On Mon, Dec 14, 2009 at 12:29 PM, Randall Whitman <909li...@whizman.com> wrote:
[snip] > For the entire workstation, or the CPU only? Entire case, measured through the power cable. Kill-a-watt units are kinda handy. > Exactly - and for doing other work while a debugger is grinding away. > So yes, I was specifically comparing *4* cores ARM vs *2* cores x86, > for exactly these reasons (together with watt-draw comparison). > For my usage scenario, I was thinking i'd getting substantial benefit > from the third cpu core. I don't know about your usage scenario, but most schedulers tend to load balance the workload across all cores, unless it is possible to define the core affinity in the program (rarely seen, IMO). Also, the difference in raw power between processors should be taken into consideration, especially if the software is optimized for SSE4 and the likes. Once the Cortex becomes readily available I suppose we'll see benchmarks popping up on the internet. > That's the other thing i'd *really* like to know :) and the other potential > fly in the ointment for this idea... First I'd have to find a supplier > willing and able to produce and sell such a rig (or at least a motherboard > that can take 4-core ARM Cortex A9 and 4G RAM), and then I can see what > the price is. I imagine there will be custom boards for this, as today you can find 1U servers based on Atom processors, or very small boards targeted to the firewall market, also Atom-based. Quad core setups would depend on the Cortex supporting dual-socket scenarios though, and although this processor seems to be pretty good in terms of expandability, dual socket boards would be a bit too much, I think. Price would be another story though, since this is sort of commodity hardware, and I'd expect it to be a bit pricier than a regular C2D set, although you would easily offset the cost in power savings. -- Dante _______________________________________________ LinuxUsers mailing list LinuxUsers@socallinux.org http://socallinux.org/cgi-bin/mailman/listinfo/linuxusers
