Beast III

“Number of Cores Quad-Core
Processor Number 9850
64-bit Computing Yes
Compatible Processor Socket Socket AM2+
Clock Speed 2.5 GHz
Thermal Design Power 125 W
Type / Form Factor AMD Phenom X4 9850
Cache Memory Details L2 – 4 x 512 KB L3 – 2 MB”
 
See AMD Phenom X4 9850 Specs
The original “Beast” had a 32-bit CPU running at 1.8gHz, with a modest 512KB L2 cache. That tiny thing actually would do more for 30 students running sessions on it in 1.5GB RAM than Lose ’98 on thick clients. It was faster to log in, and open an application and it never lost students’ files. About 2008, I was in the arctic and preparing for a conference where I was going to demonstrate a cluster of thin clients running on a single decent PC with GNU/Linux, something unknown thereabouts. I bought a 64-bit quad core thingy simply because the supplier delayed a couple of weeks hemming and hawing about delivering a nice dual-core Opteron… Sigh. So I got 125W TDP and have regretted it ever since. Anyway, it is time to fix that.

These days, I am contemplating buying a Lemaker Cello, an ARMed server board intended for developers. It has an AMD A1120 CPU crippled quite a bit by lack of interfaces with these specs:
Quad Cortex A57
L2 cache 1MB (local)
L3 cache 8MB (global)
TDP 25W
The A57s have caches for data and instructions not too different than the original Beast (48KB instruction and 32KB data whereas the original Beast had only a 64KB + 64KB L1 and 512 KL L2. The A57 has 1.7gHz clock in the A1120 and the original Beast had 1.8gHz. Want to compare 1.5GB DDR400 RAM versus 32gB DDR3 1600? I’d say we have enough computing power for what we do today, with no labs full of twitchy students to serve… just me and TLW most of the time with a few visitors on special occasions. For what it’s worth, ARM64 code seems to be smaller by a considerable margin than AMD64, so those small caches may not be much of a problem. I expect with faster hard drives, Beast III will easily beat Beast II (7m to build a kernel). Beast I cost about $200 for CPU, $200 for mobo, $200 for RAM, and a bit more for video card and hard drives. Beast III will cost a similar amount in inflated dollars and give a lot more. I expect it may well be the last “PC” I buy except for the thin clients, Odroid-C2s. Sadly, it may still require a fan or two in the Cello. The Odroid-C2 is probably competitive for Beast I and it will cost less than $100 delivered.

I intend to run Debian GNU/Linux Stretch on these machines. It’s usable already, supports ARM64 on a wide variety of systems and is relatively bug-free. So, the world of IT turns and I get to enjoy the cooperative products of the whole world, standardizing hardware and software, and supplying it at marvellous prices.

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.

35 Responses to Beast III

  1. Modular Sunfish wrote, “I’d encourage you to give them a quick try before the new hardware gets committed to production use, even if you do feel likely to go back to Debian GNU/Linux.”

    You can bet I will play with the new ARMed systems before they go into production. I was thinking more of the file-systems and utilities but OS is fair game. There are two goals for my “play”: familiarization with the hardware and detecting infant mortality. When in production I want to keep this going. That means extensive testing of RAM, I/O, building kernels, testing individual applications, and creating scripts for backup and system maintenance. I will be able to start by hanging a huge new drive on Beast and using it via NFS over the network from thin clients. Beast has multiple gigabit/s NICs so I can check out the new networking components too. I can use one of several wide-screen TVs for monitors.

  2. Modular sunfish says:

    When you get the new hardware, you’ll have a window of opportunity to experiment with the software and even the operating system while setting it up. Since you’re a fan of Debian GNU/Linux, you might take the chance to look at two related variants.

    The one would be Debian GNU/kFreeBSD. It’s Debian GNU with the FreeBSD kernel and is the same except for three improvements: ZFS, PF, and no systemd. The other would be Devuan Jessie. It’s GNU/Linux and specifically a fork of Debian Jessie, but improved through the removal of systemd with all else being the same. So it’s closer to what you’ve been using before.

