Intel: That Other OS is a Slow Dog

No, Intel did not say that but they did implied that other OS on ARM will give a disappointing experience.
“In Salvator’s view, ARM may be just fine for such consumer experiences as those provided by Apple’s iPhone or any one of a squillion Android smartphones, but at low power-consumption levels it can’t have the oomph to provide what customers expect when it comes to a Windows-based machine.”

I think M$ and partners will compensate for that by developing more complex ARMed devices but they will then be more expensive and power-hungry. Of course, GNU/Linux on ARM already gives a great experience.

With the biggest “partner” of Wintel criticizing the developments of the mother-ship, how long can the Wintel monopoly last? Intel will find that GNU/Linux runs well on Intel as on ARM and Intel will find it profitable to be partner of GNU/Linux. It’s all good.

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. Bookmark the permalink.

9 Responses to Intel: That Other OS is a Slow Dog

  1. Ray says:

    They already exist ๐Ÿ˜€

    http://www.seagate.com/www/en-us/products/laptops/laptop-hdd

    The speed of an SSD, with the Capacity of a hard drive.

    Too bad it’s only available with Seagate ๐Ÿ™

  2. We are not likely to see those on consumer-PCs anytime soon, eh? Even business will not likely deploy those widely. They will install them on servers where they will earn their pay. My favourite supplier does not even sell them although he caters to geeks.

    see Cheetah 15K datasheet.pdf

    Seek time is 2 milliseconds, so you can really move data with it. To boot or load an application and all its files, you need to seek a lot. At 2 milliseconds a seek, the machine is idling while it waits for a file to be transferred.

    SSDs do change the game but they are still too small or too expensive for widespread use. That will change in a year or two. They do make great caches for the spinning drives.

  3. oldman says:

    Pog:

    “CPUs have radically out-performed storage.”

    With the advent of affordable solid state disks, the gap is now most of the way to erased. My next system is coming with a 256Gb solid state disk. the cost to add it was not that high, and this cost will go even lower as time goes on. In fact I believe that spinning disk is well on its way to becoming obsolete.

    “These days, a 500gB SATA transfers at 120 MB/s continuously, an increase of 12X”

    You are a bit behind Pog. 15K SAS 2 disks have a throughput of almost 600MB/s or 60x your base, and they and the controllers to support them are alls becoming common and affordable.

    “I can give that huge performance increase to processes of a user sitting at that old system and, with his files cached in RAM, his performance is far superior to what any reasonable thick client can give from the latest drive”

    But you can only deliver the Linux only variant of FOSS from your system and carefully chosen FOSS at that. I guarantee you that were you faced with presenting your “solution” to a population of content creators with specific application requirements and who were used to having dedicated resources at their command, you would probably have few takers.

  4. oldman wrote, “Your statement fails to take into account the reality of the difference in quality/Performance of peripherals. No manufacturer is going to put high performance peripherals in a low end system like the ones that you describe.”

    CPUs have radically out-performed storage. Ten years ago, folks used ATA33 with sustained transfers of 10MB/s. I occasionally meet one of those. Quite sad. These days, a 500gB SATA transfers at 120 MB/s continuously, an increase of 12X. Meanwhile, CPUs and motherboards have gone from 32bits/1gHz to 64bits/4gHz quad-core, an increase of 8X. That supports oldman’s argument but the seek-time has barely changed (dropping from 14 ms to 8ms average), negating it all (SSDs do better). In the meantime our software systems ask our PCs to seek more files to boot or to open an application. This is one of the main reason I like GNU/Linux terminal servers. I can give that huge performance increase to processes of a user sitting at that old system and, with his files cached in RAM, his performance is far superior to what any reasonable thick client can give from the latest drive. Even with 4X RAID, the thick client loses. Even SCSI does not overcome the problem of seeks. A millisecond is so tedious compared to sub-nanosecond access in RAM. The storage devices give the same seek-times for an ARM or a Xeon etc.

  5. oldman says:

    “An idling Atom CPU does not run any slower than an idling quad core fire-breathing dragon. They both wait on I/O.”

    Your statement fails to take into account the reality of the difference in quality/Performance of peripherals. No manufacturer is going to put high performance peripherals in a low end system like the ones that you describe.

    On a properly sized system and configured system running software that does more that allow one to “point click gawk” a Nehelem will blow doors of an atom CPU.

