Intel teases 28-core, 56-thread HEDT CPU on stage at Computex

As part of its Computex keynote today, Intel teased a 28-core, 56-thread high-end desktop CPU in a single-socket system. Incredibly—and I do mean incredibly—the company demonstrated this part with a 5-GHz all-core clock speed while running Cinebench. The company was apparently able to do that without using exotic cryo-coolants, too. The systems on display appeared to be in standard desktop cases using chilled water cooling.

Yes, that's 56 logical cores at 5 GHz

No word on the power supply, cooling apparatus, or wiring upgrades regular folks might need to run one of these systems at similar speeds, but the answer is no doubt “prodigious” in all regards. Intel says these parts are coming in Q4 of this year, so we'll have plenty of time to get ready.

Comments closed
    • tipoo
    • 1 year ago

    Interview with the engineer

    [url<]https://www.youtube.com/watch?v=ozcEel1rNKM[/url<]

    • Leader952
    • 1 year ago

    Intel: We ‘Forgot’ to Mention 28-Core, 5-GHz CPU Demo Was Overclocked

    [url<]https://www.tomshardware.com/news/intel-28-core-cpu-5ghz,37244.html[/url<] [quote<] In the end, Intel claims the whole fiasco is merely the result of a flubbed recitation of pre-scripted lines, with the accidental omission of a single word: "Overclocked." Maybe that's the truth, but there's a lot of room for debate considering how convenient an omission this is. ... We tracked down the test system the next day and did the full deep dive analysis in our Intel's 28-Core 5GHz Processor And Test System Breaks Cover article. We learned that Intel had in fact used a water chiller to push the processor to 5GHz, which confirmed our suspicions. It certainly wasn't a garden-variety chiller, though. The industrial one-horsepower Hailea HC-1000B can pull a peak of 1000W alone, which is a shocking amount of cooling power for a processor that Intel positioned (by omission) as a chip coming to market at 5GHz. We also learned that the test system required a 1300W power supply, so total system power draw for the demo could stretch up to 2300W. In fact, the hideous amount of juice required to power the demo system prevented our contacts from showing us the performance demo privately: The area we were in simply didn't have enough dedicated circuits for the task. [/quote<]

      • NeoForever
      • 1 year ago

      Yep, just an overclocked $10,000 Xeon running at 10C.

      Anandtech did a great job revealing the industrial sized 1hp chiller.
      And I can’t believe Tom’s of all sites dug deep enough for Intel to admit “they forgot” to say it’s overclocked.

      Gamers Nexus did a great [url=https://www.youtube.com/watch?v=tRH0-QwhvVQ<]video[/url<] even before this was confirmed. Smart guy.

      • thx1138r
      • 1 year ago

      They also “Forgot” to mention the 1hp (yes, horsepower) water chiller they had hidden under the desk, but that’s ok, I’m sure every “5Ghz 28-core” headline in the press will be suitably amended.

        • NeoForever
        • 1 year ago

        Actually, if it was under the desk (rather than away in a separate room), it’s even clearer that they intended to mislead people.
        Away = out of sight out of mind. Gives them plausible deniability
        Under desk = they deliberately sound proofed it. They wanted people to think this was done without exotic cooling, as Jeff believes.

      • Mr Bill
      • 1 year ago

      I have this [url=https://www.marshallscientific.com/Neslab-CFT-75-Circulating-Chiller-p/NCFT-75.htm?gclid=EAIaIQobChMI1LegrdPE2wIVGZ7ACh0SGQ0AEAQYASABEgKrAfD_BwE<]Neslab CFT-75 chiller[/url<] in my lab. It has 2500W of cooling capacity. This one is $3200 refurbished and Its pretty loud. I have it upstairs in the storage area so that I don't have to put up with the noise in the lab.

    • LoneWolf15
    • 1 year ago

    You know, stunt or not, I’m just thankful we’re seeing innovation instead of stagnation these days.

    Thank you, AMD.

      • LoneWolf15
      • 1 year ago

      I’m thankful just that quad laptop chips are reaching consumer mainstream too.

    • BIF
    • 1 year ago

    It’s only been about 4 years since F@H took away bonus points for SMP (or any CPU based) folding. And 3D graphic cards can do a lot of GPGPU compute work now. I love the idea of having 56 threads, but I have to ask what it’s good for in this age of the GTX?

    What will this unicorn cost me, does it spin garments of gold out of its butt, make my scrambled eggs in the morning (maybe without the heatsink?) or help me cure cancer?

