Follow Slashdot stories on Twitter

 


Forgot your password?
Close
typodupeerror
×
AMD Intel Hardware

Leaked Intel i9-12900K Benchmark Shows Gains Over the Ryzen 5950X (digitaltrends.com) 90

Posted by BeauHD from the what-to-expect dept.
UnknowingFool writes: An engineering sample of Intel's next flagship processor, the i9-12900K, was shown to beat AMD's current flagship 5950X in Cinebench R20 by 18% in multi-core and 28% in single-core tests. The next generation of Intel processors is believed to use a hybrid big.LITTLE design where 8 of its 16 cores are for low power usage and 8 are for full power. The low power cores only run in single thread where the high power cores can run 2 threads. No official word on pricing or release date from Intel though but engineering samples and B600 motherboards are being sold in China for $1,250 and $1,150, respectively. According to leaker OneRaichu, the results for the 12900K were gathered using water-cooling and without overclocking, so it's possible the final score could be even higher. The rumors suggest the processor will come with 16 cores and 24 threads with a boost clock speed of up to 5.3GHz.
This discussion has been archived. No new comments can be posted.

Leaked Intel i9-12900K Benchmark Shows Gains Over the Ryzen 5950X More

Leaked Intel i9-12900K Benchmark Shows Gains Over the Ryzen 5950X

Comments Filter:

  • Crysis (Score:4, Funny)

    by anonymouscoward52236 ( 6163996 ) on Thursday July 22, 2021 @05:54PM (#61609759)
    But... can it play Crysis? (Is that joke still allowed? Let me brush the dust off of it.)
    • Still makes me laugh, so I'll allow it

    • Re:Crysis (Score:5, Funny)

      by base3 ( 539820 ) on Thursday July 22, 2021 @06:09PM (#61609805)
      Imagine a Beowulf cluster of these rendering a 3D model of a naked and petrified Natalie Portman covered in hot grits!

    • But... can it play Crysis? (Is that joke still allowed? Let me brush the dust off of it.)

      Actually with Crysis Remastered bringing modern high end gaming PCs to its knees the joke is more relevant now than ever.
      With highest graphics settings and ray tracing on an RTX 3080FE is lucky to break 45fps.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      Really depends on the system it's in. These days CPUs throttle when they get too hot, and a lot of systems do throttle. Laptops are particularly bad (the Intel Macbooks were notorious for it) but even desktops from big names like Alienware often throttle because of really bad cooling setups.

      GPUs are the same, there is a theoretical maximum performance and then there is what any given system is capable of due to thermal constraints.

    • Re: (Score:2)

      by Z80a ( 971949 )

      The original game is a mostly single core application, so it's still a decent test of CPU IPC.

    • My car can go at 150mph, but thus far, the most I usually do is 100 to 105mph. On some highways I do 120mph. So, if my existing hardware can execute jobs that I need to have done within a timely manner, do I need the speed devil. I would say, only if I am a teenager playing games, or some university researcher doing numerical analysis.

  • Not even released yet... (Score:5, Funny)

    by sidekick2 ( 852519 ) on Thursday July 22, 2021 @06:17PM (#61609825)
    And only vulnerable to 4 critical CVEs. I'm joking, I'm joking.... it's 6. :)

  • "leaked" benchmarks of cherry-picked samples (Score:5, Insightful)

    by wgoodman ( 1109297 ) on Thursday July 22, 2021 @06:40PM (#61609879)

    This is a pr move, nothing else. Intel does this all the time to try to stay relevant. They picked the best of the best and used it on whatever synthetic benchmarks make it look best against a processor that's almost a year old.

    • also no mention of real power draw - using avx-infinity or whatever can be used to win benchmarks, but its at the cost of enormous power draw where the lesser coolers cant keep up at all, not even a little bit

      • Re: (Score:3, Informative)

        by willy_me ( 212994 )

        From the article:

        The Core i9-12900K is rated with a 125W PL1 and 228W PL2.

        And they also mention it is not overclocked and that power requirements are about the same, or a little bit less, then the current i9.

