ARMed Servery

“After many years of ecosystem development and processor designs that failed to gain traction in the data center, ARM processor vendors, such as Applied Micro and Cavium, have garnered notable design wins and partnerships from communications service providers (CSPs) and systems vendors representing a wide spectrum of end customers and workloads. IDC expects that in 2017, ARM vendors will begin to gain market traction with their newest generation of designs.”
 
See Gradual Change in Server Microprocessor Market; IDC Expects Competition and Evolving Workloads to Change Supply Ecosystem in 2017
Yep, GNU/Linux on ARM is gaining traction on servers. My replacement for Beast won’t have much impact but it’s a sign that the ecosystem is growing that has reached the retail level where I shop.

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology and tagged , , , , , , , . Bookmark the permalink.

5 Responses to ARMed Servery

  1. oiaohm says:

    Deaf Spy quote something that review was against the older tech.
    The A57 is probably not strong enough for the “non-micro server” market and it remains to be seen if the A72 will a large enough improvement.
    Yep the arm64 A57 chips released in Oct 2015 after that review proved that point wrong using a Jun 2015 review shows you missed the moonshot hardware rebuild.

    There are 3 revisions to the arm moonshot module each has the same general specs but reducing power usage.
    http://www.cnx-software.com/2015/04/14/server-performance-power-usage-comparison-intel-xeon-e3-vs-intel-atom-c2720-vs-applied-micro-x-gene-1-vs-ibm-power-8/
    Do take note this review is intentionally bias against the micro-x-gene by using a distribution with different and older revision kernel to everything else so missing major performance fixes. Yes tie a applied micro x gene chip hand behind back and atom still lost by large margins. Note the atom in that was 14nm so go darn what did intel do wrong.

    Deaf Spy when you were suggesting atom something this is why I responded with that is worthless crap. Atoms are bad TCO.

    This and yours are first generation arm64. 40nm. Yes 40nm vs 14nm so of course Xeon was winning. Yes first generation arm64 like the applied micro-x-gene beat atom chips in every metric.
    http://www.androidauthority.com/14nm-exynos-7420-power-efficiency-621010/
    Change from Applied Micro X-Gene 1 40nm to Applied Micro X-Gene 3 16nnm cuts the power usage of the Applied Micro by 1/4.

    Deaf Spy so the Xeon power advantage has turned out to be nm of production. http://www.nextplatform.com/2016/04/28/arm-server-chips-xeon-class-matter/
    There is quite a increase in performance as well 40nm limits top clockspeed as well.

    Moonshot arm modules come in two versions old Applied Micro X-Gene 1 and newish Applied Micro X-Gene 2. The X-Gene 2 cut power usage in half. So idle better than Xeon-E full load slightly worse than a Xeon-E if you have X-Gene 2.

    In performance per watt the Applied Micro X-Gene 3 cleanly wins over Xeon-D same with all the other arm64 chips out there made at 16nm and like the Applied Micro X-Gene 2 at 28 nm is around the same as a Xeon-D at 14 nm. To beat 16nm arm64 chips using intel you have to use a Xeon-E5 and then the difference in full load performance may be completely undermined by the X-Gene 3 having way better idle and true core per thread stability.

    Robert Pogson AMD has a major problem due to their FAB for silicon production still being stuck in the 32nm stone age.

    This is the problem there are quite a few decent Xeon competitive arm64 chips that appeared end of 2015 start of 2016 problem is getting hands on them.

    The applied micro x-gene 3 modules are going to be nice just question is how affordable.

  2. Xeon-D sounds like a nice try by Intel but there’s this price thing. The cheapest server with Xeon-D? Newegg.ca has a Xeon-D 1520 mobo for $640. That’s without RAM… and it’s 45W TDP for 4 cores and 2.2gHz clock. ARM can compete with that although if I weren’t set to go ARMed into the future it would be a good choice. It’s certainly a better server board than the crippled Cello which has fewer SATA and memory slots and NICs and… but Cello is good enough.

    I suppose FaceBook would like it but if I can’t buy it in my budget, what’s the point? I could probably buy one of FaceBook’s obsolete server farms too, but I can’t afford it.

    It’s interesting that FaceBook is quoted as preferring the Xeon D. That’s last year’s news. I wonder what they will think in 2017 when ARM is pushing 7-10nm and much higher clocks? IT is not standing still. That’s why Intel bothered to develop Xeon D.

  3. Deaf Spy says:

    IDC… These who predicted that Windows Phone will surpass iOS by 2016? 🙂

    Meantime, in the real world:
    http://www.anandtech.com/show/9185/intel-xeon-d-review-performance-per-watt-server-soc-champion

  4. Deaf Spy says:

    where the Xeon only can fit 4 because the heatsink is in the way.

    Bwaha-ha-ha-ha!

  5. oiaohm says:

    There is more to it. When you look at moonshot parts at first it will seam that Intel is faster but after a little more careful looking you see a big problem.

    Moonshot arm64 is 8 core 2.4 ghz flat
    Moonshot Xeon fasts is 4 core 2.9 ghz with a turbo to 3.8 for a single core.

    Please note arm64 2.4ghz flat meaning that its does not clock up or down so it has two states 2.4ghz or asleep/stoped. Next its 8 cores is its not like the Xeon simulating 8 threads having to basically cut it processing power in half so the Xeon is only really 1.4-1.9Ghz once you split it to handle as many threads as the arm64 is so the Xeon is not as fast as it first appears. Finally the arm64 2.4ghz runs cold does not need a huge heatsink so it has 8 slots of ram where the Xeon only can fit 4 because the heatsink is in the way. Arm64 has 64G of ram and the Xeon only has 32G of ram. Yes having 8 sticks of ram instead of 4 under heavy load does slow down how fast the ram heats up as well.

    So out of this comes a stack of horible. So you have the arm64 and the Xeon do the same workload. Result is Xeon card gets hotter ends up at risk of slowing down to under 1.8Ghz to prevent over heating. So max load can equal the very time an X86 cpu wants to run slow and that is a very worst time you want a cpu to decide to do that. This means the arm64 chips run along dependable and predictable where x86 turn out to be roll the dice and pray. 8G of ram per thread the cpu can handle is a nice value yet the Xeon only has 4G per thread the cpu can handle. The lower ram value equals high risk of ram thrash yes this is another way to lose performance. Ram thrash is not having to use swap it when you are needing to transfer ram allocations repeatable between cpu cores so having to flush buffers and other thing consuming a lot of time.

    Xeon was good but now that we are seeing 8 and 16 core arm64 with accelerators Intel chip lead has disappeared. Arm64 high power effectiveness with low heat generation is coming into its own.

    That is just from moonshot with the older tech. So 2017 there is going to be a battle in the higher end for sure.

Leave a Reply