    If not, then please tell me again how this should be appealing to me?

      • Krogoth
      • 1 year ago

      It is a pure stunt part that makes no sense outside of servers and entry-level HPC applications.

      Hell, 12-16 core HEDT parts from Intel/AMD are nearly overkill. You are getting them more for I/O connectivity and memory capacity (48-60 PCIe 3.0 lanes + 128GIB of UDIMM/512GiB of RDIMM)

      • Srsly_Bro
      • 1 year ago

      World Community Grid is an option too. I’ve been running it since 2010. It has lots of great projects and is CPU based.

    • synthtel2
    • 1 year ago

    It doesn’t avoid being a marketing stunt just because COTS liquid cooling hardware can handle it. 5 GHz all-core on SSE workloads on golden SKL-X XCC dies is still a preposterous amount of power. In the vast majority of workloads that can actually make use of this in the first place, it’ll get its ass handed to it in all three of perf/efficiency/price by 2S server boxes with double-ish the cores at more sane clocks. It is cool that Intel can do this, but just because they can doesn’t mean they should.

    Alternately, it could be stealth Cascade Lake and Cascade Lake could be awesome, but I’m not holding my breath until I see something not named *Lake.

      • Srsly_Bro
      • 1 year ago

      Some say we shouldn’t have gone to the Moon. Some say it never happened. Only the brave ever did.

      #brave2k18

      • strangerguy
      • 1 year ago

      It’s almost designed, built and shipped for sake of egotistical celebrity youtubers.

    • jackbomb
    • 1 year ago

    To me, and probably only to me, these power hungry enthusiast chips just aren’t that impressive anymore. It’s what the ARM CPU vendors are doing–packing more and more performance into under 2 watts–that’s really interesting.

      • blastdoor
      • 1 year ago

      I think they are all pretty interesting.

    • chuckula
    • 1 year ago

    Videocardz has [url=https://videocardz.com/newz/asus-dominus-spotted-with-28-core-cascade-lake-x-cpu<]detailed interior photos of that system[/url<] It looks like a nice custom water loop but it ain't LN2 or anything extremely exotic. I also highly doubt that the water is being chilled below ambient or else there'd be telltale condensation on the non-insulated piping that's clearly not present. Additionally, a single loop is cooling both the CPU and two presumably large GPUs, so apparently the 5 GHz OC didn't require a dedicated cooling loop just for the CPU. The socket does look to have the dimensions of LGA-3647.

      • thx1138r
      • 1 year ago

      The RAM configuration would further indicate LGA-3647.

        • Srsly_Bro
        • 1 year ago

        Nice post, bro.

      • Srsly_Bro
      • 1 year ago

      They also pointed at the ram config being hexa channel instead of quad.

      [url<]https://videocardz.com/newz/asus-dominus-spotted-with-28-core-cascade-lake-x-cpu[/url<]

      • Jeff Kampman
      • 1 year ago

      When those systems appeared on stage, they were definitely using some form of chilled-water cooling apparatus. You can see a video replay of the press conference here: [url<]https://www.facebook.com/IntelTaiwan/videos/10156005253479017/[/url<] All of that stuff was clearly swapped out when they were put on display later.

        • chuckula
        • 1 year ago

        Thanks for the link. Don’t blink or you’ll miss the cinebench run.

        I look forward to the 32 core Zen2 Threadripper running at 5.5 GHz on a small HSF that will clearly be on display tonight.

          • cegras
          • 1 year ago

          Your fervent wish fell flat. I almost feel bad for all the backpedaling you’ll have to do.

            • chuckula
            • 1 year ago

            What backpedaling. Where’s your 32 core Epyc at 5 GHz doing these benchmarks faster?

            If 7nm Ryzen-whatever can’t hit 5GHz out of the box what are [b<]you[/b<] going to do then? I see that you failed yet again to make a single post except to insult me personally. Did it take you that long to get clean underwear after you saw the demo?

            • cegras
            • 1 year ago

            I don’t care, it’s you who cares. You made a big deal about this chip not needing chilled water but you wanted to believe so much you couldn’t even wait to confirm anything before you made the victory lap. How about a eat your shoe lap?

            You’re a drag on this community. I can find plenty of places and people to talk to that don’t have any of the raging banality you’re now infamous for. My mutual gag offer still stands, but you like doing this so much you’ll never take it.

        • Mr Bill
        • 1 year ago

        Ya, those insulation wrapped lines give it away. Thanks for linking that.

