back to article Some Intel Core chips keep crashing, game devs complain

Some recent Intel microprocessors are crashing systems – and the problem appears related to the chips' firmware and clock rates settings. RAD, a maker of game development tools for Epic Games' Unreal Engine, on Wednesday published a note that describes decompression failures with its Oodle Data software and crashes in games …

  1. Duncan Macdonald
    Unhappy

    Not surprising

    Intel processors are being outperformed by AMD processors. To try to make the Intel processors look competitive they have to run the chips faster which cuts the timing margins. Instead of using speeds and power usage that are safe for every chip they are taking them to the bleeding edge of their capability - and like most equipment run at the bleeding edge there are failures. At least SO FAR the failures do not seem to be causing actual chip meltdowns. (But I would be uneasy about using them for critical calculations (financial or engineering) without down clocking them to increase the margins.)

    Icon for users who applications give incorrect results ===>

    1. cyberdemon Silver badge
      Devil

      Re: Not surprising

      Not just critical calculations. If the CPU is prone to taking two plus two and getting five, even if very very rarely, then it's bad news.

      Where's my pointer? Oh, it's at array index 101 with element size 16, so that's offset <bzzt> 2128 .. oops, wrong data.

      Bad if it happens in your application. Worse if it occurs inside a system call, to a filesystem or block device driver, say.

  2. Anonymous Coward
    Anonymous Coward

    Once Upon A Time........

    .....in 1981, Wordstar or dBASE-II ran perfectly well on my Osborne 1 and CP/M-80:

    - chip Zilog Z80

    - clock 4 megahertz

    - memory 64,000 bytes

    Granted that some modern graphics or mathematical processes will need a bit more oomph.......but I fail to see the need for huge processors for:

    - writing a document or a python script

    - writing and sending an email

    ......where a Z80 can probably get the job done!

    I know......I really need to get with the program.......................

    1. An_Old_Dog Silver badge

      Re: Once Upon A Time........

      Any of the personal computers of that era which had decent keyboards and at least 64-column displays (TRS-80, Commodore-64, Apple II [with Videx video board], CBM-80, Atari 800, S-100-based computers with good serial terminals, etc.) were just fine for word-processors, databases, and spreadsheets of that era. I did a lot of useful work on my C-64.

      What's changed is both the nature of the data we deal with, and how we deal with it. The giant documents required for businesses applying for government contracts won't fit on a floppy diskette. (I've not heard of a microcomputer word processor which allowed documents to span multiple floppy diskettes.) These days, we casually sling around multi-gigabyte datasets, and use Perl to read the entire thing into RAM, for programming convenience. Newer processors are needed to (a) address increasing amounts of RAM, and (b) handle much-much-much faster internal data transfers and disk I/O. Manipulating large, complex graphic images, or video files, or audio files, takes CPU speed, copious RAM, and fast disk I/O.

      Modern video games demand the most of everything which our systems can provide.

    2. iron

      Re: Once Upon A Time........

      As much as I love the Z80... By your logic there was nothing wrong with a slate and chalk, you should be using them!

  3. Pascal Monett Silver badge

    "He observes that Intel seems unlikely to improve its verification story"

    Of course. AMD is now eating Intel's lunch. Now is not the time to do proper engineering procedures.

    Now is the time to boost the Ghz and sell ! sell ! sell !

    Intel will leave engineering to wait for when it has become the founder of the West. Then Intel will have time for that, especially if AMD comes, hat in hand, asking for production wafers.

  4. Anonymous Coward
    Anonymous Coward

    Disable Turbo Boost

    Yes, you’ll be leaving performance on the table, but you’ll be making your PC run far more reliably. Overclocking is bad, even if it’s a vendor-driven affair.

    1. GrumpenKraut
      Mushroom

      Re: Disable Turbo Boost

      You get like 15% more CPU performance for like 50% more energy. And 15% CPU power does not usually translate into 15% overall performance. On top of that, benchmarking becomes close to impossible. I have yet to see a compute cluster where any of this boost crap is enabled.

      And don't get me started on that hyperthreading dreck.

  5. Omnipresent Silver badge

    Time for an upgrade

    Oops, sorry you all bought those super shiny new core processors that were supposed to be speed demons everybody. AI slowed them to a crawl, and Windows 12 needs a newer processor. Everybody buy new hardware....

  6. Pete Sdev
    Windows

    Same old

    Those of us of a certain age will remember the adage "Don't divide, Intel inside"

    1. An_Old_Dog Silver badge

      Re: Same old

      I'd heard it as, "Can't divide, Intel inside."

      1. Phones Sheridan Silver badge

        Re: Same old

        I remember it as:-

        We are Pentium of Borg. Division is futile. You will all be approximated!

    2. Anonymous Coward
      Anonymous Coward

      Re: Same old

      Division is futile, you will be approximated.

  7. Jou (Mxyzptlk) Silver badge

    Golem.de looked closer than ElReg...

