Jingjia Micro JM9 Series
Product Overview
Changsha Jingjia Microelectronics (Jingjia Micro, SZSE 300474), founded in 2006, was the first company in China to successfully develop a domestic GPU chip and achieve large-scale engineering application. The company started in military graphics display and control, releasing the JM5400 (2014, 65nm, used to replace AMD/ATI solutions in domestic military aircraft) and the JM7200 (2018, 28nm, performance roughly equal to the NVIDIA GT 640 but at under 10W) — two generations of GPUs — before entering the civilian Xinchuang market.
The JM9 series is Jingjia Micro's third-generation GPU and the core achievement of the company's 2018 private-placement funded project, the "High-Performance General-Purpose Graphics Processor R&D and Industrialization Project". On 2021-11-16, Jingjia Micro officially announced that the JM9 series graphics processing chips had completed tape-out, packaging, and initial testing, with key technical specifications meeting design requirements. In May 2022, the second JM9 series graphics processing chip completed development, targeting desktops, servers, industrial PCs, self-service terminals, and similar equipment.
The publicly named models of the JM9 series are the JM9231 and JM9271: the former targets domestic office PCs, portable computers, mid-to-low-end entertainment devices, and high-end embedded systems; the latter strengthens scientific computing capabilities, aimed at artificial intelligence, security surveillance, speech recognition, cloud computing, and similar directions. The series is widely used in Xinchuang domestic office replacement projects across regions, as well as in professional surveying and mapping, geological information processing, industrial computer-aided design, and other scenarios.
As of March 2026, Jingjia Micro confirmed on its investor interaction platform that the JM9 series has achieved volume delivery; meanwhile, the company's fourth-generation GPU, the JM11 series, has launched and entered small-volume shipping, with the JM9 becoming one of the in-service mainstay products.
⚠️ Important Note: Two Circulating Spec Sets Exist Side by Side The widely circulated table of JM9231/JM9271 "2 TFLOPS / 8 TFLOPS FP32, 256/512 GB/s, 150W/200W, comparable to GTX 1050 / GTX 1080" on the internet and in research reports originally comes from 2019 media reports (MyDrivers, Tom's Hardware) and securities research reports, and is not an official specification released by Jingjia Micro. Jingjia Micro representatives have publicly stated that the JM9 series and the GTX 1080 "differ in market position, application characteristics, and focus, and are not comparable". The "Core Specifications" table on this page follows the official figures disclosed in Jingjia Micro's 2021-11-16 formal SZSE announcement; the rumored specs are listed separately in the comparison table below and clearly flagged.
Core Specifications
The following are the measured figures for the JM9 series graphics processing chip (first model) disclosed in Jingjia Micro's official 2021-11-16 announcement.
| Parameter | Value |
|---|---|
| Architecture | JM9 self-developed GPU architecture (unified rendering architecture) |
| Process Node | Not disclosed (media reports say 14nm; unconfirmed officially) |
| FP16 / BF16 Compute | Not disclosed |
| INT8 Compute | Not disclosed |
| FP32 Compute | 1.5 TFLOPS (32-bit single-precision floating point, official announcement) |
| Pixel Fill Rate | 32 GPixels/s |
| Core Clock | 1.5 GHz (dynamic frequency scaling supported) |
| Memory Capacity | 8 GB |
| Memory Type | Not disclosed (type not specified in the official announcement) |
| Memory Bandwidth | 128 GB/s |
| TDP | < 30 W (official announcement) |
| Interconnect | No multi-card interconnect (single graphics/compute card) |
| Interface | PCIe 4.0 x8 |
| Graphics APIs | OpenGL 4.0, OpenCL 3.0, Vulkan 1.1, OpenGL ES 3.2, DirectFB 1.7.7, OpenVG 1.1 |
| Video Decoding | 4 channels of 4K@60fps (H.265 / H.264 / MPEG2 / MPEG4 hardware decode) |
| Video Encoding | 1 channel of 4K@60fps (H.265 / H.264 / MPEG2 / MPEG4 hardware encode) |
| Display Outputs | 4 independent display controllers: 4×HDMI 2.0 + 2×eDP + 1×VGA |
| Platform Support | x86 / ARM / MIPS + Linux, NeoKylin, Galaxy Kylin, UOS |
| Launch | 2021-11 (announcement of tape-out, packaging, and initial testing completion) |
| Mass Production/Availability | Second chip completed development in May 2022; volume delivery officially confirmed in March 2026 |
Rumored Spec Comparison (Unofficial, For Reference Only)
| Parameter | JM9231 (rumored) | JM9271 (rumored) |
|---|---|---|
| Graphics APIs | OpenGL 4.5 / OpenCL 1.2 | OpenGL 4.5 / OpenCL 2.0 |
| Boost Clock | > 1,500 MHz | > 1,800 MHz |
| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |
| Memory | 8GB GDDR5 | 16GB HBM |
| Memory Bandwidth | 256 GB/s | 512 GB/s |
| FP32 | 2 TFLOPS | 8 TFLOPS |
| Pixel Fill Rate | > 32 GPixel/s | > 128 GPixel/s |
| TDP | 150 W | 200 W |
| Rumored Competitor | GTX 1050 | GTX 1080 |
⚠️ The table above reflects the figures circulated by media and securities research reports in 2019—2021, never officially confirmed by Jingjia Micro, and conflicts significantly with the official 2021 announcement figures (1.5 TFLOPS FP32 / 128 GB/s / PCIe 4.0 x8 / <30W). Possible explanations include: the official announcement disclosed a low-power mass-production version while the rumored table described early design targets; or the two correspond to different models within the series. When citing, be sure to distinguish official figures from rumored ones.