        • Mr Bill
        • 1 year ago

        The Anantech [url=https://www.anandtech.com/show/12907/we-got-a-sneak-peak-on-intels-28core-all-you-need-to-know<]'Sneak Peak' photo[/url<] shows those same water lines insulated right up to the CPU cooling block. There is probably a Neslab chiller hidden away somewhere. I run my instrument Neslab chiller at 58psi and 16C; sufficient to cool a plasma power of 1500W from a 40MHz RF supply for an argon plasma torch. Edit: Its a "Hailea HC-1000B" running at 4C, with a cooling power of 1770W. Just had to scroll down to read it.

          • blastdoor
          • 1 year ago

          Oh dear….

          • Waco
          • 1 year ago

          Ha. I figured chilled water would be barely enough to get this monster going, phase change makes a lot more sense.

        • Goty
        • 1 year ago

        *EDIT* Nope, I’m dumb. Didn’t know chilled water systems got up to 1700W.

    • thx1138r
    • 1 year ago

    Is this really a HEDT chip?

    According to this (more than slightly) dubious source:
    [url<]https://wccftech.com/intel-28-core-cpu-lga-3647-socket-6-channel-memory-aorus-motherboard/[/url<] it's just an overclocked server chip running on the server-class LGA 3647 socket.

      • derFunkenstein
      • 1 year ago

      The difference between “HEDT” and “server” chips is merely branding and arbitrary feature removal.

        • thx1138r
        • 1 year ago

        I do agree that the difference between HEDT and server is mostly marketing and I have no problem with that. I would though be quite surprised if the final product in this case had 6-channel RAM and a different socket, the HEDT market is small enough as it is.

      • Srsly_Bro
      • 1 year ago

      So much casual in this post.

      • NeoForever
      • 1 year ago

      You called it.

    • moose17145
    • 1 year ago

    I can pretty much promise you they were not running the stock garbage TIM that the peasant consumers are going to get.

      • chuckula
      • 1 year ago

      Stop picking on poor old Tim!

    • jarder
    • 1 year ago

    Looks like the 24-core Threadripper is ahead of schedule 😉

      • chuckula
      • 1 year ago

      Apple ARM Miracle chip…. well you know what comes next.

        • Spunjji
        • 1 year ago

        This comment is even more amusing now that the announcement of 24 and 32 core Threadripper chips is official.

      • blastdoor
      • 1 year ago

      Every few months we get another, bigger, example of just how little Intel was trying prior to Ryzen showing up on the scene.

      I would just like to say ‘thank you’ to all the people who have kept AMD afloat over the years, from the German government to Persian Gulf Oil Sheiks to the bond market. You have all made the world a better place.

        • chuckula
        • 1 year ago

        What are you talking about. According to every post in this thread this chip is a failed piece of crap*.

        Why are we thanking AMD for making Intel produce failed pieces of crap?

        * And if a 2019 Zen 2 gets close to it in any benchmarks it is proof that AMD is the only innovative company that has ever existed [AVX-512 benchmarks don’t count].

          • blastdoor
          • 1 year ago

          Oh, boo hoo.

        • Klimax
        • 1 year ago

        You mean, how little PR department did? Otherwise, your comment doesn’t ave that much in common with reality…

    • DavidC1
    • 1 year ago

    This doesn’t mean it’ll be sold as a 5GHz all-core part. This is like the overclocking demos they used to do way back then.

    • freebird
    • 1 year ago

    A picture is worth a thousand words… or is it watts…???

    [url<]https://wccftech.com/intel-28-core-cpu-lga-3647-socket-6-channel-memory-aorus-motherboard/[/url<]

    • chuckula
    • 1 year ago

    Oh please Intel… really? Cinebench?

    No self-respecting AMD employee would be [url=https://youtu.be/J3pJ_–nf5E?t=2m2s<]caught dead trying to pretend that Cinebench has any value for showing hardware performance.[/url<]

      • derFunkenstein
      • 1 year ago

      wait…is Cinebench good or bad, then? You seem to say both in this one comment.

        • chuckula
        • 1 year ago

        [quote<]is Cinebench good or bad, then?[/quote<] That's an easy question to answer. Cinebench was [b<]THE PURE EMBODIMENT OF EVIL[/b<] from the day it was released until March of last year. Then it became good and awesome for product comparisons except for those pesky > 10 core Intel HEDT parts that we just pretended didn't exist. Back to the present, Cinebench is gradually returning from being Lisa Su Approved (TM) to once again being [b<]PURE EVIL[/b<]. These things go in cycles.

          • derFunkenstein
          • 1 year ago

          You seem really upset. Do you need to talk it out?

            • JustAnEngineer
            • 1 year ago

            Is there a supervillian origin story? It still seems possible that on some unspeakably dark day in the past, Jerry Sanders sullied the honor of Chuckula’s sister…. or perhaps Chuckula’s father was responsible for the street gang attack that nearly killed Sanders.

    • derFunkenstein
    • 1 year ago

    Neat, I guess. Trying to imagine what you’d do with this. Many-thread server apps is all I can come up with.

      • derFunkenstein
      • 1 year ago

      everybody downthumb that because it’s a dumb comment. i mean like duh thats what its for.

      need my coffee

        • jihadjoe
        • 1 year ago

        It’s for Cinebench, obviously!

        What else do you think people build Dual Epyc workstations or Xeon Scalable servers for?

          • derFunkenstein
          • 1 year ago

          of course. my mistake. Can’t believe I overlooked that critical reason!

      • dragontamer5788
      • 1 year ago

      Blender and 3d modeling in general. Run “Gooseberry” and you’ll know why: [url<]https://www.blender.org/download/demo-files/[/url<] Threadripper 1950x (16-core / 32-threads) completes 1-frame of Gooseberry in 35ish minutes. For the whole 10-minute movie, that'd be 14400 frames (10 minutes x 60 seconds x 24 frames per second), or roughly 300+ days to complete rendering the short-film.

        • Klimax
        • 1 year ago

        And for added fun, try physically-based rendering. (like Iray)
        One image can take several days…

    • chuckula
    • 1 year ago

    Since the prevailing consensus that’s being copy-n-pasted and upthumbed by bots is that only a failed and inherently broken piece of silicon can run Cinenbench at 5GHz with 28 cores then this ought to be stupidly simple for AMD to counter later today.

    All they have to do is take an Epyc part out of the trash can, overclock it to 5GHz and show it running Cinebench at least 15% faster than this miserable pathetic failure of a 28 core chip.

    Oh, and they have to get enough of them out of the trash to have them on-sale this year prior to Intel.

    Go ahead, I’m waiting to see failed Epyc parts put Intel out of business.

      • Amiga500+
      • 1 year ago

      Meanwhile, Intel debuts a new “feature” for its latest HEDT.

      The cooling tower.

      Measuring just 30 metres in diameter and 20 metres in height, this superb piece of craftsmanship can* cool your new 28 core 5.0GHz Core i9-911 EmergencyEdition down to a level where it will last a FULL hour running cinebench before melting down.

      The cooling tower comes at an estimated cost of $4 million dollars – which is comparable to the price of the CPU itself**. This limited edition chip is not available from any retailers, contact Intel directly if seeking a purchase.

      *planning permission required.

      **Motherboards supporting the latest Intel HEDT cpu cost in the region of $2 million.

        • chuckula
        • 1 year ago

        Funny, it looked like a pretty boring liquid cooling loop being used in a live on-stage demo to me.

        But go ahead with pretending that Intel can’t compete with GloFo’s chip output.

        But since you don’t think this is possible let’s make a friendly bet: If AMD can get a 7nm Threadripper out in the first quarter of next year that overclocks better than this chip and that beats it in Cinebench by at least 20% I’ll buy you the AMD Threadripper.

        Otherwise you buy me one of these and stop spreading fud because you can’t process reality.

          • Leader952
          • 1 year ago

          [quote<]Funny, it looked like a pretty boring liquid cooling loop being used in a live on-stage demo to me.[/quote<] When [b<]is[/b<] your next eye appointment?

    • chuckula
    • 1 year ago

    Thank you Intel for showing us what a happens when you have a design that can scale past 8 cores.

      • Krogoth
      • 1 year ago

      They have been beyond 8-cores since Ivy Bridge-EX.

    • Krogoth
    • 1 year ago

    In other words, Intel is going to be releasing Skylake XCC “rejects” in light of rumored “Threadripper+ 12nm refresh” and they will require socket LGA3647 boards. The package is comparable in size to TR4/SP4.

      • blastdoor
      • 1 year ago

      In other other words, AMD has been willing to do whatever it takes to compete and this is forcing Intel to do the same. It’s good for everybody except Intel stockholders.

        • Questar
        • 1 year ago

        [quote<]In other other words, AMD has been willing to do whatever it takes to compete and this is forcing Intel to do the same. It's good for everybody except Intel stockholders.[/quote<] Intel's stock price is at record highs. Their shareholders are quite happy.

          • blastdoor
          • 1 year ago

          Perhaps this is too subtle of a point, but the relevant comparison is to what the profits would have been in the absence of competition from AMD. The overall market has been up a fair bit over the last several years so many companies are at or near all time highs. That doesn’t really mean that AMD hasn’t had an effect.

          Having said that, though… I agree the effect so far is not huge. But I believe Intel’s margins did slip a tad in the last quarter. I’ll hazard a guess that without AMD around, those margins would have been up a tad.

          String several quarters of down a tad rather than up a tad together, and AMD’s influence will become increasingly noticeable.

          And yet that doesn’t mean that Intel’s stock will necessarily go down. It just means that it would have been higher without the competition from AMD.

            • chuckula
            • 1 year ago

            [quote<]but the relevant comparison is to what the profits would have been in the absence of competition from AMD. [/quote<] Substantially lower because Intel would have to compete against a company that isn't AMD.

            • Spunjji
            • 1 year ago

            Yeah, because Via (and Cyrix before them) and Transmeta and all those other competitors that utterly failed to stick it out competing with Intel must have somehow been staffed with people even worse than all the idiots at AMD. /s

            Whatever it is that’s stuck in your throat, cough it up. AMD make missteps but it’s weird that you’re pouring out such vitriol for the only organisation to ever successfully compete with Intel on their home turf.

            • Questar
            • 1 year ago

            [quote<]the relevant comparison is to what the profits would have been in the absence of competition from AMD[/quote<] What would Intel's profits been without AMD forcing Intel to compete? You made the claim as fact, so please provide evidence of your claim with real data.

            • blastdoor
            • 1 year ago

            I don’t think I ever made a claim about fact. I certainly didn’t use the word fact. And what I’m talking about is actually a [b<]counter[/b<]factual condition. Estimating what would have happened under a set of circumstances that don't actually exist is very tricky. But it's also very important to try if you have any interest in trying to assess causal relationships. One approach, which implicitly the approach I was using, is to make predictions about the counterfactual using a model. A good model is one that makes good sense and/or makes good predictions. The model I was using is a model of the effects of competition. Here's some reading for you if you're interested: [url<]https://en.wikipedia.org/wiki/Counterfactual_conditional[/url<] [url<]https://en.wikipedia.org/wiki/Rubin_causal_model[/url<] [url<]https://en.wikipedia.org/wiki/Competition_(economics)[/url<] [url<]https://en.wikipedia.org/wiki/Monopoly_profit[/url<]

    • Chrispy_
    • 1 year ago

    Dual-die 14C/28T, at ~150W per die?

    That’s not unrealistic but these things are going to be the rarest, most cherry-picked dies Intel is making. Expect $4000 price tags and terrible BIOS support for the useful part of this chip’s time in the limelight.

      • Krogoth
      • 1 year ago

      Nah, they are going to be Skylake XCC “rejects” that ate too much power for “Gold/Plantium” brand.

      Intel will likely axe ECC support and certainly RDIMM/LR-DIMM support while cranking up voltage/clockspeed to be comparable to their current i9 offerings.

      They will take-up the current i9 slots while the existing stock will undergo price cuts to be more price competitive to the Threadrippers/Ryzen+ SKUs.

        • chuckula
        • 1 year ago

        What Intel calls a “reject” AMD calls “oh !@#%!”

        Incidentally, I just did a search. Where are your wall of unoriginal copy-n-paste posts calling every Threadripper part ever made a failed Epyc “reject” part?

        Where are all your posts calling every desktop RyZen a failed “reject” Threadripper part?

          • Krogoth
          • 1 year ago

          Threadrippers are Epyc rejects too for similar reasons.

          • derFunkenstein
          • 1 year ago

          you’re feisty this morning

            • chuckula
            • 1 year ago

            I’m getting Krogothed with the same copy-n-pasted braindead drivel that gets posted around here whenever Intel does something interesting.

            I sure as hell don’t recall the 2700X being called a “reject” part when it included a minor clockspeed bump and noticeable maximum power draw increase over last years 1800X.

            Funny how the “logic” that certain people post on Intel articles for easy upthumbs by the usual crowd doesn’t seem to get copy-n-pasted into articles about AMD parts where it would actually be more applicable.

