Report: Skylake-X CPUs prove hard to cool thanks to TIM change

Intel's launch of its Core X high-end desktop CPUs and the accompanying X299 platform has seen some controversy in overclocking enthusiast circles. The mish-mash of architectures, core counts, memory configurations, and PCIe lane allowances on the X299 platform has been the source of confusion ever since Intel's announcement, and complaints about apparent high system power draw and underspecced voltage regulation on X299 motherboards started soon after demanding overclockers got their hands on the first wave of boards. Igor Wallossek at Tom's Hardware rigged up a Core i9-7900X CPU and an MSI X299 Gaming Pro Carbon AC motherboard for a closer look at what happens when that gear is pushed to the limit.

The main takeaway is that cooling Intel's Core i9-7900X is a tough task even at stock clock speeds, even with a high-end Alphacool Eiszeit 2000 refrigerating cooling system. The site recorded a 71-Kelvin difference between the chip's core temperatures and the temperature at the top of its heat spreader under load, even with that exotic cooler at work. Tom's Hardware attributes this behavior to Intel's decision to abandon the practice of soldering the integrated heat spreader (IHS) onto the CPU die in its range-topping desktop chips. That move could be seen as disappointing in light of the high price tags attached to most members of Intel's Core X CPU lineup, and when considering the fact that (to our knowledge) rival AMD solders the IHS onto every Ryzen chip it produces.

The sheer power dissipation of the i9-7900X is another obstacle to proper cooling. The chip dissipates 160W under a Cinebench load using Tom's Hardware's testing configuration, and running the Prime95 Small FFTs torture test pushes those figures to 230W. The Cinebench load pushed VRM temperatures to 73° C, while Prime95 Small FFTs caused the VRMs to reach about 90° C. For reference, the site reports that its MSI motherboard will begin throttling the system when the thermistor near the VRMs hits 105° C. In either case, the site recommends at least some kind of beefy all-in-one liquid cooler to properly cool the chip. Even with Cinebench, Tom's observed thermal throttling from the i9-7900X with an air cooler on top.

Overclocking the chip only worsens these challenges. The i9-7900X dissipates nearly 250 W under the AVX-enabled Prime95 at a 4.5 GHz overclock. A test at 4.6 GHz running LuxRender again resulted in a 250 W power draw. In this scenario, the CPU started throttling before the VRMs even got toasty. The site soldiered on and pushed the 7900X to a 300W power draw anyway. After 10 minutes, the VRMs reached 105° C and the system went into a thermal throttling loop.

The range of X299 motherboards did not escape Wallossek's ire, either. He complains in the article about flaws in every board manufacturer's X299 wares, "including faulty P-states, incorrect Turbo Boost frequencies, and so much more." He goes on to detail inconsistencies in the way the MSI X299 Gaming Pro Carbon AC handles Turbo Boost and how these inconsistencies lead to higher-than-expected power consumption. While we're still developing full impressions of the X299 platform, we doubt our testing will be free from similar challenges. Overclockers and system builders alike will need to be fully aware of the demands of Skylake-X when planning an X299 build.

Comments closed
    • credible
    • 2 years ago

    Not sure anyone has mentioned it but so much focus on the gamer/enthusiast but what about the regular users of which there is a larger percentage that never clean out their cases.

    Could the dust bunnies not wreak havoc on these cpu’s.

    • hkuspc40
    • 2 years ago

    So happy I bought my 4790k when I did… The CPU market is a mess right now… I think I’m just going to continue to bide my time till all this gets sorted out. 😉

      • bfar
      • 2 years ago

      Any of Intel’s four core chips going back to sandybridge were decent buys at the time, at least from a gaming perspective. Those 6 core Ryzens are a bit of a shake up in fairness.

