GPUs

Run CUDA workloads in space with Colossus’ GPU Platforms

AI/ML Compute Efficiency 

Utilize unmatched parallel computing for efficient AI workloads in space applications.

Autonomous Decision-Making

Onboard GPUs enable satellites to make immediate, mission-critical decisions without ground uplink delays.

Onboard Data Processing

Filter and compress onboard data to optimize bandwidth by transmitting analyzed products instead of raw data.

Meet Our GPU Computing Platforms

Semi-customizable FPGA-based platforms built for the unique demands of space environments.

Falcon

Our most advanced GPU platform for demanding workloads.

  • NVIDIA Orin AGX

  • 248 TOPS

  • 45-60 W Power 

  • 600 g Weight 

Kestrel

Entry level general-purpose GPU platform for SmallSats.

  • NVIDIA Jetson TX2i

  • 1.3 TOPS

  • 10-15 W Power 

  • 402 g Weight

Space-Grade Hardware Standards

Colossus Computing systems meet strict space-grade hardware standards to ensure reliability and performance in satellites. We address issues related to extreme temperatures, radiation resistance, and mechanical strength for optimal operation in space. Customizable testing for thermal, vibration, and EMI/EMC is available, with a default setting of GSFC-STD-7000 for Falcon 9 rideshare.

  • Designed for 5-year low Earth orbit missions

  • Temperature rated for -40°C to +60°C operating; -55°C to +85°C storage

  • Multiple boot sectors for redundancy and software upgrades

  • Comprehensive telemetry and monitoring

Built for Customization

General computing platforms customizable to meet your mission needs.

Hardware Modifications

Both platforms include FPGAs, enabling semi-custom hardware configurations with optional add-on boards, adaptable for applications such as high-speed data transfer, onboard AI analysis, and specialized payload processing.

Software & OS Configurations: 

Choose an OS configuration to meet your power and performance needs, customized at the Linux layer to ensure seamless integration with your mission’s unique protocols and requirements.

Kestrel

Entry level general-purpose GPU platform for SmallSats

Achieved Flight Heritage

Standard microSD card and M.2 2280 SSD slot populated with validated, industrial-grade options.

Mass Storage

Gigabit ethernet is available for high-speed connectivity, as well as other common interfaces.

High-Speed Connectivity

Data is protected with Error Correction Codes (ECC) for RAM and the large capacity non-violate memory allows for the storage of multiple copies of key files.

Data Protection

Protection and telemetry circuits for routine monitoring, as well as fault detection and isolation.

Routine Monitoring

Subsystems protected with a resettable eFuse circuit to mitigate the effects of radiation.

Radiation Tolerance

Kestrel Tech Specs

Processing Capabilities

  • AI Performance 1.3 TOPS

  • GPU 256-Core Pascal

  • CPU Quad-Core A57

  • Dual-Core Denver 2

  • Video Codecs H.264, H.265, MPEG1/2/4

  • OS Linux-based (built w/Yocto)

Memory/Storage

  • RAM 8 GB of 128-bit DRAM with ECC

  • Storage 32 GB eMMC

  • microSD card default 64 GB

  • M.2 2280 SSD default 1 TB

  • EEPROM 512 kb

Connectivity

  • Ethernet 1000BASE-T

  • UART 3x RS-422, RS-485 compatible

  • CAN Two 1 Mbps

  • USB One USB3.0 SuperSpeed

  • Two USB2.0 Hi-Speed

  • PPS RS-422 hardware

  • GPIO 4 buffered GPIOs

  • Debug JTAG, Recovery USB, RS-232

Size, Weight, & Power

  • Size 1/3 U (96 x 89 x 30 mm)

  • Fits CubeSats and SmallSats

  • Mass 402 g

  • Power Idle: <1 W / Max Efficiency: 15 W / Max Performance: 25 W / Input Voltage 9-12 VDC

  • Protection eFuse input protection and monitoring

  • Telemetry Internal ADC for voltage/ temperature telemetry

Environmental

  • Heritage TRL-9 as of March 2024

  • Radiation up to 30 krad(Si)

  • Temperature -40 to +60 C Operating / -55 to +85 C Storage

  • Testing Thermal, Vibration, and EMI/EMC compliant to GSFC-STD-7000

Falcon

Our most advanced GPU platform for demanding workloads

Available Now

Filtering, prioritizing, and compressing data allows more efficient use of limited bandwidth by sending data products instead of raw data.

Efficiency

On-board processing enables satellites to make intelligent decisions without waiting hours for commands to be uplinked from the ground. Satellites can respond instantly to changing circumstances.

Autonomy

Data products can be small enough to allow transmission through satellite-to-satellite networks in seconds instead of waiting hours for a traditional ground station pass.

Speed

Falcon Tech Specs

Processing Capabilities

  • AI Performance: 248 TOPS

  • GPU: 2048-core Ampere @ 1.2 GHz, 64 Tensor Cores

  • CPU: 12-core A78AE @ 2.0 GHz

  • PVA: Programmable Vision Accelerator

  • FPGA: Up to 78k LUTs

  • OS Linux-based (built w/Yocto)

Memory/Storage

  • RAM: 64 GB of 256-bit DRAM with ECC

  • Storage: 64 GB eMMC

  • microSD card: 64 GB

  • Internal SSD: 1 TB

  • EEPROM: 512 kb

Connectivity

  • Ethernet: Two 1000BASE-T

  • UART: Four RS-422, RS-485 compatible

  • CAN: Two 1 Mbps

  • USB: Three USB3.2 Gen2

  • PPS: RS-422 hardware

  • GPIO: 4 buffered GPIOs

  • Debug: JTAG, Recovery USB, UART

  • Expanded Capabilities: Additional Gigabit Ethernet / SpaceWire / LVDS, Camera Link / PCIe, SpaceFibre, 40G Ethernet

Size, Weight, & Power

  • Optimized for SmallSats

  • 110 x 100 x 50 mm

  • 3U SpaceVPX planned

  • Mass: 600 g

  • Power Idle: <3 W / Max Efficiency: 45 W / Absolute Max: 90 W

  • Input Voltage: 24-36 VDC

  • Protection: eFuse input protection and monitoring

  • Telemetry Internal ADC for voltage/ temperature telemetry

Environmental

  • Radiation Minimum 5-year LEO mission

  • Temperature: -40 to +60 C Operating -55 to +85 C Storage

  • Testing: Thermal, Vibration, and EMI/EMC compliant to GSFC-STD-7000