            • derFunkenstein
            • 1 year ago

            Yep, I guess feisty was the right word.

            • Krogoth
            • 1 year ago

            R7 2700X-2700 aren’t a “rejected” part. It is fully enabled silicon of that family. The 2600-2600X are the “rejects” of that family.

            Intel isn’t doing anything interesting with this. It is just a Skylake-X XCC “reject” being shoe-horned into “HEDT market” by marketing types who think that more cores = better!

            It is no different from Threadrippers and upcoming 12nm+ refresh which are just really Epyc rejects.

            I suspect these stunts are just distractions from Intel’s growing pains with their problematic 10nm process.

      • derFunkenstein
      • 1 year ago

      it’s obviously three and a half 8-core Coffee Lake dies.

    • ermo
    • 1 year ago

    Begun, the Core Wars have.

      • juanchotazo
      • 1 year ago

      You mean the Corean War, right?

        • chuckula
        • 1 year ago

        There have been temporary ceasefires, but the Corean War has never officially ended!

    • tsk
    • 1 year ago

    Inb4 RIP AyyMD

    • blastdoor
    • 1 year ago

    Maybe the heat sink is a Stirling engine that generates electricity?

    That would increase performance per watt and give a new (as in old) meaning to the word turbo.

    • Amiga500+
    • 1 year ago

    *cough* bullsh_t *cough*

    Given they have just unveiled their first (and the worlds only x86) chip to do 5GHz – and that at a single core turbo, without any marked process jumps they expect to go from that to 5GHz across 28 cores?

    Doesn’t pass the sniff test. Not even close.

    For reference, the Xeon Platinum 8180 is a 28 core CPU, base clock of 2.5 GHz and SINGLE core turbo of 3.8 GHz and has a TDP of 205W.

    For that to even theoretically run at 5.0 GHz, your likely talking power consumption in excess of 1KW! How long would the silicon last? ***Maybe*** long enough to do a 30 second demo….

    edit: Note – my gripe is not with the demo – Intel are extremely unlikely to lie about what they have on stage. My gripe is the notion that you or me could buy one of these things in a few months time and expect to achieve a 5 GHz all core turbo!

      • Zizy
      • 1 year ago

      I generally agree it sounds BS, though that Platinum chip:
      1.) has extra pcie, memory channels and QPI. This chip is known to be 1P only.
      2.) is expected to run 24/7 at those clocks and TDP without errors – unlike this part.

      Now take binning to the next level for such halo part, up the TDP to 300W and we could see ~3.5GHz base and ~5GHz single core turbo. Finally, as mobos can OC the chip to the single core clocks on all cores if power and thermals allow you have semi-plausible claim the beast is a 5GHz@28C chip… just not for very long in actual workloads.

      • Chrispy_
      • 1 year ago

      Agreed. I think Intel is just ‘announcing’ this to rain on Threadripper’s pending refresh;

      A sub-$1000 16C/32T 4.25GHz Zen+ is really going to ruin their i9 and HEDT market, almost as much as the price cuts on 1st-Gen Threadripper will. If people really need cores, they’ll be open to running multiple boxes and that’s what we do. I can buy two 1950X machines for every equivalent i9 and it’s not even clear cut which chip is faster. In a lot of our scenarios (3D rendering) the AMD architecture is faster anyway.

      • chuckula
      • 1 year ago

      [quote<]For that to even theoretically run at 5.0 GHz, your likely talking power consumption in excess of 1KW! [/quote<] In a non-AVX workload? Not even close. I'd bet it's probably noticeably lower than the power draw of a Vega 64 during a regular game. This is just Intel showing that Cinebench is an old legacy application, which is pretty much what AMD's entire Zen line is aimed at. It's really not that hard to do SSE instructions that Intel first put on the market with the Pentium 4.

        • Krogoth
        • 1 year ago

        5.0Ghz across cores? Sorry, it will consume more power then a fully-taxed Vega 64 card (probably at least 300W if not slightly more). The real trick is trying to get utilize that much power under non-sythentic workload.

        In will likely scale back when heavily taxed to more sane power consumption levels like their Xeon brethren.

        • Antimatter
        • 1 year ago

        Tomshardware showed a 5GHz 8700K consuming >130W running Cinebench. I know this isn’t accurate but extrapolating form 6 to 28 cores you get over 600W. So if there’s a Vega 64 consuming over 600W you could be right.