    I’d encourage you to give them a quick try before the new hardware gets committed to production use, even if you do feel likely to go back to Debian GNU/Linux.

  3. oiaohm says:

    –Huskyboard is virtually identical to the Cello. —

    This is correct. But the differences are important.
    Huskyboard A1100 cpu was a small batch made at TSMC.
    Cello is A1100 cpu made at Global Foundries.

    The issue here is Global Foundries AMD default fab still lacks 16/14nm mass production. The cpu on the Cello is Global Foundries first mass production at 28nm.

    So in silicon production tech Global Foundries is at least 24 months behind TMSC. TMSC is tapping out 10nm prepping up 7nm and Global Foundries is tapping out 28nm and just prepping up 14/16nm this 2 generations out of alignment. Global Foundries is at least 12 months behind Intel fabs. Due to Intel slowing down it cycling Global Foundries might catch up to Intel but no way are they going to catch TMSC until after TMSC has hit bottom and been there for some time.

    The A1100 line from AMD has been delayed by 12 months because Global Foundries could not produce the chip.

    It does not matter what the part is that much but nm of production is important.
    http://www.phoronix.com/scan.php?page=article&item=9800gtx-gtx-1080
    As you move nm down the power required to perform same task reduces and clock speed increases. So why AMD x86 64 bit processors and GPU are so horible is why the A1100 is horrible compared to other arm64 chips.

    Most of the Internet Of Things stuff makes me cringe because in reality most of it is highly power ineffective hardware. You cannot make power effective cpu, gpu, ram with large nm values rules of physics forbid it. Add on the fact Internet of Things are making a lot of custom one off SOC designs also making the power usage worse due to design of SOC not being heavily refined. Yes items like raspberry pi 3 are cheap but they are cheap because they are horrible in performance per watt and there is nothing you can do to fix that because its all based on the nm the chip was pressed out on.

    High end mobile phones are 1 nm range newer than mid range mobile phones because these devices depend on batteries and to have good battery life and higher performance than mid range mobile phones they need it.

    With nm of production we are in a race to the bottom with the end really close. TMSC already has prototype production at 5 and 3. 3nm is still a question mark but that is the silicon limit. 1nm graphite/carbon because you are needing smaller atom sizes issue here is in elements that join up into a structure like silicon there is nothing smaller than carbon and to keep size down you only really have 8 elements to play with counting carbon. The thing is change to graphite will be interesting. 427 Ghz room temperature cpu/gpu chips(silicon room temp currently is 2-3Ghz with forcast max at 3nm of only 5ghz) will become possible with graphite so who ever can get there first will have quite an advantage(Prototypes of 1000 transistors exist mass production issues have not be solved).

    Its expected by 2022 that production will be at 3nm. Thing to remember is 7nm was forecast to go into mass production in 2018 but its going into mass production in 2017. 5nm was looking at 2020 and it looks like it going to be early as well. So at best we could be at 3nm the end of silicon by 2020. Mostly because someone worked out something sensible.

    What I am about to type will sound complete common sense and have you asking why were they not doing this sooner. Up until 7nm design used most exactly the same width of silicon tracks for everything. At 7nm someone has a bright spark moment(common sense catching up) if we us 10nm for power transport lines and 7nm and smaller for logic things will be a lot simpler. Result now is going down to 5nm and 3nm are a lot simpler it also solves one of the biggest heat and short causes in silicon. Of course attempting to stuff too many electrons through something its going to heat up due to resistance and heat equals requiring stronger insulation between tracks result a lot of complex on going research into making strong and strong insulation on silicon chips instead of seeing the heat cause and fixing it has happened with every silicon chip production up until TMSC with there prototype 7nm tap outs. This is what has opened up the rapid pace down to 3nm. Of course TMSC has been smart and patented this.

    So expect to see some issues when intel and global foundries attempt to go past 10nm.

    This is my big problem the best nm fabs are not Intel or AMD ones and its going to be that way for quite some time.

  4. Dr Loser wrote, “it wasn’t actually the Cello, it was the original HuskyBoard.” and “You, Robert, are not a developer. And you never have been.”

    Chuckle. I worked my way through university writing software, first for a statistician and then the Cyclotron Laboratory. That was for money. I also did it for fun and my own studies. Later I developed software in schools: attendance applications, Moodle modules, web-pages, web-applications and databases and lots of automation for IT-systems. Probably the most exciting time for me was my first implementation of LTSP in a school. I wrote a fancy DHCP configuration to allow That Other OS and GNU/Linux LTSP to coexist on the same LAN. Basic feature was that my DHCP server could issue IP addresses to the Lab computers faster than TOOS and mine was smart enough to ignore TOOS clients booting. Another highlight was a medium-sized K-12 school with multiple servers and numerous clients of several types. I wrote DHCP config for that, set up a local website with several web-applications, configs for the thin clients which were VIA PITA boxes, web-caching and virus-scanning at the router and created desktop profiles and an e-mail system. I taught software development there too.

    The Huskyboard is virtually identical to the Cello except for number of SATA connectors. Either would do what I want. The Huskyboard would be slightly better. The same guys participated in the development of both at 96boards.com, Linaro, Red Hat and so on. Red Hat stated that it had worked on its software on all comers in the ARMed server market including these developer’s boards. Red Hat, in particular has a financial interest to see that all GNU/Linux software is ported to these ARMed servers so they have invested time and money. See the Youtube Video. Notice the guy from Red Hat is holding a HuskyBoard in his hands as he talks about the ecosystem and what such boards can do. I particularly like that he states the boards are designed to be much like motherboards of regular AMD64 boards, you know, with PCI-e, RJ-45 Ethernet etc. I particularly liked where he states folks are making perfectly usable desktop PCs by adding a video card. Yes, Red Hat wants their own people and others to develop software that runs on Red Hat GNU/Linux on ARMed servers. This is one of their tools to expand their ecosystem.

    Why the HuskyBoard is not yet sold, let alone the Cello, is a mystery. A large and influential organization of businesses and developers want these products. Perhaps they are fixing one last bug or building factories or … but it will happen because its relatively open technology right down to the chips, board layouts and circuit diagrams. Almost any business large or small can profitably crank these things out, just like netbooks. They sold many millions of units. That will grow quite a market for Red Hat and others.

  5. Dr Loser says:

    Red Hat is running their software on it.

    Back in November 2015, yes. And by “running,” I do hope that you don’t mean to imply “verifying.” And it wasn’t actually the Cello, it was the original HuskyBoard.

    Other than that, yes, Robert, all very thrilling.

    It’s a developers’ board and Red Hat has developers.

    a) It’s marketed as a “developers’ board.” Not the same thing at all.
    b) I’m going to need a cite before I believe that a single Red Hat developer is prepared to waste their time, developing on this thing. The mere presence of a pool of developers — and I don’t believe for an instant that you have any clue whatsoever about what Red Hat developers do — is no sort of proof whatsoever.
    c) You, Robert, are not a developer. And you never have been.

    For simplicity, I think I’ll stick with (c) on this one. The other two stand on their own merits.

  6. Dr Loser says:

    Oh dear, this sounds suspicious. Preorder to me, sounds like the buyers are funding the development.

    Once again, I find myself in total agreement with Dougie.

    Dougie, I think this is the beginning of a beautiful friendship.

  7. oiaohm wrote, “I see it as the wrong time to get small volume boards mostly because what you are getting is old and very low performing compared to what is coming. A73 design redone for servers is going to be impressive.”

    That’s not wrong, of course, but in my situation with Beast idling and using tons of power for nothing, the move from 45nm at 2.5gHz to 28nm at 1.7 gHz makes sense. In ten years, what’s the cost of 70W going 24×7? 70 X 24 X 365 X 10cents/KWH = $61. That’s not much, but it’s something. A replacement with Intel could cost the same. I value freedom from Intel much more than that. The hard drives on Beast are full and becoming less reliable. There is some urgency to the move. Also, I’m not getting any younger. In a year or two I may not even be able to make the move. The time is now.

  8. oiaohm says:

    http://wccftech.com/arm-process-roadmap-leaked-10nm-volume-production-begins-late-2016/

    Robert Pogson the chart here shows you the arm classifications matched to nm. So in server at moment you should be expecting the chip to be 28 or 16/14nm production. With 28nm being discontinued lines.

    40 nm like the raspberry pi 3 is embedded/iot class that class include 180/90/60/55/40 nm sizes. Yes 180 nm is still required for particular usages due to very high radiation resistance. 40nm is clearly not server so when you see someone building a board that way you go please don’t its not going to be good. 32nm and 28nm is on the wrong side of thing.

    Smartphones have been getting all the love. With them becoming a mature market and servers by the thousands being the next target, DIYers like me may get a shot later this year or next.
    Smartphones are about 1 generation ahead of Servers. Of course that chart I just gave is a 2 years old.

    http://www.eetimes.com/document.asp?doc_id=1329202
    If you look here TSMC is considering jumping straight over 10nm and going straight to 7nm for server hardware in 2017. If this plays out this will put Intel in a new location compared to where they have been of being chasing instead of leading in the nm race.

    Intel has announced slowing to a 3 chips cycles per level cycle. In fact by the time intel gets through the current 14 then through 10 Arm could be down at 3nm the rock bottom for silicon.

    Also the A73 for mobile from arm is very interesting. Altering instruction handling a bit arm has been able to make the new A73 process 3 instructions in the time the old A72 designed could only process 2. This is not a nano meter change but arm pulling rabbit out of hat by altering design. The old A72 arm design at 16/14 nm in performance as long as you double number of cores is fairly close to xeon at 14nm. Horrible fact is A72 with double the number of cores uses less silicon area than the Xeon and the A73 design is smaller again and faster.

    Yes the problem with A57 is that it getting very old from arm standards.

    Robert Pogson my problem is I see it as the wrong time to get small volume boards mostly because what you are getting is old and very low performing compared to what is coming. A73 design redone for servers is going to be impressive.

    Robert Pogson I don’t know of a single site to buy hisilicon boards that is english. This is the problem I have at the moment with arm64 you want diversity and quality you have to leave english behind.

    The other thing interesting about arm is early proto server boards for developers normal come out 12-18 months to places like Redhat and Ubuntu before open to general acquirement.

    http://www.tomshardware.com/reviews/arm-cortex-a72-architecture,4424.html
    Yes a72 at same clock speed is faster than a57 at everything then A73 design is massively faster again. Mind you all three take exactly the same instruction set.

    Basically what is wrong with the AMD arm processor is old design and very high nm both leaving you sitting in bad power usage and performance.

  9. dougman wrote, “this sounds suspicious. Preorder to me, sounds like the buyers are funding the development.”

    The development is done. Ramping up production and building an inventory is time-consuming and labour-intensive. They may also be shopping for the best deal by some builder. For an established product line that’s easier. Two people can discuss numbers and make it happen. For a startup it’s a problem, just like me buying a Chinese tractor or alternator. Just creating a distribution channel can be a problem. It seems they are trying to do everything in-house making them the bottleneck. It’s all an open specification too. If they succeed they may be swamped by others producing an identical product. That’s probably why Lemaker Cello is about to ship in volume and Huskyboard is nowhere to be seen. I don’t think preordering is about funding, just getting ahead of the rush. There is a lot of interest out there from manufacturers, developers of hardware and software and folks like me. I’ve read lots of Debian developers are held back by not having enough hardware to build stuff on. They’ve probably got ~100 developers chomping at the bit. Then there are enthusiasts like me who want freedom or just something different. I can’t wait to build my first kernel on it. That’s a process which is hungry for memory, CPU and I/O. It would be a hoot if this tiny thing beat Beast II by even a little bit but it could with faster memory and hard drives and a competitive CPU using 1/4 of the power. RedHat says they’ve gone out of their way to make the user-experience very similar to using a legacy PC, like plugging in a PCI card to get more NICs or storage or a video card. That’s cool, somewhat like Google bootstrapping Android on Java… The world is investing a lot in ARMed stuff and payoff will begin seriously in 2016. ARM expects to get a large share of servery in a few years. That could happen if this stuff all works and the effort put in so far should ensure that it does work. That’s why I see little risk for myself dipping my toe by buying a couple of motherboards. Yes, Debian is ready too.

