After posting news yesterday of Penguin’s Altux 600 I was curious to know what sorts of “features not typically needed by HPC customers” they had actually removed.
Peguin Computing was kind enough to get back to me last night with an answer (thanks, Alan!).
Hot-swap capability for drives, fans, and PCI-cards …[and the]… PCBs necessary to support hot-swap functionality, as well as, the supporting firmware and motherboard support.
Chassis: By deploying a simple chassis design utilizing 2 fixed hard drives and electing to eliminate a “tool-less” service design the cost of Altus 600 is further reduced. By comparison Altus 1600 [Ed.: an enterprise server with some of those features that aren’t needed in HPC] utilizes a tool-less chassis, offers 4 hot-swap drive trays, a tool-less PCI card bracket, and a redundant power supply option. We elected to remove these features for the Altus 600 because cluster environments frequently return the entire system to the factory for repair.
PCI Expansion: Altus 600 has a single PCI-E x16 slot versus 2 PCI slots in the Altus 1600. This eliminates motherboard connectors, circuitry, firmware design, and risers.
These design tradeoff reduce the overall cost of the Altus 600 while preserving support for important HPC features, such as, current dual-core 2000-series Opteron processors and future quad-core processors. Altus 600’s PCI-E x16 slot bandwidth is capable of supporting the latest high-speed low-latency interconnects like Infiniband DDR.
This is an interesting feature set to remove. On the one hand in my experience HPC machines are never returned to the factory for repair, but maybe that is true in the kinds of environments that Penguin Computing typically deploys in.
On the other hand I’m sure these tradeoffs also offer substantial savings in power consumption, and that is a critical issue that all HPC centers are facing today.