          • Srsly_Bro
          • 1 year ago

          I was running a 2.8 GHz 14 core ivb Xeon. Idle the CPU was about 100w total system. Cinebench R15 on all cores raised total consumption to just under 200w.

          Extrapolating doesn’t always work. The architectures are entirely different. Look at 6 core skl vs 6 core cfl.

          You have been assigned homework.

          Srsly_bro

            • Antimatter
            • 1 year ago

            I know it’s not accurate as stated in my comment above. There are process, architecture, IO differences, etc. But extrapolating would provide a better estimate than guessing 300W or 1000W.

      • jihadjoe
      • 1 year ago

      > Doens’t pass the sniff test. Not even close.

      The CB score is a good argument that it really did 5GHz across all cores though. It’s higher than [url=http://hwbot.org/submission/3620709_blueleader_cinebench___r15_2x_epyc_7601_7252_cb<]a 64-core dual-Epyc[/url<].

      • SuperPanda
      • 1 year ago

      You’re forgetting that not all loads are created equal. Nobody is doubting this is pulling Chernobyl meltdown amounts of power (they aren’t using chilled water for giggles), but Cinebench is far from being the most strenuous load in terms of heat and power consumption.

        • Klimax
        • 1 year ago

        CB is SSEx only workload to my knowledge.

      • Takeshi7
      • 1 year ago

      [quote<]Given they have just unveiled their first (and the worlds only x86) chip to do 5GHz[/quote<] Wrong. AMD beat them to the 5 GHz x86 chip by 5 years: [url<]https://techreport.com/news/24940/amd-intros-fx-9590-processor-with-5ghz-peak-turbo-speed[/url<]

        • chuckula
        • 1 year ago

        Yeah, but even we don’t like to admit that we did that.
        — AMD marketing

      • Srsly_Bro
      • 1 year ago

      Got any sources for that, casual bro?

      Any information or just your opinion?

      Links, comparisons?

        • Waco
        • 1 year ago

        I don’t think his skepticism is unwarranted. A 6 core chip at 5 GHz all-core will pull north of 200 watts fully loaded. This architecture isn’t wildly different (from a power consumption perspective) so it isn’t unreasonably to expect a ~4.5x increase in power consumption for a 28 core part. That put it optimistically at just below a kilowatt through the socket…

          • chuckula
          • 1 year ago

          [quote<]A 6 core chip at 5 GHz all-core will pull north of 200 watts fully loaded.[/quote<] And -- as I've already pointed out to people who don't understand what the letters A V & X actualy mean -- what makes you think this CPU was even close to being "fully loaded" by an obsolete toy SSE benchmark? This test was intentionally picked to show what these parts are capable of in legacy applications since apparently Intel's work on AVX & AVX-512 is considered to be "useless" until AMD finally gets around to copying it* and we pretend that they invented it out of thin air. * And no, I don't expect Zen-2 to have real AVX-512 support even if they do an inefficient emulation of it.

            • Srsly_Bro
            • 1 year ago

            The casuals attacked me too. Stay strong, Chuck!

            • TheRazorsEdge
            • 1 year ago

            Everyone running AVX workloads knows what that does to power consumption and clocks. It’s still better for the right loads.

            In case that wasn’t clear… no one expects a 5 GHz CPU to run at that speed for AVX. Only a chip that’s binned below its potential could do that.

            There is an AVX offset setting for a reason. On cutting edge chips, you’ll probably need it.

            • Waco
            • 1 year ago

            By full load I mean all cores loaded with SSE instructions as best as Cinebench can load them. I highly doubt they could get AVX stable at those clocks with that many cores, nor did I mean to imply I was talking about an AVX workload in this context.

            I guess I was inflating the figure a bit since the measurements I’ve seen include the motherboard power draw, but it’s still well into the nearly 200 watt range for the CPU alone (Coffee Lake 6 core at 5 GHz, Cinebench workload). We’re still talking nearly a kilowatt into the CPU from the board (assuming that Coffee Lake is no more efficient than Skylake X…which we know isn’t true), so it’s not like it changes the skepticism much. Even if it was *half* that, 500 watts? That’s an awful lot to ask of a workstation/server board that normally only supports up to 205 Watt TDPs.

          • Srsly_Bro
          • 1 year ago

          Skepticism is one thing…. Making many unfounded claims is different

          You can doubt certain things. When you start making things up to ensure maximum circular logic, that’s where I it any reasonable person would take issue.

          “That CPU uses so much power. It uses 1000w. No way they can cool it. Fake.”

