On Fri, May 19, 2017 at 5:08 AM, Neil Jansen njansen1@gmail.com wrote:
On Thu, May 18, 2017 at 11:29 PM, Luke Kenneth Casson Leighton lkcl@lkcl.net wrote:
size, power budget. in about 5-8 years it won't be an issue. EOMA200 is better suited to clustering. bigger PCB size and a much higher power budget.
I'm not really concerned about size. What do you mean by power budget?
5W per card in the current standard, with an "option" to increase that to 10W *if the housing supports it*. that means having sufficient thermal cooling/
What's to stop me or anyone else from buying, say, 8x EOMA68's and networking them?
nothing.
(gigabit or otherwise)?
no ethernet. USB3 *if the Card supports it*... which the A20 doesn't.
Assuming that a backplane is the only thing in the way, that could happen pretty easily. The only other thing missing at that point would be availability of the 1st gen cards, but that should get better in time, I'd think. Cost wise it's still less than an Intel NUC, if you're not counting the cost of the backplane.
you can't get a 1000 pin 15W *processor* nor can you get 128-bit-wide DDR3 memory bus bandwidth into an EOMA68 case. ok the latter you might be able to do if you used 4x 32-bit-wide LPDDR3 RAM ICs but the entire 5W power budget would be blown on running the RAM even at only around 1066mhz.