    • abiprithtr
    • 2 years ago

    Good news for Intel fanboys! The article says “Igor Wallossek at Tom’s Hardware [u<][b<]rigged[/b<][/u<] up a Core i9-7900X CPU and...". So, the tests were deliberately rigged by Tom's! Tom's hardware had been paid by AMD using their profits! (Gets up from the floor after 5 minutes). Conspiracy theory confirmed!! (Goes to nearest friendly retailer to buy a Skylake-X)

    • Wirko
    • 2 years ago

    Some extrapolative thinking has led me to believe that the [url=https://techreport.com/gallery/32111/intel-core-i9-7900x-cpu-reviewed-part-one/93943/mysterychip-hires<]mystery chip[/url<] is a delid alarm.

      • Krogoth
      • 2 years ago

      No that’s a RDIF chip for inventory and tracking purposes.

    • jihadjoe
    • 2 years ago

    Actually, I wonder if the use of TIM happened so they could rush Skylake-X to market. I mean, squirting a gob of goop onto the die is probably a lot faster than soldering it to the IHS.

    • Major-Failure
    • 2 years ago

    Even without soldering, how bad does the TIM have to be for a 71-Kelvin difference between core and top of heat spreader? Sounds like they’re using mashed bananas as TIM.

    Thank god AMD is back and it couldn’t be at a better time either.

    • Jigar
    • 2 years ago

    Officially the worst News ever for chuckula

      • NeelyCam
      • 2 years ago

      Or the best news. Gives him a [s<]reason[/s<] excuse to get one for delidding and testing

      • chuckula
      • 2 years ago

      What are you talking about?
      I already bought three Threadrippers for the TR BBQ.

    • ronch
    • 2 years ago

    So I guess Intel’s fabrication technology is also subject to the limits of physics after all..

    • Krogoth
    • 2 years ago

    Nah, the real issue that physics has been catching up for a while. Intel grossly underestimated the returns from their 14nm process and the ride isn’t going to get any better. Normal Skylake and Kaby Lake chips get toasty fast if you push them hard. It shouldn’t really come to a surprise that a scale-up version would run into thermal issue fast if you crank up the clockspeed.

    AMD isn’t immune to the issues to the either. The 16-core Threadripper chips are going toasty if you intend on overclocking them and are going to more limited by architecture if you want to go beyond 4Ghz.

    7960K-7980K are going to be flops at launch. Intel shouldn’t even bother with them and just stick with 7900K until the next massive architecture change. Going beyond 12-cores is excessive for a HEDT/workstation platform. It only make sense if you are building a server of some kind.

      • synthtel2
      • 2 years ago

      SKL-S seems to usually be able to handle up to 1.3-1.4 W/mm[super<]2[/super<] without delidding becoming too critical, and that's including the area taken by graphics (off to the side) which may or may not help much with this case (I don't know). SKL-X as shown here is a total disaster by the time it reaches 1 W/mm[super<]2[/super<]. Alternately, a 6700K's stock power density (going by TDP) is 0.75 W/mm[super<]2[/super<], and it's still got substantial unmodded headroom. A 7900X's is 0.43 W/mm[super<]2[/super<], and even with an unobtainium cooler the headroom is sketchy. That isn't physics catching up, that's Intel managing said physics badly.

    • Kougar
    • 2 years ago

    Just wait until people start overclocking the 7920X and other HCC chips. Over a third of the 7920X is dark silicon, and with a larger die area it’ll go higher before throttling.

      • chuckula
      • 2 years ago

      There are rumors that the LCC chips will still be soldered too, although only time will tell.

        • Kougar
        • 2 years ago

        You mean the HCC chips? It would be surprising to me if Intel decided to swap to TIM for its Xeons, so maybe you’re right. Is a pretty penny to pay for a few extra cores and solder though.

        I am also wondering if most waterblocks & HSF coolers are going to have bases located directly over all the cores… most waterblocks use internal fins or structure directly over the die, so any waterblocks for HCC chips are already going to lose some performance. Swiftech is promising Sky-X and Threadripper specific waterblock layouts but it will be interesting to see how it plays out.

    • Wirko
    • 2 years ago

    [quote<]The site recorded a 71-Kelvin difference between the chip's core temperatures and the temperature at the top of its heat spreader under load, even with that exotic cooler at work.[/quote<] At equal load, no cooling system can increase or reduce this difference. Not even an exotic heatsink made of exotic wood (and yes, I still want one).

