The VPX3-XMXM is a GPGPU blade in 3U Eurocard format: it brings the graphics and computing power of an embedded GPU inside a VPX backplane, with the stated goal of AI inference in environments where there is no air conditioning and there is vibration.
What it carries
The graphics is NVIDIA RTX 5000 Ada Generation or RTX 2000 Ada Generation in embedded version, with a stated roadmap toward the Blackwell architecture. The top-tier configuration reaches 9728 CUDA core. The interface toward the system is native PCIe Gen4, and the lanes are configurable via the onboard PCIe switch: not a fixed number printed on the connector.
How it plugs into the backplane
The board declares three OpenVPX profiles, and the choice of profile changes what reaches the connectors:
- SLT3-PAY-2F2U-14.2.3, with the PCIe upstream port on P1A;
- SLT3-PAY-1F1U1S1S1U1U4F1J-14.6.13-0, with P2B depopulated;
- SLT3-PAY-1F1U1S1S1U1U2F1H-14.6.11-0, with P2 fully depopulated.
On the data plane there is a PCIe Gen4 x8 toward P1A and, as an SKU option, a second x8 toward P2A; each is configurable into two x4 on request, and the two x8 on P1B and P2A can be joined into a single x16. This is the part to settle with the technical office before ordering, because it depends on the backplane you already have.
Environment, power, management
Conduction-cooled cooling, 12 V power supply on VS1/VS2, operating temperature from -40 to 85 °C and storage from -55 to 100 °C. Management covers the mandatory command list IPMI V1.5 and that of VITA 46.11 Tier 1 and Tier 2: the board is diagnosed from the chassis manager, without opening the rack.
Standards and surrounding features
Alignment is with SOSA, on VPX (VITA 46), VITA 48 REDI and OpenVPX VITA 65. As an SKU option, a DisplayPort 1.4a and the expansion ports from the SBC come out: 2× USB 3.2 Gen1x1, 1× USB 2.0 and 1× RS-232, plus the RS-232 maintenance port. Supported operating systems are Linux and Windows. Declared markings: CE, FCC, EN 55032, EN 55035.
When it makes sense
It makes sense when the computing load is real-time vision or inference and the constraint is the slot, not the watt: a conduction-cooled 3U VPX integrates into a qualified microrack, where a desktop GPU does not fit and does not pass environmental tests. If the project is at the specification stage, write to us with the OpenVPX profile and the lane configuration you need.