  10. The Wiz wrote, ” Find me a URL to a review by someone who has hands on experience with the LeMaker Cello board. Your problem is that there does not appear to be anyone.”

    Red Hat is running their software on it. It’s a developers’ board and Red Hat has developers. As having software is important for maximizing profits from selling boards, they have priority. I get that. Red Hat has been testing and developing their software on prototypes for over a year. They’ve even made a few desktop systems by adding a graphics card to the PCI slot.

  11. dougman wrote, “why should you wait on some uber ARM board, when you can order everything you need right now and have it delivered within the week”?

    That’s easy. To be free from Intel is just as important as being free from M$. Both companies have committed foul deeds to stifle competition. M$ distributed crapware for many years, preferring to expend resources in keeping down competition. Intel actually paid OEMs not to use any products from AMD. Further, I’m free of x86-64 which was one of the tools of Wintel. Want the latest hardware? Buy M$’s latest release in a PC. Want the latest software? Buy Intel’s latest hardware in a PC. It was a racket and this year I should be completely free of it. In the ARMed world there are many designers and many sources of supply so I can shop in a competitive environment. Same with FLOSS. I like competition in IT. It allows the best ideas to rise to the top unlike Wintel where M$ stifled Intel’s ambitions in software as they did everyone else who wasn’t a “partner”.

  12. dougman says:

    In all serious, why should you wait on some uber ARM board, when you can order everything you need right now and have it delivered within the week.

  13. dougman says:

    Re: Find me a URL to a review by someone who has hands on experience with the LeMaker Cello board. Your problem is that there does not appear to be anyone.

    I could not find one either. No way would I dish out money for a board that has no reviews and/or is not sold by NewEgg or Amazon.

    “The LeMaker Cello is now only accept the pre-order, and the scheduled ship time is in Q2 of 2016.”

    http://www.lenovator.com/product/103.html#detail

    Hmmmmmm… why should someone have to preorder it? Why not wait till its in actual production and available for purchase.

    Oh dear, this sounds suspicious. Preorder to me, sounds like the buyers are funding the development.

  14. Wizard Emeritus says:

    “The chips have been around since 2014…”

    The issue is not with the chip, but with the board that you are interested in. Find me a URL to a review by someone who has hands on experience with the LeMaker Cello board. Your problem is that there does not appear to be anyone.

  15. Wizard Emeritus wrote, “it sure looks like the LeMaker Cello is actually nothing more at this point than a picture and a description. None of the “reviews” that are posted are based on hands on experience with an an actual board, and the second quarter of 2016 is coming to an end and we still don’t have a shipping product.”

    The chips have been around since 2014 with all kinds of people checking them out for use in real servers, developing software for real servers and of course, AMD. Folks with enough money and connections could buy them in 2015.

    Anandtech reported AMD promised A1100s for March 2014, complete with a developer’s board. I wanted that but it was ~$3K…

    The URI for the announcement from AMD states, “The AMD Opteron A1100 SoC has been in advanced development with technology partners and customers for several quarters and is available in mass production quantities today.” The URI contains 1/14/2015 while the announcement contains 1/14/2016. Clearly, AMD and “partners” were slow to launch but it’s definitely been around for years already. AMD announced the A-57 core about 2014 but this document lists it as January 2015 and later that A-57 would be available at 16nm. A1100 is really old tech, but it’s good enough for me.

  16. Wizard Emeritus wrote, “I don’t see any support for ECC RAM explicitly mentioned”.

    It’s an Opteron. All of them support ECC. In fact, on the site, it is written that it must be ECC. I don’t think that’s true. AMD64 and Opteron all have memory controllers built in the chip. That’s an innovation ahead of Intel which was using a separate chip until AMD did that.