    “I think that comment was an Intel-salesman doing what salesmen do, upsell”

    Perhaps, but it does nonetheless match experiences that I have had.

  6. I have seen a few PCs running Atoms. One was a netbook. It did not run at all because XP had died. We put Debian GNU/Linux on it and it ran very well, far faster than XP according to the owner. One was an eeeBox from ASUS. It was a small cute box that bolted on the back of an LCD monitor. Ran like a rocket with XP or ASUS quick-start GNU/Linux. The third was a multimedia PC in a living room. The motherboard had died and an Atom mini-ATX board replaced it. The thing works very well with GNU/Linux. So, I have seen three very different Atomic systems and they all ran pretty well. I think that comment was an Intel-salesman doing what salesmen do, upsell. An idling Atom CPU does not run any slower than an idling quad core fire-breathing dragon. They both wait on I/O.

  7. oldman says:

    “The most compute-intensive stuff folks do with PCs is decoding video usually. ARM puts a whole core to work at that and they still have one or more cores ready to do other tasks”

    Even if your assumptions are true, there is no reason to think that the manufacturers who will be putting their livelyhood on the line will see it that way. I take you back once again to another portion of the same interview that you did not quote:

    “…when asked if ARM could be an Atom competitor in the lower rungs of the PC market however, he had a story to tell.

    “Let me give you one quick anecdote,” he said, which he titled Aventures in Mispositioning. “When we first shipped Atom, a couple of clever system makers thought that they could basically take a chassis like that one” รขโ‚ฌโ€œ he pointed to a plain-vanilla 15-inch laptop รขโ‚ฌโ€œ “put an Atom inside it and call it a notebook, and sell it, and try to capture a premium for it sort of between the typical netbook and notebook price points. And they really kind of patted themselves on the back and said, ‘Aren’t we clever’. Until the returns started. Because people brought it back, saying: ‘You know, I bought this expecting the full-fledged PC experience, but I didn’t get it’.”

    As far as your demonstrations are concerned. What some hacker does with their phone or on hardware that no non technical user would ever see for sale, does not impress me. I’ve done hacks likethis myself at several points in my career, and they did work, but at no point did I kid myself that they were viable.

  8. “negate” is the wrong term. “reduce” would be better. The most compute-intensive stuff folks do with PCs is decoding video usually. ARM puts a whole core to work at that and they still have one or more cores ready to do other tasks. There is no need to do much more than they have done already to increase performance of ARM except advance along Moore’s Law.

    Here’s a YouTube video of OpenOffice.org loading in 45s on a Nokia N900: slow but usable

    The N900 has a CPU with ARM Cortex A8 600 MHz, PowerVR SGX530 graphics and 256 MB RAM. According to ARM that processor uses less than 300mW and runs performance-optimized consumer applications requiring 2000 Dhrystone MIPS. The chip is less than 4mm^2 at 65nm. What do you think the performance will be at 20nm? They already have A9 CPUs with 10000 Dhrystone MIPS in less than 2W with graphics processor included @2gHz.

    Further, the performance with OpenOffice.org is more likely limited by RAM throughput/SSD throughput than CPU. Just loading the application is limited by some bottleneck unrelated to CPU performance.

    Here’s an A9 doing video.

    Here’s an A9 at 500MHz competing nicely against an Atom (with video acceleration) at 1.6gHz”

    Here’s an A9 playing 3D games on a smart-phone

  9. oldman says:

    “GNU/Linux on ARM already gives a great experience.”

    Are you kidding Pog? Some hobbyist shows a stripped linux configuration and running xfce4 desktop on top of what looks like a developer prototype board. There is no demonstration of anything running, nothing but a few screens showing snapshots of the boot process.

    Where is the demonstration of experience Pog? Can you show me the experience of Openoffice running? of Firefox, of thunderbird.

    The interesting thing is that in the part of the Salavator interview you don’t quote he says the following:

    “From Salvator’s point of view, creating an ARM-equipped device that could compete with a PC would negate ARM’s power advantage. “For them to really come and do anything serious that would go into into a bona fide PC device and deliver a full-fledged PC platform, they’re going to have to do some different things … like considering things like an out-of-order processing execution engine, going multi-core, and cranking up their clocks.”

    Sound familiar?

Leave a Reply