Skip to main content

Vastai Semiconductor SV Series

Vendor: Vastai Semiconductor

Category: GPU Graphics Processor

Architecture: Vastai

Introduction​

Vastai Semiconductor high-end GPU chips with two generations of proprietary GPU cores. SV series covers three product lines: graphics rendering GPU, data center AI GPU, and edge AI GPU. One of the few domestic vendors with commercial deployment in both general AI and rendering. SV chip deep learning inference throughput reaches 2× that of general-purpose GPUs.

Specifications​

ModelComputeMemoryInterfaceTDPProcess
VA12160 TFLOPS (FP16) / 512 (INT8)32GB HBM2ePCIe 4.0250W7nm
VA1L72 TFLOPS (FP16) / 200 (INT8)16GB GDDR6PCIe 4.0150W7nm

Official Website​

Visit Official Website

Driver Downloads​

Linux​

OS Support​

WindowsLinuxmacOSAndroid
❌✅❌❌

Version History​

VersionRelease DateDescription
SDK 1.22023Video encoding optimization

Performance Benchmarks​

ModelTaskPerformance Metric
VA100Video Transcoding/AI InferenceINT8 optimized
VA1PEdge InferenceLow-power scenarios

Pricing​

ModelReference PriceNotes
VA100Contact vendorVideo processing + AI inference
VA1PContact vendorEdge inference dedicated

Quick Installation​

Linux​

# Install Vastai driver and SDK
sudo rpm -ivh vastai-driver-*.rpm

Code Examples​

Python (Vastai VA100)​

import torch
# Vastai GPU provides CUDA-compatible backend
assert torch.cuda.is_available()

Architecture Highlights​

  • Vastai Architecture: Vastai proprietary GPU architecture, focused on video processing and AI inference
  • Video + AI Dual Capability: Simultaneously supports video transcoding and AI inference, suitable for streaming media scenarios
  • VA1P: Low-power edge inference card

Model Compatibility​

Model/FrameworkSupportNotes
PyTorch⚠️ CompatibleCUDA-compatible backend
Video Transcoding✅Core capability
AI Inference⚠️Primary promoted scenario

If you're evaluating alternatives, the following products may also fit your scenario: