ARM unveils 64-bit Cortex-A50 server processors

ARM unveils 64-bit Cortex-A50 server processors

Summary: AMR prepares to go up against chip making titan Intel with a Cortex-A50 series of processors is based on the ARMv8 architecture


British chip designer ARM has unveiled a new series of 64-bit microprocessors aimed at the server market.

The ARM Cortex-A50 series of processors is based on the ARMv8 architecture, with two designs slated for initial release. These new chips are the Cortex-A53 and the Cortex-A57. Both chips will feature energy-efficient 64-bit processing technology, as well as support for existing 32-bit processing.

The Cortex-A53 is designed as a high-performance part, while the Cortex-A57 has been designed with power efficiency in mind. ARM also claims that the Cortex-A57 is world's smallest 64-bit processor.

The chips can operate separately, or combined into a ARM big.LITTLE processor combination. In this scenario the Cortex-A53 does the heavy lifting while the Cortex-A57 is used for any low-power applications.

Licensees of the new processor series include AMD, Broadcom, Calxeda, HiSilicon, Samsung and STMicroelectronics.

U.S.-based chip-maker AMD, which has resisted licensing ARM designs until 64-bit processors were available, has already announced plans to manufacture 64-bit server processors based on ARM's Cortex-A50 series designs, beginning in 2014.

According to AMD, these chips will fill a demand for denser sever clusters that pack more computing power into a smaller space. This, in turn, allows server rooms to be more efficient because less cooling is required.

Topics: ARM, Hardware, Processors, Servers

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.


Log in or register to join the discussion
  • By 2014, if nothing slips, these probably wont even be competitve.

    I think intel will have lower power and higher density than this, while retaining a significant perf advantage. And whos running data centers full of underutilized servers anymore? If a workload isnt fully using the cpu throw a few hundred more vms on that box. We dont need low performance servers.
    Johnny Vegas
    • Intel is trailing

      Has been trailing, and will be trailing into the foreseeable future.

      Challenge for you:
      Name one time point at which Intel has produced more flops per watt than Arm.
      • and the point is?

        Since when was FLOPS per watt a measure of anything in servers. Servers and floating point operations? I think not.
      • Only in that and power are they trailing

        There manufacturing process lead has been growing. I've heard about 3 years now. I don't expect that will continue.

        Intel right now simply outperforms ARM per core, bottom line. That said, I do think Intel needs to make a course change or 4 years from now they may really be hurting.
    • It isn't an underutilization problem.

      The problem is that the type of workloads being processed have changed over the years. The WEB creates many (thousands and even tens of thousands) lightweight workloads. It is more efficient and achieves lower latencies to run these on a weak processor for each workload than to queue them up to run on fewer stronger processors.

      I agree that Intel isn't sitting still but so far they really haven't done much to significantly address this. This is what SUN Microsystem's (now Oracle) Niagra and Rock SPARC processors. Weaker than Intel for big workloads but better at lots of little workloads. AMD has continued to up its cores 16, but with Intel most of the Xeons are still 8 core with 2 threads per core.

      Intel has done work on their Teracomputing initiative but so far no real product addressing the many core scenario has come out.

      Even if they get the Xeon where it is running at the same power level as the ARMs A53 (not likely). It still won't be near as cheap or have the same package density. Also Intel is in danger of being surrounded by ARM manufacturers.
    • Intel's trouble

      The primary trouble Intel has is that they have optimised their chips to run Windows "faster". Traditionally, Windows has been mostly single or limited threading system and this sort of worked. But even Windows has now changed and other than Windows workloads actually can use huge number of cores, running at lower clock rates.
      The thing about power consumption is that for a given technology it more or less grows with clock rate. If you can reduce the clock rate, you can have higher per-watt total performance.

      None of these *huge* datacenters use Windows (or anything Microsoft) technology and so far they only use Intel CPUs because of associated benefits, such as good networking etc. But this is changing too --- Intel is busy on too many fronts and again, with the primary goal to please Windows.

      Now, as they head towards a divorce, we could see wonderful new chips from Intel, leveraging "traditional" technology and what they can do best: manufacturing.
      • Chips are NOT optimized simply for Windows

        UNIX, Netware OSes were running on Intel even before Windows 3. And you are confusing threads execution with cores. Windows has been able to run thousands of threads. If you are on an old single core with single thread processor they were simply switched. If you have more cores these threads can run in parallel. The number of threads doesn't change but you can execute more of them in parallel.

        Intel is not just trying to please windows. There are many many linux servers running on Intel or x86 if you will. Even the super computers are mostly running some linux variant and most of these are x86. x86 has taken market share from Power, SPARC and other RISC based super computers. And many of these x86s are Intel though a lot are AMD Opterons.
  • I think

    the DESIGNS were unveiled, rather than the chips (microprocessors). ARM does not make chips.

    It may be a while before the chips appear.
    • Yes and even when released they are sometimes not available in quantity.

      So perhaps, for example, you can get the 2 Ghz part initially but have to wait until 2015 until you get the 2.5 Ghz part. I never see a table indicating the power consumption at each frequency. They specify a potential range of frequencies and then only one power consumption (presumably at the lowest clock speed)
  • Typo

    Mr Kingsley-Hughes - you said " the Cortex-A53 does the heavy lifting while the Cortex-A53 is used for any low-power applications", I think you meant for one of those to be another processor number?