GB10/DGX Spark Parts
Stacking and clustering cables for NVIDIA DGX Spark and other GB10 systems: ASUS Ascent GX10, Dell Pro Max GB10 and HP ZGX Nano. Choose 400G QSFP112 DACs for direct two-node stacking, 200G QSFP56 DACs for the ConnectX-7 ports, or QSFP-DD breakout cables to connect three or more nodes through a switch. Every cable is tested before it ships and covered by our lifetime warranty.
From the blog
One inference server needs only a 10GbE front end, and an Intel X710 covers it. A back-end RDMA fabric on...
Dell requires 1100W-or-higher supplies for any GPU in an R740xd, and a 1600W supply delivers only 800W on a 120V...
A GPU alone doesn't make a refurbished server an AI box. The seven parts an R740xd or DL380 Gen10 inference...
One inference server needs only a 10GbE front end, and an Intel X710 covers it. A back-end RDMA fabric on ConnectX matters only when a model or training job spans servers. Which card and cable fits each, and why 100G is the ceiling in a PCIe Gen3 server.
Dell requires 1100W-or-higher supplies for any GPU in an R740xd, and a 1600W supply delivers only 800W on a 120V outlet. How to size power supplies for one, two or three GPUs in a Dell R740xd or HPE DL380 Gen10, with worked examples and prices.
A GPU alone doesn't make a refurbished server an AI box. The seven parts an R740xd or DL380 Gen10 inference build needs, from the GPU enablement kit to power, memory, boot, storage and network, with part numbers and prices.
Testimonials
Got to me in 2 days, looks as professional as CDW.