We reached yesterday a milestone in the development program of the MD series, "MiniDedicated" hardware platform.

The "node plate" side hardware component selection is now complete and has moved into final qualification phase. We expect only minor adjustments and fine-tuning to the platform in a iterative fashion as we scale up production.

In this most power efficient, highest density platform, we are now ready start software development to bring automation to you. Main software component selection has been already chosen, development required on interfacing/interoperability, documentation/structural and qualification.

This will be leading efficiency dedicated server platform by far, we do not expect there are many coming close to these level of efficiencies from start to finish. From acquiring the nodes all the way when Your Bits and Bytes leave the datacenter. Efficiency is a key factor each step of the way.

"Plate" Key Features in a Nutshell:
  • 1RU Form factor
  • Node scalability: Up to 16 nodes per 1RU, 672 nodes per 42U Rack
  • Power scalability: Up to 1000W per 1RU, 42KW Per Rack. PSU Options: 350W, 500W, 700W (Redundant), 1000W (Redudant)
  • Industrial Mean Well PSUs with High Power Efficiency of up to 94%
  • Built-in managed network switching, 10G Uplink per 8x1Gbit nodes (Upgradeable once new managed small form factor switches come available)
  • Power management per individual node, 0.5W standby power consumption, max 3.84W
  • Designed for "consumer grade" hardware without traditional BMC/IPMI features.
  • Supports efficient cooling for standard mini-ITX platforms, height can be increased conveniently for high power draw CPUs with big coolers (No definite limit on height)
  • Current version supports standard 100x100 VESA Mount and Mini-ITX mounting, future version will incorporate on as needed basis 4x4"/NUC, SBC mounts etc.
  • Designed for rack and datacenter level super efficient cooling, bringing cooling costs significantly down. BIG Fans, BIG Efficiency.
Rack level cooling still needs some work, but all main components are in stock or in freight, we expect zero hiccups on that since it is "just temperature controlled fans". Current design plan iteration consists of 24-56x 120mm industrial fans per rack, but later iteration will move to using larger than 120mm fans depending upon availability, characteristics etc.

Our current typical setup will utilize 8x35W nodes for maximum power draw of 280W on nodes, 20W on the managed switch, 4W on power management for total on-plate power draw maximum of 304W. Typical expected ~200W mark depending on load based on real world, in-production testing so far. Latest generation "35W" setups has been seen to consume up-to 90W in reality, typical max in the 60W ballpark.

Top Of The Rack (TOR) switch selected consumes only 400W giving 48x10G ports for the nodes itself and 2x100G Uplinks, giving each plate of 8 nodes average dedicated bandwidth of 4.17Gbps, or after error correction dedicated bandwidth of roughly 500Mbps per node! That also is industry leading, typical contention ratios in the industry are many times higher than 2:1 we have chosen to pursue.

This will make typical current generation full rack power consumption, rack level cooling included, of ONLY ~9100W for 336 NODES! 27.08W/Node.

ETA for full scale production is still unknown, aspirationally we would get all of this done by end of the year. It depends largely on when the new datacenter build is finished.

Thursday, June 1, 2023

« Back