        So yes it uses lots of power but that will be with all cores active. If it is just a single core then power consumption should be reasonable. And it sounds like that single threaded performance could be quite good.

        • Who cares about single thread performance any more? Everything CPU-intensive is multithreaded and/or multiprocess. Even games, except for old ones; and because they are old, modern CPUs still have plenty of performance for them.

          Corollary, if everything is using multiple cores now, who cares about single thread TDP? It's irrelevant to anyone but benchmarkers.

          • Who cares about single thread performance any more? Everything CPU-intensive is multithreaded and/or multiprocess.

            This is just incorrect. There are plenty of CPU-bound computing tasks that can't be broken into small pieces after a certain point, so parallel processing doesn't help. A simple example is OCR and page layout recognition. You can't take chunks smaller than the individual page level because the software needs to understand the layout of the entire page -- breaking it into little pieces durin

            • A simple example is OCR and page layout recognition. You can't take chunks smaller than the individual page level because the software needs to understand the layout of the entire page

              So what? OCR of a single page is not an intensive task by modern standards, while OCR of an entire document is easily split into pages. That example is overly simple; it's so simple it's inapplicable.

              • Once again, you're incorrect and talking about stuff you know absolutely nothing about. OCRing pages with Tesseract using the 5950X (again, one of the fastest single-threaded CPUs available today) can average 13-15 *minutes* per page or more depending on the size and complexity of the page layout. Try scaling that up to tens of millions of pages or more that need to get done in a tight time window and then tell me single thread speed doesn't matter.

                • Try scaling that up to tens of millions of pages or more that need to get done in a tight time window and then tell me single thread speed doesn't matter.

                  Your argument is idiotic. The more pages you have, the more jobs it can be split into, and the less single thread performance matters. You have literally proven my point for me, thanks.

                  • Ahh, sweet emotion. Hit a nerve, I see. Have a good day and I hope you think long and hard about how I schooled your ass on a topic you know nothing about. :-)
                  • Yes if you have millions of scanners however the limiting factor of parallelising OCR work is that you cannot multiple the equipment with code changes alone. A scanner can only do so much is his point and you are limited by physical hardware. Your argument is the same as saying if 1 woman takes 9 months to birth a baby, then 9 women can do it in a month. If you want 9 babies, you need 9 women.

                    • You are 100% correct. Sadly, we must operate in the real world and not Happy-Fun Make Believe Land, where we have a finite and limited hardware and people budget (as in, yeah, someone actually has to pay for this stuff), potentially limited electricity capacity ("how many amps are you gonna pull?"), certainly a lack of physical space for all of these machines, an actual time window to complete these tasks (we don't have a server with 10 million images just sitting there; they get created over x weeks / mon

                    • If it takes 15 minutes to OCR a page, which it does according to numpty over there, then you're going to be able to parallelize that because it only takes a few seconds to get a scan if you're not previewing. You dump a stack into the ADF and you walk away.

                      Also ironically Tesseract OCR actually supports multithreaded OCR, you CAN recognize multiple parts of the same page at once, that's what OCR fucking DOES. It breaks the page up into logical blocks before recognizing them. Even very old OCR worked this wa

                    • Re: (Score:2)

                      by tlhIngan ( 30335 )

                      Encryption is also either embarrasingly parallel, or stunningly serial.

                      It's parallel when using the notoriously bad CBC mode, or XTS mode (because of the way it operates - disk reads are non-linear so you have to be able to decrypt a black at a random location on disk).

                      But the other modes are serialized chains - the key gets permuted, a block encrypted, and the permuted key permuted again to produce another key. At best, you have two threads - one permuting keys, one encrypting (or decrypting) the plaintext

                • Why can't you OCR multiple pages at the same time using different cores?

                  • You can, of course -- so on a 5950X, for example, you can run 16 pages at a time (hyperthreading doesn't do squat for long-lived CPU-bound tasks, so we only count actual physical cores). But when it takes minutes per page per core, the bottleneck quickly becomes single-thread performance -- literally, how fast can the machine get a single page done.

                • Re: (Score:2)

                  by AvitarX ( 172628 )

                  Single core speed doesn't matter.

                  You're bound by total computation, not the single core.

                  You could distribute the task to 10 million clients, have them dedicate idle CPU to it and have the 10 million pages done within an hour. That is pretty much the definition of single thread performance doesn't matter.

                  It would matter if there was a reason you needed 1 page OCRd in less than 10 minutes, but your example of ten million pages absolutely doesn't care.

                  • How about finite hardware budget? Electricity costs? Available space? A real time window in which x pages must be processed to keep the project on schedule? Have you ever actually tried to do something like this in the real world?

                    • Re: (Score:2)

                      by AvitarX ( 172628 )

                      I used to batch process PDFs to tiffs, similar but about 100x the speed

                      I would pull round robin from a pile of thousands of PDFs, and hand off, with one PDF converting per CPU. I would go for raw power instead of single thread speed since that is what got the quickest result, even though each PDF was done individually at the speed of a single thread, it was far faster to have twice as many going at once. If the investment choice was twice as many cores on a system, or 50% more peak single thread the choice

                  • There are two better examples in the engineering world: FPGA design and 3D CAD.

                    The FPGA design process has some parts that are embarrassingly parallel - compiling many source modules into RTL for example. There are also some phases that must be done sequentially, like place-and-route. Medium-sized FPGA designs (something that takes ~70% of an Artix-7 200k gate device) take about 30-40 minutes on my I9-9880H. Larger designs, like something targeting the multi-million gate Virtex and UltraScale chips (and

                    • So the answer to the question asked is vanishingly few people.

                      I never said that it was nobody, but there really are very few people who need to maximize single thread performance today. What most people need is maximum performance for 4-8 threads.

          • Re: (Score:2)

            by EvilSS ( 557649 )

            Who cares about single thread performance any more?

            Gamers for one. Even though most new games take advantage of multiple cores, they can still end up CPU bound on single core performance if they are single threading specific functions or are only using a handful of threads for everything so they are not using all of the available cores. I have a 18c/36t CPU and it gets CPU bound on some games compared to my older 4790K system. You can see them maxing out a few of the cores while the rest are all sitting near idle.

            • I have a 18c/36t CPU and it gets CPU bound on some games compared to my older 4790K system.

              Indeed. For years, the venerable 4790K was the single-threaded top dog. I don't think Intel came out with anything faster per core until either the 7700K or 8700K, and even then it was a tiny marginal improvement -- certainly not 20%. That was a disappointing stretch in terms of CPU innovation -- we are fortunate that AMD is finally making Intel get off their ass and do something!

            • Gamers for one.

              Not really. Maybe for some ancient games built on crappy, single threaded engines. Modern engines are efficiently multithreaded with DX12 or Vulkan. The CPU's just is now pretty much just to set up DMA. CPU performance is no longer particularly relevant to game performance and that trned is accelerating.

              Just move on from that old broken crap.

          • Umm lots of people. Pretty much all flight sims are bound by single-thread perf. TLDR: You cant parallelize everything, single-thread perf matters a hell of a lot more than you are intimating.

        • ...and yet, it's still not a product they've gotten to market, let alone had on the market for months.

          At this point Intel needs to stop screwing around or go ahead and say "I'll make my own flagship CPU... with blackjack and hookers" instead of continuing to try to curry favor in the press for things they haven't managed to bring to market. This business of pretending to still be making the fastest hardware is quickly growing stale.

        • But it could also mean there is little performance gains in normal overclocking that a consumer could do. Sure using liquid helium cooling could probably get performance but not something a consumer or professional would do daily in workloads.

  • This is Intel, this is no leak it is a marketing stunt, of course the liquid nitrogen cooled overclocked processor is going to run faster than a Ryzen, a last gasp attempt to keep their stock price from plummeting even further, to try an trick investors into thinking Intel has a hope.

    • Comment removed based on user account deletion

      • Because a very large company such as Intel has a huge amount of inertia.
        It takes years for a direction change. Look at how long it took AMD, which is a much smaller company.

      • Because Intel became so complacent that they fell asleep at the wheel, and drove their car head-on against a wall?
        Because their fab is an unrecoverable trainwreck at this point?
        Because they fired a large part of their key management staff and are in the long tail of a massive implosion right now?

        (All because they have been massive dicks since forever, and now that they are down, everybody wants to have a go. Like the school bully who just got kicked in the nuts by the nerd boy, and is now recoiling on the f

      • Intel was producing the leading CPUs from the mid 2000s until the mid to late 2010s.

        Right, in part because they were cheating, and in part because they had superior process technology.

        They're still cheating, but they no longer have a superior process. Now they have to compete on merit and as it turns out, this seems to be very sparse at intel.

      • what is with all the Fanboism?

        Realism. Intel is still spinning its wheels, stuck in 4-6 year old process technology with no salvation on the horizon.

    • Now now, it clearly says water cooling. So no LN. However we do probably know where that water chiller from 2018 got off to. As for the idiots whining about there being no overclock, Intel CPUs ignore TDP power limits for a given period of time until restricting themselves.

  • Alder Lake is 10nm (Score:5, Informative)

    by Solandri ( 704621 ) on Thursday July 22, 2021 @07:29PM (#61609991)
    This is manufactured on Intel's 10 nm process, with a density of 100.8 million transistors per mm^2. That's competitive with TSMC's 7FF process that AMD is currently using (96.5 MT/mm^2). So this is a pretty apples-to-apples comparison of Alder Lake vs Zen 3. And falls in line with my theory that there's really not much speed/power difference between the different companies' architectures, and most of the performance differences we're seeing are due to manufacturing process size. Intel's 14nm process was 37.5 MT/mm^2, which was a massive disadvantage.

    AMD opted not to pay for TSMC's 5nm process (173 MT/mm^2) for Zen 3, which is what Apple's processors are being manufactured on. That's a big enough gap with Intel's 10nm that AMD probably could've left Intel in the dust had they chosen to pay TSMC more (Apple is only expected to take half of TSMC's 5nm capacity this year). Likewise, Intel could leapfrog AMD and possibly even Apple by paying TSMC to use their 5nm process. (And if you want more fun things to think about, TSMC's 3nm process is supposed to be about 250 MT/mm^2, Intel's 7nm process is estimated to come in between 200-250 MT/mm^2.)

    If you go back the last 30 years, you see that Intel always had an advantage in manufacturing process density. It wasn't that their processors were better. They simply could build them smaller (and thus faster and using less power). Up until a few years ago when they stumbled and ran into problems rolling out their 10nm process.

    • Re:Alder Lake is 10nm (Score:5, Interesting)

      by alvinrod ( 889928 ) on Thursday July 22, 2021 @08:55PM (#61610157)
      AMD is using TSMC’s 5nm process, just not for their current parts. The Ryzen 6000 series using Zen 4 will be fabricated on TSMC 5nm and likely regain the top position for that reason alone, but there will certainly be some architectural improvements as well if that weren’t enough for whatever reason.

      AMD is already having difficulties keeping the market supplied as is with manufacturing being done on the 7nm process node from TSMC which has significantly more wafers available. It also doesn’t have a company (Apple) that probably makes more revenue in certain weeks than AMD does all year bidding against them for those limited wafers either.

    • > this is a pretty apples-to-apples comparison

      Since Apple is now using M1 chips, this is at best a Lenovo-to-Lenovo comparison.

    • A lot of what enabled TSMCs process is ASMLs technology. Intel could put in an order with them and then develop the process around it and I suspect this is what they plan to do when they build out new fabs in Europe which is where ASML is located.

      • Re: (Score:2)

        by swilver ( 617741 )

        Ah, so next month then?

      • Where to do you think Intel buys equipment form currently? Hint: the only place that sells it is ASML.

        • If thats the case how come they have not been able to get to 7nm or 5nm since equipment wise TSMC and Intel should have exactly the same gear? There was a post on Slashdot a while back about how the US Gov blocked the sale of ASML equipment to Chinese companies to hold them back, hence why if having the equipment is all it takes I would think Intel/Samsung would be able to keep up with TSMC?

          • "if having the equipment is all it takes"

            Who said having the equipment is all it takes, besides you?

            • I am not saying that, just curious what else they could be missing. Please educate me?

