ARMs head Moonshot bodies: HP pops Applied Micro, TI chips into carts HP taking another shot at its Project Moonshot by today announcing two new ARM-powered servers, one 64-bit and the other 32-bit. The US giant hopes its two new ProLiant models will live up to the hype: for years now the ARM world has claimed the architecture will provide high-density, power efficient systems for data centers …
I can see telcos buying these boxes to replace existing proprietary ones. I can't imagine anywhere is using x86 for this kind of workload as it is suited to dedicated workloads. Buyers will have to weigh the potential extra costs of patching and deploying software themselves against any savings and independence from suppliers.
Intel doesn't need to worry, yet because it's not in those markets. But, of course, if the boxes turn out to deliver the right performance with low power draw and at an acceptable price then there will be appetite for more and it will presumably be easy enough to plugin whichever modules are required whether its multimedia or cryptography. At some point someone will try them as file or web servers.
Why would it get removed? What would be your technical reasons as a paid professional for enacting such a wholesale change?
Given that it's based on Debian (which is used by a great many as a solid server OS) and Canonical builds on Debian's solid foundation with some pretty good enterprise-scale deployment and management tools, where does your problem lie?
Do you have an actual technical reason? Did you only ever use the desktop version and it didn't support your WiFi card? Or just distro-specific bigotry?
Because if you took a working commercial solution on an infrastructure that while you may run, isn't actually your property and ran the risk of endangering shareholder value with a wholesale OS change simply becuase "I don't like it.", then YOU would be "the first thing to get removed" at my shop.
This is a tool which is designed to provide DSP processing resources which will ship with development tools and support under one OS. Most of the features involved included shared memory access and synchronization between an ARM processor and 8 DSP cores.
I just can't imagine why the OS would even matter for you. I guess you want to buy a really expensive DSP box and redesign it from scratch?
BTW... do you actually have any idea what a DSP is or how it works and where you would use it?
From what I was told on a course recently, replacing the OS may not be as easy as all that as these blades are pretty bespoke to the task. The idea being that if you need one to do X then they design one around X with hardware and software. If you have a software stack they can design and test you a blade in a couple of months ready for deployment so it's quite flexible despite the inflexibility.
TI's DSP toolchain has sucked fantastically for years. Their GCC tools for DSP are horrible. Their C++ compiler is trash. Debuggers are even worse in most cases. They keep building their DSPs as if they are trying to make them general purpose computing friendly. It attracts shitty developers who can't write pipelined code and then complain about how poorly the DSP performs.
The TI DSP just doesn't offer good performance per watt per dollar spent developing anymore.
These days, it is 100 times smarter to implement a system based for example on Altera's tools which start by simply providing a general purpose processor in a relatively small footprint. Then you can profile your code and develop VHDL, Verilog or SystemC pipelines to improve performance. This allows multi-stage single clock pipelines to be implemented directly in logic which will provide substantially more performance per watt. Even better, you don't even feel like you've been robbed because the compiler sucked.
It's a shame that HP did this... but I guess some dumb ass will buy a pile of these and write code that will never run on anything new. Seems like a stupid investment to use this technology and a very unlikely investment to pay off.
Seems to me the whole point of the moonshot. Though I'm pretty sure the idea of the TI DSP comes from an existing or prospective customer.
I'm sure the idea is to support the kind of architecture you're talking about with the standardisation of modules an attack on proprietary solutions. ARM can be a baseline for booting the rest of the kit which might be more ARM or other stuff. Good luck to them if it works.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
