++luu for an interesting read. One thought on hardware acceleration: it may be kind of happening, piece by piece. AES-NI, carry-less multiply, and CRC32 instru

++luu for an interesting read. One thought on hardware acceleration: it may be kind of happening, piece by piece. AES-NI, carry-less multiply, and CRC32 instructions are kind of like smallish (in gate area) accelerators of operations that go much easier in hardware. AMD talked about a compression accelerator for their (now-delayed) A1100 ARM server chip. These are marketed as performance features, not as power-saving, but if you can do something in fewer clock cycles without a ton of area, you are often saving on power as well. It's also possible that some savings come through finding places where slower chips will do the job because RAM and/or disk and/or bandwidth is the limiter (storage, caching, maybe some proxies and scale-out DBs now running on faster chips). Intel has server Atoms (Avoton, Rangeley) and there are experiments with ARM servers (Baidu cloud storage tried them out: https://gigaom.com/2013/02/20/first-arm-based-servers-in-production-support-baidus-cloud-storage/ https://gigaom.com/2013/02/20/first-arm-based-servers-in-pro...). There's lots of work on ARMs and Atoms on the consumer side and some of the benefits may spill over to low-power servers. Don't know how widespread this will be, though; I've been rooting for the concept of servers on low-power chips for years and haven't seen much yet. You also pay for higher power dissipation with lower density. If you're Amazon or some huge provider, you can be talking about a lot of space in absolute terms. It's interesting to eyeball a latest-gen Facebook Open Compute server with high-wattage CPUs in it, and how much of its height seems to be due to those heatsinks and fans: http://images.anandtech.com/doci/9138/leopard-mob.jpg http://images.anandtech.com/doci/9138/leopard-mob.jpg SeaMicro sadly didn't succeed, but it's interesting they thought people wanted density enough to make it a key part of their marketing, and AMD thought enough of the idea to buy them: http://www.brightsideofnews.com/Data/2012_9_11/AMD-Expands-SeaMicro-Big-Data-Gets-a-High-Performance-Home/AMD_SeaMicro7_689.jpg http://www.brightsideofnews.com/Data/2012_9_11/AMD-Expands-S... Of course, companies using and selling blade servers and the popularity of <1U configs (e.g., SuperMicro's twin servers: http://www.supermicro.com/products/nfo/1UTwin.cfm http://www.supermicro.com/products/nfo/1UTwin.cfm, Microsoft's blade-y variation on the idea: http://www.anandtech.com/show/9138/open-compute-hardware-tried-and-tested/9 http://www.anandtech.com/show/9138/open-compute-hardware-tri..., even that Facebook Open Compute layout which fits 3 boxes in 2U) demonstrate some interest in density.