwEAmpwYAAAA3ElEQVR4nDXQyysvcBAF8M9WWbC53iTvkhQKKQvKI4+iKLHyupIiWYiwu24plCX/q6bOb3Ga73dmzjnNgXWsYB4zWMAixjGJNdzhHhuYxghmszeHVSxhFN7wH0+4DR7xF1d4wRe+8YxL7OMEF9jFMXbQU4I/+MQrPkIq8X/pvUes8IDrmJ7H8BQ3ubSlBGvhENs4w16GBziKexGKWHFUNFs5dzPcmi2jtQSncns/xjCAvtSh5FV5DuMP2tGLruxMJOtBNJdgDbpTO4LGvzLpDKrXeLcFtVPmxSmzpl+xaiJ0OhcdTAAAAABJRU5ErkJggg=="
        },
        "url": "https://cdn.sanity.io/images/ult5g8gw/production/819c44eb44fce3651561956490ebc726f71c1e2c-300x90.svg"
      },
      "crop": null,
      "hotspot": null
    },
    "companyName": "Innoviz",
    "jobTitle": "IT Manager",
    "text": "\"WEKA's storage scalability and ability to grow the infrastructure without losing performance, was a key factor in the decision.\""
  }
}
```

[Download PDF](/api/resource-pdf?slug=neuralmesh-for-autonomous-vehicles)
