Optimizing SWaP is a foundational element for any electronics designer. As consumers, all we have to do is look at the mobile devices in our hands to see the ongoing evolution of packing more power, faster speeds and greater functionality into smaller, lighter gadgets. However, taking that level of optimization from a mobile device up to that of a rugged, high-performance compute solution requires a whole other level of engineering ingenuity and expertise.
So, when a major aerospace & defense OEM needed the power and reliability of a rugged 4U server compressed into a system with the size requirements of a 2U, our team was ready to take on the challenge…or more accurately, challenges.
Break it down to build it up
It’s very common for a customer to need a customized server to accommodate specific operating parameters or application needs. That’s why Crystal Group specializes in creating custom solutions that meet each of those unique needs. But the combination of requirements our engineers needed to address for this particular situation called for more than a few modifications to a standard system:
- Cut the overall size by 50%
- Increase the power supply
- Incorporate two A100 GPUs
- Include dual processors
- Add two U.2 NVMe drives
Given the scope and complexity of converting the performance and capabilities of a 4U server into a 2U, they broke down and addressed the needs one by one to build the desired end product. Since the footprint was the primary limitation, they started with a standard 2U chassis. Knowing the thermal risks of packing the needed electronics within such a small space, the team used our patented cooling and anti-shock and vibration techniques, as well as creative fan and air plenum use to alleviate any heat-related risks.
Next up: Power. This posed a significant obstacle because the DC power supply needed was much larger than anything typically offered for a standard 2U form factor. Our team dug in and got creative. Applying the technology from our RCS7850 switch, they created a completely new power supply. Not only did this solve the increased power requirement, it also provided critical power redundancy.
Now that the size and power elements were in place, the engineers turned their attention to the interior design challenge. Traditionally, motherboards take up a lot of valuable space inside a 2U, which was not feasible for this project. They made more room in the chassis with a twin-sized, dual-processor motherboard, but unfortunately, full-height cards don’t fit vertically in a 2U system. So, special risers were designed to connect the GPUs and peripheral cards horizontally next to the motherboard.
The last major challenge with this design is adding the two U.2 NVMe cards to the system without blocking the airflow needed to cool the CPUs and GPUs. This meant the PC boards had to be custom designed to provide a rugged connection while also enabling the necessary airflow needs. Using computation, fluid-dynamic modeling, the design was validated before it was even built.
Capabilities without compromise
Challenges are often resolved through compromise or tradeoffs. The more challenges linked to a common objective, the more negotiating, often to the point of diluting the overall solution to “good enough.” However, settling for “good enough” today can increase costs, jeopardize safety and performance, and erode trust tomorrow.
By collaborating closely with the customer, creatively combining elements from different systems, and maximizing use of available chassis space, the team provided a solution that fit the customer’s operational needs in the desired footprint. Now, the customer has a rugged, high-power server that delivers seamless, reliable performance at the edge while saving valuable space (and weight) on every installation. Proof that reducing size doesn’t mean reducing capability.