Performance and efficiencies of micro-servers based on system-on-chips with ARM architecture exceed expectations and there is enthusiasm towards those products among server makers. The commercial machines are on-track to emerge next-year, as planned, but the takeoff will hardly be really quick, even though the industry push towards energy-efficiency is growing.
“We are seeing increased momentum [on servers]. We are twelve months closer to 2014 and, therefore, the appearance of sensible, commercially available ARM-based servers, we are more certain the timescales are not actually changing. So we are transitioning from there being no servers to some servers, and 2014 is when it's going to be something that you can actually notice,” said Warren East, chief executive officer of ARM, during a conference call with financial analysts on Tuesday.
The market of server processors based on ARM’s 64-bit architecture with server-class enhancements will be pretty broad as numerous companies, including AMD, Applied Micro, Cavium, Calxeda, Marvell, Nvidia, Samsung and Qualcomm, are developing ARMv8-based system-on-chips aimed at emerging market of micro-servers.
While there are naturally no performance benchmarks of ARM Cortex-A53 or Cortex-A57 processors, it is noteworthy that the feedback ARM gets from its partners is generally positive. Apparently, micro-servers powered by ARM architecture clearly consume less power than x86 machines and can do the job at decent pace.
“In terms of the feedback that is coming back, so far the experimental work is actually exceeding expectations. The original theoretical stuff has been proved and more and that is fuelling some enthusiasm, and you saw some announcements in 2012 and I think we will see some more in 2013,” added Mr. East.
Micro-servers are currently developed by Dell (project Copper), Hewlett-Packard (project Moonshot), SeaMicro/AMD and Quanta.
Tags: ARM, ARMv8, Cortex, AMD, Applied Micro, Cavium, Calxeda, Marvell, Nvidia, Samsung, Qualcomm, Dell, HP, Quanta, SeaMicro, Copper, Moonshot
Comments currently: 2
Discussion started: 02/06/13 05:06:38 PM
Latest comment: 02/06/13 09:44:52 PM
Expand all threads
| Collapse all threads
Does anyone know why pending 64-bit ARM chips are still limited to four cores? It appears that further scaling will instead be achieved by connecting many chips to a fabric. Given the transistor budgets afforded by modern processes though, why not pile on the cores--they are relatively small, are they not?
I'd like to see something like that available for home usage as well, for those of us who aren't tied to x86 in any way. It is disappointing that both clock speed and core count have leveled off in recent times.
02/06/13 05:06:38 PM]
- collapse thread
There are no reasons to be limited to just 4 cores, you can go custom multicores for server applications. But preference is for smaller dies to increase yields as much as possible to offer much more packing and density.
Seamicro fabric relies on packing a CPU (of any ISA), RAM and FreedomFabric ASIC on a very small board (the mobo). As long as the die is smaller and you still squeeze in more cores, then its all fine. But studies have shown the data center workloads don’t really need super-fast single threaded performance todays x86 chips offer but rather more threads are needed on less complex workloads.
One way is obviously a chip like an 8/16 arm cores for instance or you can simply have much smaller core density per chip but link up many more of those chips to form a dense fabric 3D-torus.
If you can do tomorrow for the same die size today of a 4 core arm and then squeeze in 16 cores, sure no problem its all fine. But there is a set idea die size to maintain the Seamicro miniboard’s size advantage in terms overall CPU package area consumption.
Surely for the die size of today’s Ivy Bridge chips used in some of the SeamMicro offerings, we can squeeze in a larger multicore ARM solution. Nothing technically stopping anyone from doing it, but there is that ideal die size/package size restriction in order to maintain Seamicro’s density and packing targets.
Seamicro is all about offering 50-75% power reductions, space savings and fully manageable super fast network fabric. So in line with all that a quad core ARM 64 solution will do a fine job of handling the tasks at a much lower power density per chip and over all compared to a best in class x86 solution. This is an evolving need so there is nothing to say a more than quad core solution won’t come out in the near future.
For commercial usage: If there are more mainstream apps in the future that truly stress that much on more TLP, such cores will come.
Battery life is still tough and all sort of new thinking is needed just to extend battery life. Battery tech is not progressing fast enough to keep up.
Real world usage analysis of most quad cores on smartphones and tablets show that a large percentage of whole cores or partial cores remain idle or unused due to the good scheduling schemes in OS and optimizations in apps which are ongoing.
So its not much use to go for more cores to satisfy a small market of apps that need even more threads or say scale well with more threads, even if say such battery tech was available today to power such devices well beyond a day.
Such battery tech is not available today, so pretty much the progress is ruled by what battery tech can offer today and what the limiting factors are. Heat generation is another issue but no:1 limitation is still battery tech.
02/06/13 09:44:52 PM]
Add your Comment
Enter your username and e-mail address. Password will be sent to you.