Deep Learning, Artificial Intelligence, Rugged, Conduction-Cooled, Fully Sealed GPGPU Co-Processor
SYSTEM HIGHLIGHTS
- PCIe-over-cable local bus extension for inter-/intra chassis expansion for two full-size PCIe co-processors
- Supports up to two full size commercially-available General-Purpose Computing on Graphics Processing Units (GPGPUs)
- Accommodates specialized FPGA, ASIC, co-processor cards
- Conduction-cooled rugged enclosure provides 600 W of power
- FlexIO™ backplane provides internal/external PCIe Gen 3 at 8 Gbps
- Independent GPGPU outputs (interface is GPU/card specific)
- GPGPU modules support a total of 32 PCIe Gen 3 lanes, 16 lanes of input and 16 lanes of output
- X422 system daisy chains to multiple Lightning GPGPU co-processor units
- Software support for NVIDIA™ CUDA proprietary framework and OpenCL open framework
I/O AND EXPANSION OPTIONS
- PCIe-over-cable bus extension for co-processor cards
- Up to two full size GPGPUs (dual x16 PCIe 3.0 slots)
- 32 lanes PCIe Gen 3, dual x16
- Two independent card-specific fabric or I/O ports (one per PCI slot). Example: Titan V provides 2x DisplayPort and 1x HD Video (A), and 3x DisplayPort and 1x HD Video (B)
AVAILABLE NVIDIA® GPGPUs
The X422 supports the following NVIDIA® GPGPU cards.
(Model / Architecture) | NVIDIA CUDA® Cores | Single Precision GFLOPS | TDP (W) |
Jetson TX2 / Pascal | 256 | 1500 | 8 |
GTX 1060 / Pascal | 1280 | 3594 | 80 |
GTX 1080 / Pascal | 2560 | 7967 | 150 |
P4 PCIe – Tesla / Pascal | 2560 | 4800 | 67 |
P100 PCIe – Tesla / Pascal | 3584 | 10000 | 250 |
TITAN Xp / Pascal | 3840 | 10790 | 250 |
V100 PCIe – Tesla / Volta | 5120 | 14026 | 250 |
TITAN V - Volta | 5120 | 12288 | 250 |
TITAN V – Volta [Tensor] | 5120 | 110592 | 250 |
ENVIRONMENTAL SPECS
Size: 12.45" x 11.6" x 2.54" (including fins)
Power: 300 W (one card) to 600 W (two cards)
MIL-STD: MIL-STD-810G, MIL-STD-1275D, MIL-S-901D, MIL-STD-461F, DO-160D, IP64
Temperature: Operates up to extended temp -20° C to +55° C (Optional)
Ruggedness: Available in ruggedization levels R1-R2
SYSTEM OVERVIEW
The X422 is a GPGPU Co-Processor System that supports GMS’ companion S422 SW Server. Up to two full size GPGPU cards can be installed, either as two independent processing units or they can be intelligently linked together to form one virtual processing unit. In addition to the GPGPU capabilities, “Lightning” offers bus expansion capabilities to virtually any full size, full height PCIe 3.0 card, but is optimized to harness multiple GPGPU systems. The X422 affords the use of commercially available, full size, GPGPU products in a completely sealed and protected, extended temperature fully ruggedized enclosure that utilizes GMS’ patented RuggedCool™ technology.
Owing to the underlying GPGPU architecture with multiple cores and threads, the X422 can accelerate large data processing tasks such as image recognition, digital signal processing, data mining, block chain computation, artificial intelligence, machine vision, image processing, vector processing, and other compute-intensive tasks. PCIe 3.0 bus expansion allows upward-scaling by cascading (daisy chaining) additional GPGPUs as as necessary via additional X422 chassis.
The X422 utilizes GMS’s FlexIO™ flow-through architecture which is based upon wide PCIe 3.0 lanes (x16) operating at 8 Gbps. When connected via x16 PCIe 3.0 to the companion S422 dual Xeon® conduction-cooled rugged server, the GPGPUs (or any PCIe cards) appear to be within the same “bus” of the main server. This closely-coupled architecture allows for rapid data passing, RDMA and “atomic” operations, or provides a fully autonomous GPGPU co-processor where only data is shared asymmetrically from the main CPUs.
The onboard intelligent PCIe 3.0 switch provides packet processing and local “routing”, allowing the user to customize a homogeneous or heterogeneous architecture between the local GPGPU resources. For example, the X422 can be partitioned for GPGPU A to operate separately and independently from GPGPU B—or they can pass data between each other via the dual x16 PCIe 3.0 fabric. Separate, GPGPU-specific, and customizable I/O “pipes” are available to the front panel for each PCIe slot. This allows sensor-to-GPGPU processing, down- or upstream processing, or an additional data path that bypasses the companion S422 dual-Xeon® server.
APPLICATIONS
The X422 is a high performance GPGPU co-processor and bus expansion system designed for use with the S422 SW Server.
The X422 is an ideal forwardly-deployed, vehicle-mounted, high performance GPGPU co-processing system. Applications include computing clusters and parallel computing, digital signal processing, digital image processing, video processing, neural networks, data mining, cryptography, and intrusion detection.
The X422 system is fully compliant to MIL-STD-810G, MIL-STD-1275D, MIL-S-901D, DO-160D, MIL-STD-461E and has ingress protection up to IP64.