GSC6204 6U OpenVPX GPGPU Coprocessing Board
Mercury Systems
The GSC6204 OpenVPX GPGPU Coprocessing Board off-loads compute-intense functions to GPU coprocessors, providing scalable processing power for complex workloads in commercial and defense applications.
Typed specifications23 values
Discloses 7 of 13 fields engineers filter by in this category (category median 5).
- Processor
- 2x embedded NVIDIA Quadro TU104 GPU with the Turing architecture datasheet p.3
- RAM
- 16 384 MB 16 GB GDDR6 per processor · datasheet p.3 32 768 MB 32 GB total GDDR6 memory
- Operating system
- Linux Linux development and run-time environment · datasheet p.2
- Power consumption
- 170 W typically per accelerator, power is configurable · datasheet p.4 170 W per accelerator
- Data interface
- PCIe Gen 3 datasheet p.1 other 2x analog VGA display outputs (1 per processor) | other 4x DisplayPort display outputs (2 per GPU) | other NVIDIA NVlink high-speed bus between GPUs (50 GB/s) | I2C standard I2C bus for IPMC / IPMI system management +3
- Processor architecture
- GPU / NPU NVIDIA Quadro TU104 GPU, Turing architecture · datasheet p.3
- Board format
- 6U VPX 6U OpenVPX (ANSI/VITA 65.0) and VPX-REDI, 1.0" pitch single-slot · datasheet p.4
- Flight software support
- OpenCL, NVIDIA CUDA, OFED, OpenMPI datasheet p.2
- Cores
- 384 384 Tensor cores · datasheet p.3 48 48 Ray Tracing cores | 6 144 6144 total processing cores (3072 lanes per GPU)
- AI performance
- 30 TOPS 384 Tensor Cores, mixed-precision matrix-multiply and accumul… · datasheet p.1
- RAM type
- other GDDR6 · datasheet p.3
- ram_gb
- 32 datasheet p.3
Each value is typed with its unit and measurement condition and cites the datasheet page it was taken from. Nothing is guessed: a field that is absent here has not been published by the manufacturer. machine-extracted ✓ checked by NewSpace Market ✓ manufacturer confirmed by the manufacturer.
As published by the supplier7 lines, in the manufacturer's own wording
- GPU
- NVIDIA® Quadro® TU104
- CUDA Cores
- 6144
- Memory
- 32 GB GDDR6
- Interconnects
- PCIe Gen3 and NVIDIA NVLink™
- Software Support
- NVIDIA, CUDA®, OpenCL™
- Environmental Protection
- MOTS+ extreme
- Compliance
- ANSI/VITA 65.0
About
The GSC6204 modules are powered by NVIDIA® Quadro® TU104 GPUs, delivering the floating-point math compute performance required for efficient off-loading of compute-intensive applications. These include radar, EO/IR imagery, AI, urban air mobility, autonomous driving, cognitive EW, and sensor fusion, which generate huge volumes of sensor and network data requiring powerful processing resources to extract actionable information in real time. Available with MOTS+ extreme environmental protection, GSC6204 OpenVPX coprocessor modules are ruggedized for applications operating in harsh environments. They incorporate the NVIDIA Turing™ GPU architecture, bringing the latest commercial processing advancements and scalability to the embedded domain. Powered by dual Quadro TU104 processors and incorporating NVIDIA’s NVLink™ high-speed direct GPU-to-GPU interconnect technology and a PCIe architecture, these modules embed massive parallel processing capability into edge applications. Each GPU features NVIDIA’s Tensor Cores for mixed-precision matrix multiply and accumulate calculations, key for AI, deep learning, and signal processing workloads.
Documentation
Need the full ICD, test reports or a specific revision? Ask the supplier directly.