NVIDIA & AMD — NATIVE DUAL-PLATFORM SUPPORT

Checkpointing That
Doesn't Stop Training.

The async I/O engine that eliminates GPU idle time during checkpointing. 16+ GB/s effective throughput verified on NVIDIA Blackwell (RTX 5080), H100/B200 and AMD MI300X with bit-perfect integrity. Native low-level binaries for Linux & Windows.

Effective Throughput
16+ GB/s

Verified effective write speed on RTX 5080. Near-zero I/O overhead for incremental checkpointing through asynchronous hardware-native pipelining.

Efficiency & Integrity
10x+ Deduplication

Bit-Perfect Accuracy: Atomic commitment guarantees 100% restoration.
10x+ Ratio: Validated on LLaMA 8B sharded checkpoints and LoRA workloads.

Enterprise Ready & Hardware Native
NVIDIA + AMD Python 3.10–3.12 Support
NVIDIA: Native support for Hopper, Blackwell, Ampere, Turing. Fat binary: SM75→SM100+PTX
AMD: Native ROCm 6.x for MI300X (gfx942) and MI250X (gfx90a).
Distributed: DDP and FSDP natively supported. Slurm compatible.
Backends: Native io_uring (Linux) and IOCP (Windows). Fully offline operation for enterprise.
DeepSpeed and HF Accelerate native interception planned — contact us to prioritise.

Compiled Native Architecture

Pre-compiled fat binaries for NVIDIA CUDA and AMD ROCm. No source code distribution — native GPU performance with full IP protection.

SESSION: 7f3a9e2b
admin@gpu-cluster
user@lab:~$ pip install neuralio-2.2.5-cp312-linux_x86_64.whl
[Neural:IO] License: ENTERPRISE | Platform: CUDA + ROCm
[Kernel] NVIDIA FatBinary (SM75→SM100+PTX) | AMD (gfx942)
[Save] File: ckpt_step_4200.pt (4.0 GB)
[MARKET SUMMARY] Unique Data: 0.40 GB | Dedupe: 10.00x | Effective: 16.2 GB/s
[✓] Incremental checkpoint — 90% I/O eliminated

Command Center Dashboard

Monitor effective throughput, verifiable deduplication ratios, and track literal dollar savings with the integrated real-time telemetry dashboard.

Neural:IO Live Monitor Dashboard
Neural:IO Audit Trail Neural:IO Deep Configuration

Join the Private Pilot

Currently accepting partners for pilot programs on NVIDIA H100/B200 and AMD MI300X infrastructure. Windows & Linux.

Evaluate

24h Demo Tier

For rapid PoCs & Local Hardware Validation

$0
  • 24 Hour Token Lifespan
  • Full Dashboard Telemetry
  • Fat Binary Architecture
  • Best-effort support
Pilot Program

Enterprise Pilot

For H100/B200 & MI300X Clusters

Node-based (Pilot is Free)

Direct access to labs engineering. Custom kernel tuning for your specific GPU topology and storage stack.

  • Unlimited Multi-Node Scaling
  • NVIDIA + AMD Kernel Tuning
  • Windows & Linux Binaries
  • 24h Priority Email Support

Ready to Accelerate?

Get in touch for pilot access, partnership inquiries, or to schedule a technical demo.