              • "what else they could be missing"

                A well designed process. There's isn't just one process at a given node that everyone discovers. If you pick a parameter value that turns out to be too hard to actually manufacture, to back out of that choice may mean you have to back out of choices for other parameters and might wind up back at the beginning.

                For example, say you're trying to make the resistance of diffusion contacts lower, so you make them wider. But that reduces the spacing between the diffusion contact an

    • AMD opted not to pay for TSMC's 5nm process

      That is not my take. AMD opted not to pay to be first on TSMC’s 5nm process. Apple secured the majority of current 5nm production with loans of several hundred million dollars and bookings. Intel appears to be in talks to book TSMC's 3nm in a similar fashion. AMD will use 5nm eventually when TSMC has more capacity.

  • without a supported motherboard and ddr5 aint it?

  • Buyer beware. They've been sued for lying before and lost.

    • It was not Intel that posted the benchmark results; however, they are engineering samples so they may not reflect the performance of released products.

  • Somehow this doesn't pass the sniff test for me. If this really is a big.LITTLE like architecture, then on the one hand you're comparing a true 16c32t CPU (the 5950X) with a 16c24t CPU, where 8 of the cores that don't have HT could also be weaker in other regards for efficiency sake. (Though that's not clear yet.) And while the 28% in single core score might be real (we've seen similar numbers the other way in some cases around when Zen3 was released), the 18% in multi-core doesn't seem right here. If that

    • From what I can tell from Intel has sacrificed efficiency for power: "TDP values are maintained at 125W (PL1) and 228W (PL2)." The 5950X TDP is 105W. This is the same trick Intel did to get better single core performance with the 11900K; use MOAR POWR! While the leaker did not overclock the chip, it was water cooled so this might be a considerable change if the next gen from Intel requires water cooling.
      • We have two air-cooled 5950Xs that literally scream like jet engines when they get put under load and run at 85-88C (consistently, without thermal shutdown apparently AMD rates the 5950X at 90C under load and shuts it down at 94C); in contrast, our liquid-cooled 5950Xs run under load in the low 60s and are a whole lot quieter. I think liquid cooling will become a requirement for all of these high-end desktop processors in the near future.
        • Yes the flagship CPUs are getting harder to air cool. The main concern with next gen Intel is that all models require water cooling if Intel feels that prioritizes benchmarks over practical usage.

        • Re: (Score:1)

          by chx496 ( 6973044 )

          I have a 5950X (stock configuration, no OC) in combination with a Noctua NH-D15 SE-AM4 and while that is definitely noticeable under full load (and I wouldn't want to sleep in the same room as that noise), I would not call it a jet engine. Temperatures are also fine:

          Could I get that quieter with liquid cooling and a large rad? Sure. But since the temperatures are perfectly fine even under full load and in my workloads I have short bursts of the fan being noticeable, I don't really mind.

          Now obviously we're

          • Re: (Score:1)

            by chx496 ( 6973044 )

            Sorry, I forgot that there's HTML here and that dropped part of my post. Temperatures are < 40C in idle, and ~ 70C under full load (all 32 threads as close to 100% as SMT allows me, frequency still clocks up to 4.3 GHz there, albeit with a lot of jitter). Takes less than 30s to cool back down to < 45C after the load has stopped and the CPU is back in idle.

    • Yeah, they are comparing it to the wrong processor. The 5900X clocks higher than the 5950x, and the thread count is more appropriate than comparing to the 5950x. Obviously they didn't want to show that, as their smoke and mirrors would be seen through
  • This was first leaked in March [pcmag.com]. The more interesting part not mentioned here is that Alder Lake supports PCIE 5.0 and DDR5 RAM. Heck, we just got PCIE 4 after being stuck on 3 for what seems like forever and it's a huge performance leap. Competition forces vendors to improve their products -- whooda thunk it?!?
  • No thanks, I won't be putting Israel Inside! That 4% of the GDP to an apartheid state that we can eliminate by just not buying expensive over rated Intel chips. I won't be using ethnic cleanser on my keyboard either.

Slashdot Top Deals

10 to the minus 6th power Movie = 1 Microfilm

Close