      • Chrispy_
      • 2 years ago

      What you say is true, but would you not also agree that a 71-over-ambient temperature is pretty ridiculous regardless of the cooling?

        • Wirko
        • 2 years ago

        Pretty ridiculous (to me)
        Ridiculous but not pretty (to those buying a SKL-X CPU)
        Not pretty and not funny (for the EA who already bought a SKL-X CPU)

    • jcw122
    • 2 years ago

    Sounds like PrescHOTt 2.0. High clocks with high temps.

      • maxxcool
      • 2 years ago

      Not exactly. Prescott suffered from to many repeaters that upped the draw from the socket and never turned off. it was pure architecture of the silicon. This failure is good old stupid.

    • Axiomatic
    • 2 years ago

    Got it, skip Skylake-X.

      • Dudeface
      • 2 years ago

      Yup. Was going to get SKL-X and a 1080 Ti. Now I think I’ll just wait for Coffee Lake and Volta.

      • Klimax
      • 2 years ago

      Looks like I’ll have to go into that camp. Pity, I was looking forward to AVX-512 and large L1d ports.

    • The Egg
    • 2 years ago

    I think we’re all being a little too rough on Intel. It’s important that they save 63 cents on their processors that cost $1,000 minimum.

    • Chrispy_
    • 2 years ago

    [b<]FFS Intel.[/b<] Sure, you can get away with cheap TIM on your mainstream CPUs but for the love of all that is sacred, don't ruin your outrageously expensive enthusiast CPUs with $0.08 corner-cutting like you have at least twice in the past. I mean, you'd think Intel would learn by now, right? Are they stupid or something? Honestly, I always want healthy competition but I seriously hope Ryzen-based chips tear Intel a new one this coming generation, like they did in the Pentium 4 days. God, they deserve it because they treat us chumps like crap most of the time....

      • jodiuh
      • 2 years ago

      Didn’t we go through this with Haswell? I had three 4670K’s. One was defective from the get go, the other two could barely make 4Ghz and ran very hot.

      Then they made a big hullabaloo about Devils Canyon w/ it’s improved TIM. Bummer for enthusiasts. :/

        • Chrispy_
        • 2 years ago

        Ivy was the first awful TIM and everyone pretty much confirmed that the K-series were being outperformed by the previous generation due to TIM limitations. Haswell was worse and it was so bad that Intel effectively re-named the product – Devil’s Canyon was nothing other than vanilla Haswell with better (but still cheap polymer-based) TIM!

        Having learned their lesson they then used the Devil’s Canyon TIM in Skylake rather than the toothpaste they were using in Haswell and Ivy, before inexplicably going back to toothpaste with Kaby, despite kaby being a 14nm chip and thus even more sensitive to poor TIM than the last time they cheaped-out on Haswell at 22nm.

        They have been using solder in the Sandy-E, Ivy-E, Haswell-E and Broadwell-E models to date, for good reason – when producing a $1000+ product, don’t ruin it by choosing $0.05 TIM instead of $0.12 TIM or $0.50 solder. What they’re doing is effectively flipping the bird to buyers. They want you to de-lid and invalidate your warranty, because they know you’ll have to.

          • RAGEPRO
          • 2 years ago

          I mean, it’s never been about the TIM quality. The stuff Intel used, even on Ivy Bridge, is very high quality. It’s just a matter of thermal paste being actually quite awful at transferring heat. If you compare the thermal conductivity of metal-on-metal or silicon-on-metal versus even the best thermal pastes in the world it’s a factor-of-hundred difference. Thermal paste is just no substitute for having the cooling surface directly against the heating surface; it’s intended to fill micro-gaps that would otherwise be filled with air, not to be the primary heat transfer medium.

          Intel did indeed say that they were improving the quality of the thermal solution for Devil’s Canyon but it didn’t really make any difference. Delidding the 4790Ks was so popular and difficult that there are a bunch of aftermarket devices for doing it. Intel knows using paste instead of solder is bad, they obviously know. There really is just no way to see this as anything other than a shameless and blatant anti-consumer move from Intel, and I’m not one to say things like that lightly.

            • ermo
            • 2 years ago

            Why don’t intel just use something similar to Coollaboratory Liquid Pro on each and every CPU?

            That would bring the thermal conductance up by quite a bit, wouldn’t it? If intel already has a good automated TIM application process running, switching paste ought to be (relatively) simple?

            Arctic MX-4 is 8.5 W/mK. CLP is 84 W/mK. An order of magnitude is nothing to sneeze at, and since solder is also electrically conductive, that’s not really a good argument against liquid metal thermal paste, is it?

            So why doesn’t intel use liquid metal of *some* sort as paste…?

        • maxxcool
        • 2 years ago

        weird. My 4690k runs all day folding and never breaks 70c …

          • Chrispy_
          • 2 years ago

          Of course it does, that’s Devil’s Canyon.

          DC was basically a Haswell refresh where the only thing they refreshed was the TIM and IHS to stop the overheating issues.

          EDIT!
          Sorry, it wasn’t just new TIM, they added 8 tiny capacitors to the bottom too, to prop up their dubious (and then abandoned in Skylake) on-chip integrated voltage regulators. Another thing that could have been avoided if they’d just spent an extra $0.02 per CPU in the first place :\

      • anubis44
      • 2 years ago

      “I mean, you’d think Intel would learn by now, right?
      Are they stupid or something?”

      Yes. Yes they are stupid. And arrogant. They thought they ruled the x86 universe, but Mr. Keller has proven otherwise. I did say he was the ‘Luke Skywalker’ of CPU design, back in 2012, but of course, everybody laughed at that; said it was ‘impossible’ for one man, helping a company with such limited financial resources, to EVER compete with the gigantic, invincible Goliath that is Intel. Well it wasn’t.

      I’m now laughing all the way to the bank with my AMD shares which have quintupled since March of 2016 (yes, I’ve sold and bought them back again several times over–AMD’ll probably hit $20/share by the end of this year. Within two years, I believe it’s a $45/share stock) and rocking to two Ryzen 7 1700 systems, one of which is in my HTPC because it only draws 65W of power, and I thought ‘what the hell! They’ve more than made me enough money, I’ll splurge on the HTPC upgrade! Maybe I’ll just transcode videos on it instead of my desktop!’

    • tipoo
    • 2 years ago

    Still with the TIM issues…I’d rather they just solder them like AMD is now.

      • brucethemoose
      • 2 years ago

      What’s the official excuse anyway? Wasn’t it something about the die being too thin?

        • xeridea
        • 2 years ago

        It costs slightly more to solder than to use TIM, so they gotta save $0.50 on a $1000 CPU.

          • brucethemoose
          • 2 years ago

          No, I mean what is Intel PR’s official excuse?

        • swaaye
        • 2 years ago

        Well there is this:
        [url<]http://overclocking.guide/the-truth-about-cpu-soldering/[/url<]

          • brucethemoose
          • 2 years ago

          Great read.

          The article mentions that the problem is magnified by small dies though. Skylake-E is bigger than Broadwell-E, yet it’s fully soldered and is made on (basically) the same process, and it seems to have held up just fine.

            • Krogoth
            • 2 years ago

            It is mostly due to cutting cost and long-term reliability issues. I only see it being really an issue with laptop processors though due to their thermal cycles on their platform.

            Overclockers have been torturing their soldier-chips for years now and there haven’t been incidents with “cracking” which leads me to believe more to that Intel went away from solder to cover their behinds on “what ifs” as seen with Nvidia’s “Bumpgate” debacle while reducing bottom line costs.

            • brucethemoose
            • 2 years ago

            Reliability concerns would have to be on the server end, since these desktop chips are just the leftover scraps.

            Thermal cycling shouldn’t be a problem for them though, right? They aren’t getting turned on/off constantly, all they do is alternate between “warm” and “hot” when the load changes.

            I guess it could be pre-emptive, like you said. Maybe an incident or 2 with big customers is motivating the change.

            • swaaye
            • 2 years ago

            Mobile CPUs rarely have an IHS or use solder.

            I imagine part of the reason there haven’t been failures from cracking die is Intel validated it very extensively and know what they are doing. From the complexity of the process to solder die to IHS, and cost of materials, yeah cost reduction seems quite plausible.

            And really it only matters to overclockers in the end. It’s just a shame that they can’t somehow eliminate the gap between die and IHS that makes the paste so ineffective.

            I did the liquid metal paste change out and that is extremely effective. I do wonder what the long term reliability might look like there though. It seems too good to be true, and AMD and Intel don’t use that approach ….

            • blahsaysblah
            • 2 years ago

            I thought it wasn’t the TIM that was the issue, it was acceptable(inside, not the one on standard Intel cooler, i just swapped some other day from brand new i7-7700 mini-stx build and got 10C+ drop under stress load).

            What folks get when putting liquid metal TIMs is only a few degrees versus something like Noctua NT-H1, its rather the gap on the IHS and subtrate is vastly reduced from Intel’s sealant and that also reduces the gap between IHS and CPU. Thats were the difference comes from and why a Noctua or Kryo(not recommended, its for water cooling, bad above 80C) paste vs a Liquid Ultra/Pro liquid metal…

          • Spunjji
          • 2 years ago

          The author fails to justify why (1) we should give a damn about an issue that occurs when cycling from -50 degrees celcius to 125 degrees and back again, and (2) why their claim that the change to do with die size doesn’t bear out either with older, smaller dies and newer, larger ones.

          That second part is plausible but fails the smell test. It’s clearly about cost, and sure the cost and complexity is higher than most people think it is, but on a $1000 CPU… no, no thanks.

    • chuckula
    • 2 years ago

    Some of this has to do with MSI apparently applying a 1 GHz all-core factory overclock to the 7900X:

    From TFA:
    [quote<] Since Prime95 does not output any results, you need to be a bit more critical about where the differences in power consumption come from. To start, MSI has a BIOS entry called Enhanced Turbo, which is set to <Auto> by default, meaning it's active. In order to find (and disable) it, you have to open the Advanced BIOS menu and activate the Expert mode. Turbo Boost normally allows two cores to hit 4.3 GHz in lightly-threaded tasks, but with this switch in its default position, all cores are automatically pushed to 4.3 GHz under load. This is a nice factory overclock, which is of course stabilized by an automatic voltage adjustment.[/quote<] Something tells me that the power consumption numbers of other chips that receive a 1 GHz all-core overclock that probably isn't optimizing the voltage input to boot (at 4.3 GHz most 7900X parts are quite happy with 1.15V or even less) aren't going to be all that amazing either. [FYI: If you read Tom's article they note that Asus boards, like the one TR used in its review, don't appear to apply the auto-overclock like that MSI board did. Additionally, as TR has noted, the actual power efficiency of the 7900X under a real-world heavy workload was quite similar to the 1800X. Phoronix has additional results where the 7900X is actually substantially ahead of the 1800X in real-world power efficiency in most CPU-bound tasks: [url<]http://www.phoronix.com/scan.php?page=article&item=skylakex-thermal-pow&num=1[/url<] ]

      • cegras
      • 2 years ago

      Why not quote the rest?

      [quote<]At 4 GHz with Enhanced Turbo turned off, our CPU sucks down an astonishing 230W. The 4.3 GHz setting imposed by Enhanced Turbo mode looks a lot more conservative in comparison at 160W (with peaks just over 230W) under Prime95, curiously enough. This can be explained when you see Enhanced Turbo mode pulling clock rates back more sharply in AVX-heavy workloads. Rendering with Cinebench, which doesn't utilize AVX instructions, changes this picture dramatically. The 4 GHz configuration consumes 145W for a score of 2169, while every core running at 4.3 GHz via "Enhanced Turbo" hits 190W for a modest jump to 2312 points. That's worrisome. You get a 6.6% performance benefit from 7.5%-higher clock rates at a cost of 31% more power consumption. Talk about a poor bang for your power/heat/cooling buck.[/quote<] And none of what MSI did has to do with the TIM bottlenecking heat transfer out of the chip.

        • chuckula
        • 2 years ago

        OK. So a 700MHZ 10-core overclock under non-AVX-512 workloads is basically running at the stated TDP number of the chip. In Prime95 that is apparently using AVX-512 (or is it? Tom’s can’t be bothered to give the information) under an overclocked workload the power consumption is higher. I’m not that surprised.

          • mat9v
          • 2 years ago

          Now we know how it is possible for nominal 7900X to be so much faster then Ryzen 😉 It is running overclocked without user intervention. No wonder the differences between tests are so large with some scoring (stock) 2100p in Cinebench while others score just about 2350p at the same “stock” setting (and in reality OC 4.3Ghz for all cores) and suddenly using 300W+ (system).
          Oh and where did you get this 700Mhz overclock? Default All Core Boost for 7900X is 4Ghz and they were running at 4.4Ghz during OC (Prime95) and up to 4.8Ghz for other tasks.
          To quote: “At a Vcore of 1.4V, the system stayed stable for 10 benchmark runs. Intel’s Core i9-7900X consumed an average of 261W, while individual peaks jumped as high as 293W. A test at 4.8 GHz using 20 instances of a year-long shading computation for a rooftop photovoltaic array, including profit calculation, pushed power consumption all the way to 335W. The motherboard shut down after we started Prime95 without limiting AVX. The last recorded value was 364W.”
          You do not wany to OC this high with power hungry CPU like this one. At least if you are planning to run anything more important and heavy then games on it.

            • chuckula
            • 2 years ago

            The Asus boards weren’t running overclocked and that’s what TR used in its review to show the 7900X being quite a bit faster than its predecessor (much less the 1800X).

      • mat9v
      • 2 years ago

      The problem is that in a lot of tests 7700K is also faster then 7900X and 1800X is strangely, I think, on average the slowest CPU in those tests. The only tests that actually stressed CPU there were kernel compiling (7900X reached 93C and 7700K was only 20% slower then 1800X???) and OpenSSL (that one is reasonable, 1800X is faster then 7700K and 7900X is faster still, power use when normalized is the best on 7900X) . Other tests did not stress the CPU, or how else can you explain average power use for the tests of 120W (whole system) being in contrast to kernel compilation using 234W on 7900X. When you have single threaded workloads highly optimized for Intel you should not expect anything else – the whole test is strange for me.

        • chuckula
        • 2 years ago

        In the gaming benchmarks the 7700K is ahead but not by a large margin and the only really oddball result in there are the apache benchmarks. GCC & OpenSSL are both excellent real-world benchmarks that use all the cores heavily.

        As for the temperatures of in the compiling tests, have a look at the tower coolers they used and note that the “high” 93C temperature was actually measured on the cooler that had an average temperature of only 69C under load, meaning it was probably a small spike that occurred before the fan could spin up on the heatsink. It’s not bad to get results like that with two average tower coolers when the hype suggests that even liquid cooling can’t keep the 7900X operating properly at stock speeds.

    • brucethemoose
    • 2 years ago

    Silicon Lottery is gonna have a good year.

    I’m not a shill or anything, but it is a great service… If I was dropping that much on a platform, I would definitely pay a little extra for a nice bin and a delid/TIM replacement with a warranty.

      • blahsaysblah
      • 2 years ago

      Uhh, if you were paying that much money, Intel should have soldered with best in class TIM in the first place?

        • derFunkenstein
        • 2 years ago

        Agreed. Still, Silicon Lottery shouldn’t’ be expected to do their fix for free, so you can’t be too hard on them. Be very, very hard on Intel, though. They’ve earned it.

          • brucethemoose
          • 2 years ago

          They’re also living proof that this was a bad business decision.