    AMD A1120 Product Brief

    The 96Boards EE spec does not require ECC, but I think that it would be silly to omit ECC when using huge RAM on servers. I intend to hold a lot of data in there for weeks at a time and I don’t want it degrading. ECC is cheap insurance.

  17. Wizard Emeritus says:

    “It’s brand new manufacture of a reliable and tested product.”

    Actually Robert Pogson, it sure looks like the LeMaker Cello is actually nothing more at this point than a picture and a description. None of the “reviews” that are posted are based on hands on experience with an an actual board, and the second quarter of 2016 is coming to an end and we still don’t have a shipping product.

    Perhaps your enthusiasm is a bit, er, misplaced at this time?

  18. Wizard Emeritus says:

    “I’ll be able to have most of my files in ECC-RAM blowing away any advantage of SSD. ”

    Uh, Robert Pogson. I’ve been looking over the specs for the cello, and I don’t see any support for ECC RAM explicitly mentioned. Do you have a URL to where it says that ECC RAM is supported by the Cello board?

  19. DrLoser, knowing the price of everything and the value of nothing, wrote, “expensive piece of garbage”.

    It’s not expensive. It’s brand new manufacture of a reliable and tested product. I really don’t care if it’s not state of the art as long as it’s fast enough and reliable. All reports are that it works for people who have it and Lord knows AMD has had plenty of time to develop and test the chips.

  20. DrLoser wrote, “It’s the 16 GB two stick solution, isn’t it? Not that I would put it past you to be even more miserly and settle for two 4GB sticks.”

    Certainly not. I’m really short of disc-space at the moment and at times RAM is in short supply with 4gB. I want this to be future-resistant so instead of 16gB, I’m going for 32gB. RAM is pretty cheap these days. Besides the day to day needs, I’m going to want to play with some virtual machines that eat up RAM. Essentially, I’ll be able to have most of my files in ECC-RAM blowing away any advantage of SSD. I’m also looking at using 2X2TB in RAID-0 using XFS so I’ll have great speed compared to today for random-access stuff.

  21. Dr Loser says:

    To reinforce Dougie, btw, and I think I surprise both of us by doing so:

    What is the purpose of this server?

    In all my days, I am still surprised when I ask someone this and they cannot fathom a 10-second response.

    That, Dougie, is the best question asked so far on this thread.

  22. Dr Loser says:

    Dougman wrote, “How are you going to handle backups, duplication, etc..”

    Dougie is right on this point, Robert. After all, we are theoretically talking about a server. And I have to admit that everything Dougie said afterwards made a lot of sense.

    And incidentally you didn’t answer the question, predicated upon your purchase of a Cello.

    $200 for 32 GB of RAM with no hope of future expansion? Or $500 for 64GB, which by sheer dint of capacity might just about work for you?

    Or, and I am convinced that you are going to do this, because no matter what you say, your sole objective is to be as cheap as you can … 16 GB of RAM.

    Fess up, you cheapskate miser (one who can claim a $500 gain in a single day on his self-administered pension fund).

    It’s the 16 GB two stick solution, isn’t it? Not that I would put it past you to be even more miserly and settle for two 4GB sticks.

  23. Dr Loser says:

    Some people even are booting it …

    How very comforting. In my old age, should I ever reach the advanced old age that you have already reached, Robert, I can hope to indulge my senile fantasies by booting an otherwise pointless “servlet” board. Perhaps I can even spend my twilight years, divided between spinning up yet another Linux kernel and home-welding my broken cheap Chinese tractor substitute?

    Or maybe … actually, I can think of a million better ways to spend my retirement.

    Not all of them make sense. But, if I had to choose, buying one useless expensive piece of garbage like the Cello whilst dreaming of other useless pieces of garbage that can only just squeak past the boot phase … would not be quite the thing.

    Have you tried crochet, Robert? Crochet is about your talent level at the moment.

