ASUS ESC A8A-E12U Power Consumption
In terms of power, we did not have this hooked up to our standard monitoring at the PDU. At the same time, you can see a few interesting things from the telemetry data provided by the machine itself.
For example, even at idle, the eight GPUs combine for 1kW+ of power consumption alone. That does not include the CPUs, NICs, optics, SSDs, PCIe switches, cooling, or other components. We could also hit 12kW at the higher end on this system, and there is probably room to go up from there. Those GPUs each can hit 1kW for 8kW total.
The 5+1 redundancy design is also a bit different but it cuts down on the need for more power supplies and probably reduces the height of the system by 1U.
STH Server Spider: ASUS ESC A8A-E12U
In the second half of 2018, we introduced the STH Server Spider as a quick reference to where a server system’s aptitude lies. Our goal is to start giving a quick visual depiction of the types of parameters that a server is targeted at.
This is a server designed for maximum AI compute. Perhaps the better metric is actually HBM3E memory capacity, since that is a major optimization point of the AMD Instinct MI325X GPUs. Given that we started this in 2018, we did not have HBM capacity/ U of rack space as a metric.
Final Words
Overall, this was a really neat system to get to take apart and look at.
With eight AMD Instinct MI325X GPUs, and 2TB of total HBM3E memory, there is clearly a lot happening in this system.
ASUS has a really neat design that both makes the system compact but also makes servicing the system fairly easy. Hopefully, folks can see this in our reviews. There is a lot of thought that goes into packaging so many components into such a tight space while also air-cooling the system.
One thing is for sure: with up to eleven PCIe card slots, two big CPUs, eight big GPUs, and lots of memory both on the CPU and GPU sides, this the ASUS ESC A8A-E12U one large AI machine.
