The rugged VPX3-4924 NVIDIA Tesla Pascal GPU processor is part of a family of GPGPU modules available from Curtiss-Wright Defense Solutions to enable the development of High-Performance Embedded Computing (HPEC) systems.
- NVIDIA Tesla Pascal 16 nm GPU
- 2048 NVIDIA CUDA cores
- 6.2 TFLOPS
- 67 GFLOPS/watt
- 16 GB GDDR5 with NVIDIA GPUDirect DMA technology
- Max memory bandwidth: 192 GB/s
- High-Performance Compute Mode with auto-switching
- In HPC mode: full 16 GB BAR
- NVIDIA GRID vGPU virtualization
- PCIe x16 Gen
- ISR and EW applications where TFLOPs of accelerated processing is required
- Massive data ingest of modern Radar, SIGINT, EO/IR sensors
- Unparalleled HPEC performance in cross-cueing applications
VPX3-4924 3U VPX GPGPU Processor Card with NVIDIA Tesla Pascal P6
The rugged VPX3-4924 NVIDIA Tesla Pascal GPU processor board is designed and manufactured by Wolf Advanced Technology. This board is part of a family of GPGPU modules available from Curtiss-Wright Defense Solutions to enable development of High-Performance Embedded Computing (HPEC) systems.
Leveraging the NVIDIA Tesla Pascal 16 nm GPU technology to bring extreme high performance, the VPX3-4924 brings GPUDirect DMA and GRID virtualization technology to ruggedized embedded platforms. NVIDIA GRID is the industry’s most advanced technology for sharing true virtual GPU (GRID 2.0 vGPU) hardware acceleration between multiple users; max of 16 instances. This technology ensures complete application compatibility, which means features and experiences are the same as they would be on a physical device.
The VPX3-4924 is especially suited for use as an advanced GPGPU compute engine because it utilizes a large 16GB BAR for direct access from the CPU and other PCI devices. Tesla Pascal GPU architecture also provides a more powerful Unified Memory feature. Pascal’s larger virtual memory address space enables GPUs to access the entire system memory plus the memory of all GPUs in the system. The on-demand page migration engine allows the system to migrate pages from anywhere in the system to the GPU’s memory for efficient processing. This improved memory handling results in significantly improved algorithm efficiency.