Announcement

Collapse
No announcement yet.

Qualcomm Reportedly Wanting To Exit ARM Server CPU Business

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Qualcomm Reportedly Wanting To Exit ARM Server CPU Business

    Phoronix: Qualcomm Reportedly Wanting To Exit ARM Server CPU Business

    Calxeda as the first interesting ARM-based servers didn't pan out and the company went bust, attempts by the likes of AMD at ARM server CPUs so far have not panned out, and now today is a report that Qualcomm is looking to end its Centriq server CPU line or sell off that division...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Find some less fixable spectre bugs – that should force all cloud computing vendors over to ARM Cortex-A53/55.

    Comment


    • #3
      This is pretty disappointing. ARM has a real chance of gaining marketshare in the data centre. Unfortunately, it requires companies who are willing to take a risk. QCOMM no longer wants to take any risks. They want to play it safe .

      Comment


      • #4
        Also Thunder X2 looks like a better choice than Falkor.

        Comment


        • #5
          Originally posted by SyXbiT View Post
          This is pretty disappointing. ARM has a real chance of gaining marketshare in the data centre. Unfortunately, it requires companies who are willing to take a risk. QCOMM no longer wants to take any risks. They want to play it safe .
          The issue wasn't risk.

          Qualcomm put up some serious dollars in R&D and was also funding the Microsoft Windows Server for Centriq effort. The risk they took was pretty significant.

          The issue is strictly the pressure Broadcom put on them in the boardroom as part of the NXP acquisition.

          While Broadcom is smaller than Qualcomm, they could see how much of QCOM cell phone revenue was being diverted into Centriq and M&A and took advantage of the situation.

          With Apple on the verge of developing their own radio, dropping QCOM and Intel, QCOM needed a new market to penetrate. (servers)

          If Centriq is indeed cancelled or sold, it will become another footnote in ARM like Calxeda. QCOM will have to buy off Broadcom to fend them off and retrench where they are going to go post cell phones.

          Comment


          • #6
            Originally posted by SyXbiT View Post
            This is pretty disappointing. ARM has a real chance of gaining marketshare in the data centre. Unfortunately, it requires companies who are willing to take a risk. QCOMM no longer wants to take any risks. They want to play it safe .
            I'm sure Qualcomm being risk adverse is part of the problem, but not the whole story. Love it or hate it, but the server space is dominated by Linux these days and Qualcomm is absolutely notorious for not playing nice with any kind of open source efforts. Qualcomm offers driver blobs but they don't maintain them properly, leaving proper upgrade paths out of the question. This probably plays into a lack of adoption by server providers and definitely does with Linux kernel devs.

            One other reason would be that hundreds of millions of lines of software is quite literally dependent upon the largely standardized, defacto if not official, x86 and AMD64 architectures. Corporations aren't going to change all that over to a new incompatible architecture without a great deal of foot dragging, political dog fights, and basically having a gun to their heads to do so. This is largely what doomed Intel's Itanium but AMD recognized by creating AMD64 as an extension of the legacy x86 architecture. It was mostly seamless to move from x86 to x86-64. Needed legacy compatibility? No problem, keep running 32 bit OS. Want the performance and memory capabilities? Migrate to AMD64 as needed. I know Itanium initially contained a backwards compatibility layer, but it had worse performance than hardware already on the market and it was slated to be removed soon into the product life cycle. Corporations and research institutions don't like that.

            Comment


            • #7
              As hinted above, it all comes down to a lack of investment by any of the ARM server hardware vendors in software support.

              Any alternative architecture that wants to stand a chance needs to have an affordable, capable platform for developers to play with given that much of the software today is open source and somewhat hobbyist developed.

              But no one wants to offer the hardware.

              If you can't get an affordable platform into the hands of the public, then you don't get developers, then the development and testing isn't going to get done and the server customers aren't going to have any confidence in the platform and refuse to buy it.

              Comment


              • #8
                good, now the better choice will undoubtedly be epyc

                and when rome arrives... (insert maniac laugh)
                Last edited by davidbepo; 09 May 2018, 02:53 PM.

                Comment


                • #9
                  Originally posted by stormcrow View Post

                  I'm sure Qualcomm being risk adverse is part of the problem, but not the whole story. Love it or hate it, but the server space is dominated by Linux these days and Qualcomm is absolutely notorious for not playing nice with any kind of open source efforts. Qualcomm offers driver blobs but they don't maintain them properly, leaving proper upgrade paths out of the question. This probably plays into a lack of adoption by server providers and definitely does with Linux kernel devs.
                  don't confuse their server effort w/ their android/mobile effort[1].. everything was upstream for their server chips (had to be, else no RHEL, etc). The arm server effort for all involved is 100% about linux, so not having upstream kernel support is not an option.

                  I think this all comes down to improving profits in the short term to make wall street happy, basically the same thing that broadcom would have done. I'm not sure it is healthy in the long run.

                  [1] although situation on mobile side has been significantly improving in last couple years, and they are migrating to upstream drivers/frameworks which should significantly reduce their technical debt and difficulty in upstreaming drivers and migrating to new kernels, but I guess it will be another generation or two of mobile SoC's and android kernel versions before you really see that pay out.. but if you look at the progress of various enthusiasts in the community getting upstream kernels running on devices, not to mention upstream kernel support for various qcom arm boards, you can see that this is already happening

                  Comment


                  • #10
                    Maybe Linaro + Socionext could do it: That 24-core ARM "desktop"

                    Comment

                    Working...
                    X