Vastai Tech QianYuan SG100 (2023)
Product Overview
QianYuan SG100 is Vastai Tech's second-generation GPU chip officially released on July 6, 2023 at the 2023 World Artificial Intelligence Conference, using 7nm advanced process, is a full-function GPU integrating rendering, AI, and video, with industry-leading rendering performance, while also having low-latency high-throughput AI computing power and powerful video processing capabilities.
Positioning: Full-function GPU (graphics rendering + AI inference + video processing), targeting some scenarios of NVIDIA A100, is an important breakthrough for China's full-function GPU.
Core Specifications
| Item | Parameter |
|---|
| Architecture | Vastai second-generation GPU architecture (QianYuan series) |
| Process | 7nm (TSMC) |
| Function | Rendering + AI + Video three-in-one |
| FP32 Compute | Not disclosed (estimated ~20 TFLOPS) |
| FP16/INT8 Compute | Not disclosed (estimated ~200 TFLOPS FP16 / ~400 TOPS INT8) |
| Ray Tracing | Supported (hardware ray tracing) |
| Video Codec | Powerful video processing capabilities (specific channels not disclosed) |
| TDP | Not disclosed (estimated 200-300W) |
| Form Factor | NanYu series (VG1600/VG1800/VG14) |
| Release | 2023-07-06 (2023 World AI Conference) |
| Mass Production | 2023-Q4 |
| API Support | DirectX 11, OpenGL, Vulkan (Windows/Linux) |
⚠️ Specification Note: Specific compute values of SG100 are not fully disclosed by official sources. Some values above are estimates based on public reports, subject to Vastai Tech's subsequent official data sheet.
NanYu Series GPU Accelerator Cards
NanYu series accelerator cards released synchronously with SG100:
| Model | Positioning | Scenario |
|---|
| VG1600 | Cloud gaming | Immersive cloud gaming experience |
| VG1800 | Cloud desktop/remote office | Smooth office software, education APP, industrial design software |
| VG14 | Domestic workstation | DirectX, OpenGL and other API support under Windows system |
Vastai Product Matrix
| Product | Release | Positioning | Status |
|---|
| SV100 (ZaiTian VA10) | 2022-09 | AI inference accelerator | ✅ Mass production |
| SG100 (QianYuan) | 2023-07 | Full-function GPU | ✅ Released |
| VA1L | 2023-07 | LLM large model AI accelerator | ✅ Released |
| VA12 | 2023-07 | High-performance generative AI accelerator | ✅ Released |
Comparison with ZaiTian VA10
| Metric | ZaiTian VA10 (SV100) | QianYuan SG100 |
|---|
| Positioning | AI inference acceleration | Full-function GPU |
| Rendering | ❌ Not supported | ✅ Supported (hardware ray tracing) |
| AI Compute | INT8 400 TOPS | Estimated ~200 TFLOPS FP16 |
| Video Processing | ✅ Supported | ✅ Powerful video processing |
| API Support | VastStream (AI) | DirectX/OpenGL/Vulkan |
| Item | Content |
|---|
| Company | Vastai Tech |
| Founded | 2018-12 |
| Headquarters | Shanghai, China |
| Founder | Qian Jun (former AMD executive) |
| Product Page | https://www.vastaitech.com/product/general/va10 |
| Software Stack | VastStream (CUDA-like API) |
| Status | Mass production, second-generation SG100 released |
Application Scenarios
- ✅ Cloud gaming (VG1600, rendering+AI+video)
- ✅ Cloud desktop/remote office (VG1800)
- ✅ Digital twin (leading rendering performance)
- ✅ AI inference (low latency high throughput)
- ✅ Video processing (powerful video encoding/decoding capabilities)
- ❌ AI training (not training-specialized)
- ❌ CUDA ecosystem strong dependency (needs migration to VastStream) |
References