          Silicon Lottery is making money replacing TIM. With shipping and so on. It’s clearly a feature people want and will pay for, so Intel’s just losing money by not doing it themselves.

            • derFunkenstein
            • 2 years ago

            Intel is only “losing” money if people choose THRDRPR instead of Skylake-X.

            They could charge marginally more and use a better TIM, but as many CPUs as SL will sell, the vast majority won’t care enough to do anything about it.

    • Anovoca
    • 2 years ago

    So early x299 adapater steps now include:
    1.) Delidding their $1000 CPUs and adding better thermal paste
    2.) Finding a way to mount a 140mm fan to their VRM heatsinks.

    Rmmm, yeah – No thanks.

      • blahsaysblah
      • 2 years ago

      3.) Finding a thermal response(not just monitor) solution for key power connections/wires.

    • Mr Bill
    • 2 years ago

    Market segmentation (.) You will not overclock, you will buy the next faster CPU? When you think about it that way, its just a way to use thermal dissipation to segment your bin speeds in a way that can’t be gotten around by overclocking.

      • K-L-Waster
      • 2 years ago

      Alternatively, x299 may have been pushed out early to respond to RyZen and not really be ready for prime(95) time…

      Either way, not making me want to jump up and buy one.

        • xeridea
        • 2 years ago

        Intel started using toothpaste to cool CPUs a few years ago, has been a trainwreck ever since. Sad they refuse to not use toothpaste on $1000+ CPUs. It is hard to keep a 7700k cool, so they figured double TDP would be no problem….

      • xeridea
      • 2 years ago

      Intel marketing geniuses at work again!

      • ImSpartacus
      • 2 years ago

      Yep, Intel is very comfortable with market segmentation.

      They are frighteningly good at making a profit.

        • bfar
        • 2 years ago

        Especially when there’s basically no competition. They’ve got complacent though.

    • shank15217
    • 2 years ago

    Damn it AMD, why did you have to come out with that Threadripper CPU? – Intel

      • ultima_trev
      • 2 years ago

      To be fair, leaks (whether they are to be trusted or not) show Threadripper’s multicore performance to be hardly an upgrade to Ryzen 7. Makes me wonder if it’s an early engineering sample where the CPU is throttling to save its life or if it too will have thermal problems like Skylake-X:

      [url<]https://browser.primatelabs.com/v4/cpu/3324737[/url<] My stock clocked 1800X for comparison: [url<]https://browser.geekbench.com/v4/cpu/3342097[/url<]

        • shank15217
        • 2 years ago

        If that is the case, their entire EPYC line of cpu is in serious trouble. Threadripper is simply a 2 module EPYC processor.

          • K-L-Waster
          • 2 years ago

          Not necessarily. For servers you don’t tend to run them on the ragged edge of thermal and power capacity.

        • K-L-Waster
        • 2 years ago

        For that matter, the 1600 and 1600X aren’t far off either — even better single thread and comparable multi-thread.

        [url<]http://browser.primatelabs.com/v4/cpu/search?utf8=%E2%9C%93&q=ryzen+1600[/url<]

          • ultima_trev
          • 2 years ago

          Those seem overclocked going by the stated clock speeds. My 1800X at 4.05:

          [url<]https://browser.geekbench.com/v4/cpu/3373757[/url<] Therefore I hope this result is either an engineering sample or outright spoofed.

        • brucethemoose
        • 2 years ago

        That’s just an Amdahl’s law demonstration.

        Everyone thinks these CPUs with insane core counts are the best thing since sliced bread. I think they’ll be in for a rude awakening when they discover how poorly most programs scale, and how much overhead these HUGE interconnects (particularly AMD’s multi chip ones) have.

          • blahsaysblah
          • 2 years ago

          🙁 I am Amdahl, I bought a dual Pentium Pro back in the day and got that exact rude awakening.

            • Wirko
            • 2 years ago

            Now everyone who buys an X with two Zs is an Amdahl?