  24. oiaohm wrote about D03 at Estuary…

    Yeah. That would rock but nowhere on the page do they say who sells those, the price etc. It’s a setup where you can test software on such boards in a lab. D03 is 32-core, almost certainly above my pay-grade…

    UPDATE I dug around the Internet and found the chips are HiSilicon but the motherboards are Huawei. Here’s a picture of the DO-2.

    That looks pretty neat with two sockets for CPU and a bunch of connections for DRAM and SATA… That’s overkill for me but it shows the diversity of products out there. The DO2 is out there. Some people even are booting it but I have no idea how/where to buy one. The part number is CH02TEVBC and Google has ONE hit…

  25. oiaohm wrote, “40nm is out dated production”

    Beast I was 90nm. Beast II is 45nm. I sure would like to die with Beast running ~10nm, but I don’t think ARMed processors on mobos at such resolutions will be out soon for consumers. Smartphones have been getting all the love. With them becoming a mature market and servers by the thousands being the next target, DIYers like me may get a shot later this year or next. The best ARMed motherboards for servers are like $3000 this year. It’s supply and demand. I’m not seeing any supply and lots of demand.

  26. oiaohm says:

    dougman Raspberry pi 3 is 40 nm tech or basically 2008 generenation stuff. So either slow or in need of a heat sink of some form to perform decently then add on it stuffed width memory bus and it totally stuffed. Yes built on too tight of a budget.

    A1100 is 32nm so a little better but not great but decent width memory bus helps.

    http://open-estuary.org/d03/ I did point to a D02 for a reason. Its 14nm on the d02 boards so fairly power effective Yes the d03 is is DDR4 but still has the same problem of only have pci-e x8 slots that requires some mucking around to allow a normal video card in. Yes x8 is enough to turn over ati and nvidia graphics cards with a quite minor performance hits. Only rare cases with video cards are you in fact using the full x16 bandwidth..
    http://eshop.sintech.cn/flexible-pci-express-x8-to-x16-riser-card-p-609.html
    About 10 dollar converter.

    Issue with these boards is finding places selling them.

    Dr Loser –3) A “high-end” mobile phone Cortex chip– This is wrong. High-end mobile phone Cortex chip is 16-14nm production.
    http://www.anandtech.com/show/9330/exynos-7420-deep-dive Yes and that was last year samsung phones were 14nm.

    This is what is annoying me so much so production tech worse than a high-end mobile phone. 32nm is what you see in mid range to low end mobile phones that are old stock. New arm processors for phones are 16nm or better.

    Intel went 32nm in 2010. So 6 years behind production tech. Amd is only now starting to convert it fabs over to 14nm. 40nm is out dated production 32nm is not much better.

  27. Dougman wrote, “How are you going to handle backups, duplication, etc..”

    I plan on buying 3 2TB hard drives. Two will be RAIDed and one will be backup. LTSP is very efficient, cheap and easy. Why question it? I’m retired. I don’t want to maintain three or more systems when one will do. These clients boot from flash memory cards but can run sessions on Beast III. So, every client gets the speed of Beast for a fraction of the cost. I pay $300 + 3 X $40 instead of 3 X ~$100 for clients and whatever for a server. I save and get much better performance. Web applications work very well when they are on the same machine as the application is running. Zero network-lag. With all the RAM I will have, files cached require microsecond seek-times too. It’s all good.

    Beast runs 35 databases on MariaDB and a few on PostgreSQL. Some have not been used in a while but several are very active, like reloading, contacts, recipes and such. I use a few web-applications like phpMyAdmin, phpBB, Coppermine Photo Gallery and a few scripts I’ve written to find recipes I’ve collected from the web. We don’t put any of that on the web because of copyright and load on the server. MrPogson.com runs on a minimal configuration. Mostly Beast enhances desktops here. I used to have a snapshot of Wikipedia for schools but that got out of date and I dropped it. Same with Moodle. I won’t be teaching so there’s no need for my own copy.

    I use SSH to access Beast from any of the clients for system administration or to run my usual browser settings from any client. Beast out-performs any of the clients so this is helpful. I can surf from my LazyBoy in the living room for example more comfortably than in my den on a hard chair…

  28. dougman says:

    GNU/Linux terminal server, at home? For what??

    database, web, DHCP, TFTP, NFS, SSH. It’ll be mostly idling…..what about redundancy? Bonded Ethernet? No UPS?

