Start Page

NEOX® | GA100

Industry’s first RISC-V GPGPU for 3D graphics and AI at the edge


NEOX® | GA100 is the leading GPGPU solution for the MCU market and industry's first GPGPU architecture based on RISC-V, offering the unique combination of performing both graphics and ΑΙ on the same IP. Ultra-low power, tiny silicon area & memory requirements are key benefits that position NEOX ® | GA100 as the ultimate choice for battery powered and cost-constrained wearable and edge AIoT devices.

Additional information

Download Product Brief
Unparallered RISC-V flexibility & programmability

NEOX® | GA100 is the world’s first RISC-V GPGPU, having full entitlement of the RISC-V RV64 architecture and enabling greater flexibility and programmability for Graphics, AI & compute. Offering the unique opportunity to leverage the open ecosystem of tools (profilers/debuggers etc.) and mature compiler infrastructure (LLVM), it is a highly configurable, turnkey accelerator equipped with custom graphics and AI instructions.

Compelling Ultra-Low Power 3D for wearables & AIOT

NEOX® | GA100 multithreaded nature enables 3D graphics rendering on a variety of target systems, from low-power bare-metal systems to high-performance, Linux based systems. It introduces new hardware and a new set of software extensions to accelerate 3D graphics. 3D support is introduced through a royalty free file format called glTFTM developed by the Khronos® that enables the efficient transmission and loading of 3D models by engines and applications. Once the 3D object has been imported, effects like rotation, lighting, animation can be applied before rendering it to the screen. The entire 3D, VG & 2.5D API requires less than 150KB of memory and 10mW of power.

Enabling your AI Application without additional silicon cost

NEOX® | GA100 is delivered with an AI SDK that can perform offline graph optimization and AI inference based on TensorFlow Lite for MCU. It includes various open source, proprietary tools, and algorithms for analyzing, visualizing, converting, compressing, and deploying Deep Neural Networks (DNNs) on NEOX® architecture. It allows to perform various iterative steps in model compression and model analysis, until the desired balance between “accuracy-performance-memory” is achieved. NEOX® | AI SDK is also integrated into Edge Impulse platform, enabling ML developers to easily target NEOX® | GA100 for edge applications on wearable and AIoT devices.


NEOX® | GA100 is addressing diverse vertical markets including next generation ultra-low power devices, such as wearables (smartwatches, AI smart glasses, hearables), smart home (appliances, thermostats, speakers, cameras), mobility (e-bikes, e-scooters) and industrial HMI/IoT.

Example graphics applications for NEOX® | GA100 include rendering 3D / VG / 2.5D watch faces, VG or raster fonts, VG maps, notifications and alerts, 3D avatars for video calls, and more. Example AI applications include human activity recognition, health monitoring and diagnostics, keyword spotting, audio and speech enhancement, among others.

3D, Vector & 2.5D Graphics, AI
RV64IMFC + custom GFX & AI extensions
Vertex / Shader Processing
Unified Fully Programmable LLVM C/C++ RISCV
Max Core Clock Frequency (MHz)
Characterization Configuration
32 threads; AXI4:64 or AXI4:128
16.7Million (24-bit)
Texture Read Performance @100MHz
100 Mpixel/sec
Frame Buffer Write @100MHz
100 Mpixel/sec
Power Consumption
<10mW (workload dependent)
Typical watch face performance @200 MHz (454x454, 32bpp)
2D >60fps (Vsync bound), VG >60fps (Vsync bound), 3D >30fps
AI Performance
Performance reported by NEOX® | AI SDK available upon request
Typical CPU utilization
Up to 4K
Small to mid-sized displays (1,5" - 10")
Wearables: smartwatches/smart bands/ AI smart glasses/hearables
Mobility: E-bikes & E-scooters
Smart Home: smart appliances/thermostats/speakers
Industrial: PoS/Pol terminals


The software tools for graphics include

- NEOX® | pix-presso, a utility software for converting images to/from Think Silicon's proprietary lossy compression formats (TSCTM4, TSCTM6, TSCTM12) suitable for low power embedded devices.

- An offline utility, embedded inside NEOX® | Pixpresso, which converts files from Scalable Vector Graphics (SVG) format into a very compact internal custom binary format (TSVG). The TSVG file format can be executed very efficiently with a single call of the NEMA®|vg API.

- An offline utility that converts the vector font file (TTF format) into a very compact custom binary file that can be executed very efficiently with a single prin function call of the NEMA®|vg API. With this utility the developer can choose the suitable font range, e.g., Greek, Latin, Asian, etc. The key reasons for choosing multiple font resolutions without impacting the quality and enabling more efficient storage, especially for Asian fonts.

The software tools for AI include

- NEOX® | AI SDK, a collection of open source and proprietary tools for converting, optimising, analyzing and deploying pre-trained and post-trained DNN models on the NEOX® architecture. The SDK delivers the desired balance between "accuracy-performance-memory".

- The user can run the TensorFlow Lite model using the NEOX® | HW simulation platform. After the test execution, the user gathers insightful information, such as number of simulated instructions, memory transactions, or cache statistics. The AI-SDK package is delivered in a docker image that contains all the above tools as executable files.



  • Fully Programmable Alpha blending
  • Source/Destination color keying
  • Porter-Duff Blending


  • Texture mapping
  • Point sampling
  • Bilinear filtering
  • Blit support
  • Rotation any angle
  • Mirroring
  • Stretch (independently on x and y axis)
  • Format conversions on the fly
  • 3D Perspective Correct Projections


  • Clamp
  • Border (single color)
  • Repeat
  • Mirro


  • 3x3 matrix transformations


  • Bitmap antialiased (A1 / A2 / A4 / A8 )
  • Vector fonts - offline conversion from TTF
  • Font Kerning, Unicode (UTF8)
  • Text Alignment:


  • 32-bit RGBA8888 / BGRA8888 / ABGR8888
  • 24-bit RGB
  • 16-bit RGBA5551 / RGB565
  • 8-bit Α8 / L8 / RGB332
  • 4-bit A4 / L4
  • 2-bit A2 / L2
  • 1-bit A1 / L1
  • TSCTM (Optional) (Read Only)
  • YUV Read Only


  • TSCTM4 (4 bits per pixel)
  • TSCTM6 / TSCTM6a (6 bits per pixel w/o alpha)
  • TSCTM12 / TSCTM12a (12 bits per pixel w/o alpha)


  • 8x MSAA
  • 3D Anti-aliasing
  • Quadrilaterals / Triangles /
    Thick lines / Cubic Bezier /
    Quadratic Bezier per edge

Software development kit for offline graph optimization and AI inference

Copyright © 2021 - 2024 Think Silicon | ULTRA-LOW POWER GPUs