I agree. In microservers, I think it's nearer than many people think, but beyond that, Intel's dominance is still there to stay for a good while. Still, it's a start and if that means reduced prices and better service, great!
Exiting times, yes, but It's easy for Intel's PR to make marketing claims just like they did with the previous Atom (and look how "good" it was in reality...). Note 2nd generation 3GHz X-Gene is 28nm and will be available mid-2014. If you look at today's devices then current Avoton at 22nm doesn't look all that great vs ECX-2000 at 28nm (http://www.theregister.co.uk/2013/10/28/calxeda_midway_launch/). Given 2014 will bring even faster and lower power 64-bit ARMs as well as TSMC 20nm, and the new entrants go after a multitude of markets which Intel doesn't have covered today, I'd say that Intel will have a hard time competing.
Correct. BTW, I personally do not understand some people's obsession with the node used. The actual performance, energy effiency etc. depend on much more than that. Let us examine what really matters: performance per dollar spent, performance per Watt consumed, total cost, total power, typical performance, maintainability, diversity of supply etc. etc. of the full system. We are no longer in the era where a computer system could be summed up by one single number!
I don't see how lower power server chips based on ARM cores can replace (in the immediate future) Intel-based ones in these days where much of the computing & storage has already moved to the cloud. Computing bandwidth and latency (excluding the communication layer) are two criteria where Intel clearly leads.
Perhaps by mid to late 2014 we will see comparable ARM-based servers' performance but I remain skeptical.
WIth all due respect. HP has talked about shipping servers 1H 2014 for production with ARM based server SoCs from Applied Micro (64bit), TI and Calxeda. Partners like AMD have announced that they will be shipping next year. We already have hyperscale players like Baidu deploying with Marvell's server solutions. We are already quite far down the long and winding road with the gate to deployment just a few steps away.
we didn't really expect Intel to roll over and to give all their server business to ARM, didn't we?...it goes without saying that you have the work hard to win new biz and the incument will fight back with all possible means, price, technology and PR etc
Dell timelines should be an year behind HP and other OEMs. Dell is hardly a trend-setter in the server space. It has been over 2 years since HP announced moonshot and has servers shipping today. Dell has nothing to show yet.
To me, it is always long shot for ARM to get into Intel dominant server business. But I don't think Intel will be winner either. Under pressue, Intel will be forced to come out low cost server chips to fend off Competition from ARM. This will hurt Intel Business overall with falling ASP. At end, winners are likely system Companies such as HP and Dell since they have better price negotiation power againt Intel, or end users to have access on low cost servers. Although it is hard for ARM to get into general purpose server market, I still believe ARM could be used by large service Companies such as Google, Facebook, Microsoft to build purpose build servers to address their own specific need.
I am not sure OEMs are really leveraging this opportunity to get back into more value-added system design. My observation is Dell and HP seem to be letting the chip makers do a fair amount of the board design and just creating passive backplane chassis for them to ride in. Hopefully, this won't be their missed opportunity.
As for Intel, it's actually the only company shipping microserver SoCs now, admittedly at lower ASPs than it is used to.
uummm strange post. It seems like all servers load can be accomodate with large clusters of tiny socks. Unfortunately, most of real world workloads are very sensitive to monothread performance. So yes, in this upcoming 10/15 percent o server world intel will lose a bit, still the bulk of server sold will remain an high ipc cpus territory.
uummm strange post. It seems like all servers load can be accomodate with large clusters of tiny socks. Unfortunately, most of real world workloads are very sensitive to monothread performance. So yes, in this upcoming 10/15 percent of server world intel will lose a bit asp, still the bulk of server sold will remain an high ipc cpus territory.
Most servers are i/o limited, not CPU limited unlike desktop PCs, so ARM server chips would be fine for the majority of servers and server farms.
Also note that most servers handle many requests per second, and so having multiple cores or multiple servers in a server farm fed by a round robin router works very well, again allowing use of more lower powered ARM servers, or lower powered ARM cores. Indeed far more so than with desktop computers.
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...