Embedded MXM modules are first to load Nvidia Turing GPUs
May 11, 2021 — by Eric Brown 886 viewsAdlink announced the first MXM 3.1 modules with Nvidia Turing GPUs, ranging from the Quadro T1000 to the RTX5000. Aimed at edge AI applications, the modules are only a fifth the size of a PCIe card.
Adlink announced three PCIe graphics expansion modules that adopt the compact, PCIe x16-enabled MXM 3.1 form-factor: the EGX-MXM-T1000 with Nvidia’s Quadro Embedded T1000, the EGX-MXM-RTX3000 with the Quadro Embedded RTX3000, and the EGX-MXM-RTX5000 with the Quadro Embedded RTX5000. The Nvidia collaboration represents the first time such high-level graphics have been available on MXM (Mobile PCI Express Module).


EGX-MXM-RTX3000, front and back
(click images to enlarge)
Adlink has been a major booster of MXM, offering compatible slots on systems such as its Coffee Lake based Matrix edge computers and ROScube-I robotics controller. While those systems ship with optional Nvidia Pascal architecture Quadro P1000 (47W) to P5000 (100W) graphics cards, they can now load the new modules with more powerful, Turing architecture Nvidia Quadro RTX and Nvidia Quadro T1000 graphics technology.
The modules are aimed at edge AI projects with significant size, weight, and power (SWaP) constraints, including medical imaging, industrial automation, biometric access control, autonomous mobile robots, transportation, and aerospace and defense systems. The products offer 5-year availability, which Adlink says is more than three times the product lifecycle of typical graphics cards.
— ADVERTISEMENT —
As with Adlink’s earlier Pascal Quadro equipped EGX-MXM modules, the new products enable 4x additional DP 1.4a ports with 4K at 120Hz or 8K at 60Hz. Linux and Windows drivers are available.


EGX-MXM-RTX5000 (left) and EGX-MXM-T1000
(click images to enlarge)
The entry-level EGX-MXM-T1000 has a Quadro T1000 with 896 CUDA cores and 2.6 TFLOPS peak FP32 performance. The module uses an 82 x 70mm MXM 3.1 Type A form-factor and is equipped with 4GB of 128-bit GDDR6 with 192 GB/s bandwidth. The 0 to 55°C tolerant module runs at a low of 50W.
The EGX-MXM-RTX3000 is equipped with a Quadro RTX3000 with 1,920 CUDA cores, 240 Tensor cores, and 5.3 TFLOPS Peak FP32. The module adopts the 105 x 82mm MXM 3.1 Type B form-factor and integrates 6GB of 192-bit GDDR6 with 336 GB/s bandwidth. The 0 to 55°C tolerant module runs at 80W.
The top-of-the-line EGX-MXM-RTX5000 is equipped with a Quadro RTX5000 with 3,072 CUDA cores and 384 Tensor cores, and offers 9.4 TFLOPS Peak FP32 performance. The module uses the MXM 3.1 Type B+ variant at 110 x 82mm, which is big enough to fit 16GB of 256-bit GDDR6 with 448 GB/s bandwidth. The 0 to 45°C tolerant module can run at 110W.
Adlink says that the modules can be used by its own customers with MXM equipped systems. It is unclear if they would work on other systems with MXM 3.1 connectors such as Cincoze’s Coffee Lake based GM-1000 system or Ibase’s Coffee Lake powered MT800M-P board.
Further information
No pricing or availability information was provided for the EGX-MXM-T1000, EGX-MXM-RTX3000, or EGX-MXM-RTX5000. More information may be found in Adlink’s announcement and MXM page.
Please comment here...