Key Features
- Fully self-developed GPU architecture: mastery of full-chain key technologies including chip-level logic/physical design, analog interface design, and GPU driver development — not a modified IP-licensed design
- Full-stack Xinchuang support: officially supports x86 / ARM / MIPS processors, as well as domestic operating systems such as NeoKylin, Galaxy Kylin, and UOS
- Extremely low power: official figure < 30W, supporting fanless/low-cooling designs, suitable for government office terminals and embedded devices
- Strong multi-display output: 4 independent graphics display controllers, supporting 4×HDMI 2.0 + 2×eDP + 1×VGA
- Complete video codec support: 4 channels of 4K@60 hardware decode + 1 channel of 4K@60 hardware encode
- Military-grade reliability heritage: carries forward the high-reliability design experience of the JM5400/JM7200 in demanding aerospace environments
- GIS/CAD scenario strengths: deeply adapted with domestic industrial software such as SuperMap, forming a "CPU+GPU+OS+GIS" self-controlled closed loop
- Limitations: AI compute (FP16/INT8) never officially disclosed, unsuitable as an AI training or large-model inference card; no single-card interconnect capability
Vendor Information
| Parameter | Value |
|---|---|
| Company | Changsha Jingjia Microelectronics Co., Ltd. (Changsha Jingjia Microelectronics Co., Ltd.) |
| Headquarters | Changsha, Hunan Province |
| Founded | 2006 |
| Listing | Listed on the SZSE ChiNext in March 2016, ticker 300474 |
| Business | Integrated circuits (GPUs), signal processing, compute and storage, broadband ad-hoc networking, small radar, electromagnetic spectrum |
| 2025 Revenue | RMB 720 million |
| 2025 R&D Spend | RMB 428 million (59.39% of revenue) |
| R&D Team | 1,060 people (68% of employees, end of 2025) |
| Granted Patents | 165 (as of end of 2025) |
| GPU Product Line | JM5 series → JM7 series → JM9 series → JM11 series |
| Official Website | https://jingjiamicro.com |
Use Cases
- ✅ Xinchuang domestic office PC / portable computer graphics replacement
- ✅ Desktop terminal domestic substitution in government, finance, transportation, and telecom industries
- ✅ Geographic information systems (GIS), 3D real-scene map rendering
- ✅ CAD-assisted design, industrial drafting, industrial digital twins
- ✅ High-end embedded systems, industrial PCs, self-service terminals
- ✅ Multimedia processing and 4K video encode/decode
- ✅ Military/special display and control equipment (the company's traditional strength)
- ❌ Large-model AI training (no public FP16/BF16 compute, no high-speed interconnect)
- ❌ Large-model inference clusters (8GB memory and 128 GB/s bandwidth insufficient)
- ❌ High-end AAA gaming (graphics API support level and ecosystem limited)
- ❌ CUDA ecosystem workloads (requires re-porting)
Related Cards
- Jingjia Micro JM9271 — High-end model in the same series, oriented toward AI/cloud computing
- Moore Threads MTT S5000 — Fellow domestic full-function GPU, one of the "Five Tigers of Domestic GPUs"
- Innosilicon Fenghua 1 — Contemporary domestic GPU with data center/cloud office positioning
- Innosilicon Fenghua 2 — Contemporary domestic low-power desktop GPU with positioning highly overlapping the JM9
- MetaX C600 — Domestic GPU high-compute route comparison
References
- Changsha Jingjia Microelectronics announcement on the R&D progress of its JM9 series graphics processing chips (2021-11-16, SZSE)
- Jingjia Micro official website — Integrated circuits (JM5/JM7/JM9 series)
- Jingjia Micro: commercial GPUs urgently need breakthroughs amid the AI boom (Sina Finance, 2023-04-04)
- Jingjia Micro Tapes Out JM9-Series GPUs, Aiming for GTX 1080 Performance (Tom's Hardware, 2021-09-15, source of the rumored specs)