          Do you see the idiocy?

            • Waco
            • 1 year ago

            His language is stronger than I would have used, but I don’t think the initial response of “BS” is totally unwarranted.

            EDIT: To be clear, I have no idea why he thinks it’ll be a shipping product at 5 GHz, that’s clearly not the case. I’m skeptical of anything being able to run that chip at 5 GHz for any real period of time.

            • Redocbew
            • 1 year ago

            The fact that they had to use a chiller in the loop should be all the proof anyone needs of that. If they had been able to get away with using an off the shelf CLC I’m sure they would have used one.

            • Waco
            • 1 year ago

            Yep. It’s a stunt, take it for nothing more than that.

    • LocalCitizen
    • 1 year ago

    does anyone make 480 radiators?
    does anyone make cases that fit 480 radiators?
    just doing some planning…

      • VincentHanna
      • 1 year ago

      [url<]https://youtu.be/Q2SIrV_4-dM?t=2m36s[/url<]

        • LocalCitizen
        • 1 year ago

        4x4x120
        you win
        tho i think it could be improved if 4 radiators are used in parallel rather than in series.

      • DrDominodog51
      • 1 year ago

      1080mm radiators (3x3x120mm) have been made.

      [url<]http://www.performance-pcs.com/phobya-xtreme-nova-1080-9-x-120mm-radiator.html[/url<]

        • LocalCitizen
        • 1 year ago

        and it can be made to fit most mid tower cases if you use this in place of the side panels. who needs to see those rbgs? heat from the 5ghz 28 core is for the real men.

        • UberGerbil
        • 1 year ago

        4K radiators or GTFO

          • Redocbew
          • 1 year ago

          4k fans on a 4k radiator.

          Woah, that’s like 16k.

            • Beahmont
            • 1 year ago

            [url=https://www.youtube.com/watch?v=IagZIM9MtLo<] Hey Hey 16K [/url<]

      • moose17145
      • 1 year ago

      Corsair Obsidian 900D can accept a pair of 480 rads.

    • Anonymous Coward
    • 1 year ago

    So Intel has decided to demonstrate they can make something that not just everyone can make. A good move, I say.

      • Leader952
      • 1 year ago

      I remember the Intel Pentium demo years ago that has a single display showing 4.0 GHz.

      Want to know when it was released: NEVER

        • ig0012
        • 1 year ago

        I guess you mean demo of Pentium IV running 4 GHz. Indeed, it was never released. But Intel did release Pentium IV 3.8GHz. So if Intel could manage to get retail 28-core CPU running at 4.8GHz, I think, it will be good enough.

          • Leader952
          • 1 year ago

          Yea that was it.

          I also remember that this 4 GHz P4 was just the beginning and was going to ramp over time to 10 GHz. Well we all know that didn’t happen.

          Much like what was seen in this demo.

          EDIT: Link to Intel pulling the P4 4GHz

          Intel pulls the plug on 4GHz Pentium 4
          [url<]https://arstechnica.com/uncategorized/2004/10/4311-2[/url<]

          • Spunjji
          • 1 year ago

          They won’t, though. Not without pre-chilled water anyway.

      • freebird
      • 1 year ago

      They should’ve just run it in a big floor freezer… more people will have one of those in their homes than the cooler Intel used in their demo…

      [url<]https://www.tomshardware.com/news/intel-cpu-28-core-5-ghz,37201.html[/url<]

    • brucethemoose
    • 1 year ago

    I love how they have a guy with a gamer headset on in the background.

    [quote<]Now you can stream 5 game clients AND run your private server all at once![/quote<] Also, Jeff, if ya'll manage to get one of these behemoths in the lab, you should ask AMD for a 7551P (or whatever the Epyc workstation refresh is) just for some point of reference. EDIT: Nooo, you switched the image! This is what it was before: [url<]https://cdn.vox-cdn.com/thumbor/HQmQPrU5WMJHhckSANusqdH_2rQ=/0x0:2040x1360/920x613/filters:focal(194x427:520x753):format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/59956207/intelctxvladsavov28core.0.jpg[/url<]

      • Anonymous Coward
      • 1 year ago

      Victory through 28 cores baby.

      • Kretschmer
      • 1 year ago

      Dude, pair this thing with a 1060 3GB and you have yourself a sweet gaming rig!

      (You WILL see someone do something that stupid in a NewEgg wish list. I guarantee it.)

Pin It on Pinterest

Share This