    DHCP can be done by your router.

    Database?…for what and web? This website?

    TFTP/SSH, you remote into your server from where?

    How are you going to handle backups, duplication, etc..

  29. Dougman wrote, “What is the purpose of this server?”
    It’s a GNU/Linux terminal server for a few clients like Odroid-C2 and a few services for the LAN: database, web, DHCP, TFTP, NFS, SSH. It’ll be mostly idling. Beast has 3 clients at the moment, 240 processes, with load average 0.09, RAM used 3.6GB (including 1.5GB cached files), and about 1% CPU utilization, mostly me hunting, pecking and gawking. It’s old and uses too much power to do that. Time for a change. ARM64 and Debian are ready.

  30. Dougman wrote of a “Raspbery Pi3”.

    That is even a poor chip for phones. The A1100 has huge caches for instructions and data. The RP3 has tiny memory bandwidth. It’s useful but you can’t just plunk hundreds of processes on it and expect snappiness.

  31. DrLoser wrote, ” you’d be better off buying a Wintel laptop for that price”.

    Nonsense.

    • I don’t want to buy trouble.
    • The display on some laptops might use more power.
    • I do want ECC and huge RAM and storage, stuff that’s worthwhile. I’m willing to pay for that, not Wintel, core-gHz I’ll never use, malware, forced broken upgrades and that damned registry. The display of a laptop may never be used. The motherboard of the server is just a small part of the system. If it’s too slow I can buy something better next year and the RAM, storage, and networking are reusable.
  32. dougman says:

    http://www.pcworld.com/article/3041458/hardware/lemaker-to-release-a-raspberry-pi-type-board-using-amd-server-chips.html

    Hmmmm, I ran a overclocked Raspbery Pi3 for a little while, but the web was ungainly slow. NO way in hell, would I use that board for a server. My UNRAID NAS was built with a SUPERMICRO MBD-X9SCL-F-O and uses an Intel i3 processor which is far cheaper then this board and gives you WAY more performance.

  33. dougman says:

    What is the purpose of this server?

    In all my days, I am still surprised when I ask someone this and they cannot fathom a 10-second response.

  34. DrLoser wrote, “1) Two slots for RAM
    2) No DDR4 capability
    3) A “high-end” mobile phone Cortex chip”

    Again, 1, and 2 are interfaces, just the socket. The chip is able to do DDR4. The chip also has multi-megabytes of cache. It’s not a phone chip. Most phones I’ve seen can do one or two tasks for the user and one or two for the system, hence they only have tiny caches. This chip is designed to run a lot of little processes or a few big ones, just like a desktop user wishes. My server is a glorified desktop in that it provides the usual desktop applications plus some databases, DHCP, TFTP, NFS and local web-applications. The chip should be fine. I won’t be running FaceBook from it. The only heavy loads it will have are the biweekly kernel-builds which can be done in the background or late at night or on just one core. It’s all good.

  35. Dr Loser says:

    … crippled quite a bit by lack of interfaces …

    It’s crippled utterly, Robert. I believe we have gone through this in painstaking detail. The lack of interfaces is entirely uninteresting and irrelevant.

    You’re going to be ponying up $295 for a MoBo limited to:
    1) Two slots for RAM
    2) No DDR4 capability
    3) A “high-end” mobile phone Cortex chip

    Ignore the lack of everything else, and the fact that you could do far better by actually investing, say, $500 of those stock gains you brag about in a real server-lite that will afford you upgrades for the next five years or so.

    Why are you doing this? Out of stupid ignorant obstinate pride? Your nose must seriously dislike your face.

    Are we going to go for the two-stick 64GB RAM option at $500, or the two-stick 32GB RAM option at $200?

    Let me guess. You’ll be checking out the cheapest possible two-stick 16GB option, won’t you?

    In which case, you’d be better off buying a Wintel laptop for that price.

Leave a Reply