          • xeridea
          • 2 years ago

          Multi core scaling is highly dependent on workload. Some get near perfect scaling, some so so, some not at all. Also depends on the reason you need more cores. If you run VMs, heavy multitasking, etc there is a huge benefit.

          • jts888
          • 2 years ago

          Zeppelin’s interconnects are actually surprisingly frugal. GMI is only ~40 GB/s per inter die link, or the same amount of bandwidth to/from a 2 channel DDR4 controller.

          In contrast , Skylake-E has a mesh of presumably 2-4, 64B ports per core clocked who knows how high, and a 128B datapath from each L2 through each L1D to the cores. AVX512 needs bandwidth at any and all costs to avoid bottlenecking.

        • Amiga500+
        • 2 years ago

        Those leaks are tripe.

        ST performance drops 11% for a 6% drop in clocks?

        Yep. Sounds legit.

          • Redocbew
          • 2 years ago

          A non-believer? Of Geekbench? This heretic must be burned.

          • Klimax
          • 2 years ago

          Overhead and getting below optimal operating range for given design.

        • Redocbew
        • 2 years ago

        If you’re planning on buying a Threadripper for your leet gaming machine, then yeah I wouldn’t expect a huge difference from Ryzen 7. I wouldn’t expect a huge difference between Ryzen 5 and Ryzen 7 for that either, but clearly there is benefit from using a Ryzen 7 for those workloads which scale beyond what a Ryzen 5 can provide.

          • brucethemoose
          • 2 years ago

          I can’t wait to see AMD and Intel advertise 12+ core monsters as gaming CPUs.

          If you to play several games at once, maybe…

            • anubis44
            • 2 years ago

            Speaking of games, maybe with 12+ core CPUs, we’ll finally get the highly destructable environments I’ve been waiting for since the 1980s. In a game like Company of Heroes, when tank fires at a 5 storey brick building, maybe we’ll actually get a highly realistic, well-calculated partial collapse, leaving parts of some of the floors intact, rather than a ridiculous, pre-rendered total collapse of the structure at the slightest provocation.

            For THAT, I’d buy a Threadripper CPU.

            • brucethemoose
            • 2 years ago

            Console hardware will dictate that more than anything

            • SiSiX
            • 2 years ago

            “They” could do that now. The bigger problem is this: Exactly how do you keep track of all the debris from that building you just destroyed? Sure, it seems really cool to be able to have different bullet holes (doable), destructible walls (doable), floors that collapse into random piles of rubble (doable), and random craters, destroyed trees, and piles of dirt here and there (all doable). The only question is: How much game data overhead do you want to put into keeping track of all that extraneous “stuff”? You think load times are bad know, imagine them in a war game with a couple hundred thousand random bricks, rocks, broken windows, pieces of wood, and other random debris that all has to be loaded BEFORE anything happens.

            Yeah, it’s not a CPU thread issue, it’s a database asset management issue.

            • anubis44
            • 2 years ago

            Fair enough. Although I have done some programming, I certainly haven’t programmed anything remotely approaching a game engine.

            Sure, it’s a lot of ‘stuff’ to keep track of, but that’s why we have CPUs and GPUs that can perform at several orders of magnitude faster than their equivalents of 10, 15, 20 years ago. As for load times/storage, M.2 SSDs are getting ever-faster, and we can now reasonably expect a gaming PC to require 16GB of ram. I think we’re getting to the point that highly optimized code should be able to pull off what I’m talking about with highly destructable buildings. As for the debris, I’m open to a certain amount of simplification for the sake of play-ability, but right now, the stuff just simply disappears most of the time, which is ridiculous. In a real war, that debris would almost certainly enhance the defensive options for infantry, so just making it vanish into a puff of smoke is pretty mickey mouse.

            • Amiga500+
            • 2 years ago

            They’ll probably advertise them as machines capable of opening MS Outlook in under 5 minutes 😀

      • albundy
      • 2 years ago

      its either that or go out of business. then you’ll never get intel to make anything faster.

Pin It on Pinterest

Share This