    As the German Golem article pointed out with their actual testing:

    Many BIOSes, like the one shown in picture two in the linked article, set the "Short Duration Package Power Limit" to 4095 Watt. The "long duration" was >100% above the Intel recommendation too. Result: CPU get ONLY throttled by thermal limit. But see pic 4 with corrected values, and ZONG, suddenly stable!

    I am not an Intel fan for a lot of reasons, but this time it is not Intels fault (alone).

    1. herman Silver badge

      Re: Golem.de looked closer than ElReg...

      200 Watt should be enough to heat your home…

      1. Chris Stephens

        Re: Golem.de looked closer than ElReg...

        1500 modern plastic heater watts is not enough for a space heater to heat one room these days..

        1. Jou (Mxyzptlk) Silver badge

          Re: Golem.de looked closer than ElReg...

          That depends on the room whether 1500 Watt is enough or not. After some rather optimizations I need a lot less heating. But it depends on many other factors what works on other homes.

        2. Sorry that handle is already taken. Silver badge

          Re: Golem.de looked closer than ElReg...

          1500 W of resistive heating is 1500 W of resistive heating no matter what it comes from.

          (apologies if I've missed a joke)

          1. Peter Gathercole Silver badge

            Re: Golem.de looked closer than ElReg...

            I think that you may find that some restive load heaters have thermal controls to prevent them getting hot enough to melt the enclosure. So while the device is rated 1500W maximum, if you were to run it for an extended time T, the overall heat output would be significantly less than Tx1500W.

            These controls masquerade as air temperature thermostats, supposedly to not keep consuming electricity when the air temperature reaches a set level, but also to prevent the device putting the full whack of heat out so it doesn't melt.

        3. Killfalcon

          Re: Golem.de looked closer than ElReg...

          Maybe they were thinking of a 200W heat pump, but likely not, as that's roughly equivalent to a 700W resistive heater.

    2. Kevin McMurtrie Silver badge

      Re: Golem.de looked closer than ElReg...

      The only thing sillier than 4095 Watts is trying to do that at 1 Volt, 4095 Amps.

  8. Chris Stephens

    Power supply noise.

    Noise on supply voltages inside the chip can lead to all sorts of errors.

    CPUs are CRAZY current bursty. This wide bandwidth low frequency rumble is crazy to look at on the various supply lines at the chip. While voltage regulators MOSTLY handle this noise, its much more of a impedance issue across a wide freq range below 100hz then a DC supply that feeds a chip. These supply lines need to be looked at as low freq AC rather then simple DC..

    Motherboard makers typically follow "reference designs" for power supply bypass caps and regulation / current requirements. A "reference design" is the least expensive way that works. It is not the BEST way. So you see the same circuits and even board layout across various boards and board makers for a specific socket.

    Each CPU is different in how its wired internally power wise. Each CPU has a different current spectra profile. This current spectra of course varies in realk time depending on what it is doing.

    The amount of actual noise on the power input pins can be pretty impressive that it can work at all.

    So its NO surprize that some CPUs, under some power conditions, running specific things, can glitch.

    What makes this a more complex to engineer is that modern test equipment like a $100k Tektronix MSO scope really does not do well looking at noise. Digital scopes are all about statistics. Averaging. Glith capture. They are alll designed around capturing a SIGNAL and excluding noise. So when you get a truly random signal its much harder to look at using modern gear. To see these signals WAY better you gotta go back to the 70-80s and get a Tektronix 7904 analog scope. You can imagine modern motherboard makers and CPU designers dont have a restored 7904 on thier bench. BUT THEY SHOULD..

    Plus modern engineers rely on computer modeling too much and soon AI..

    So really the issue is instrumentation in R&D.

    The solution to this issue tho is pretty easy. WAY WAY more super low ESR caps RIGHT at the CPU. So rather then a .1uF, a 33uF monolithic ceramic or more. Per pin.. Then a bit back from the socket, a giant tant cap, like 2200uF.. All this after the regulator and before the chip.

    Now this issue might indeed be in the chip in how it handles the power lines on board. But cleaning up the power coming in can only help..

    A digital system is really just a analog circuit running a few voltages that are on/off.. Modern engineers forget there is a VAST analog world that makes thier world go around.

    1. David Hicklin Silver badge

      > The solution to this issue tho is pretty easy

      Get the power consumption down, then you need a less beefy power supply.....I mean...4kW peak loads...sheesh....

    2. Steve Hersey

      Everything is analog if you look closely enough.

      The trouble with reference designs for crazily complex CPUs is that it's rather difficult for anyone but the chip maker to effectively optimize the bypassing scheme, as only the maker really knows where the critical points and needs are.

      I'd be surprised if the motherboard makers' engineers get much leeway for comparative design testing when churning out the board for the Latest Sexy Chip. There must be incredible pressure to just use the reference design.

  9. Anonymous Coward
    Anonymous Coward

    ahh the new intel

    it's been like that since at least 13th gen, when they seemed to decide to just go balls to wall and MB makers followed them putting in stupid high power limits as default

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like