Here’s a sneak peek at our Ryzen Threadripper results

Bad news: despite consuming the better part of a Coke Zero 12-pack overnight and typing until my fingers went numb, my Ryzen Threadripper review is just not ready for prime time. I am no more happy about this state of affairs than the rest of you. That said, I do have all of the gaming and productivity results I need to finish our review; it's just putting them all in context that's the obstacle at this stage. As a consolation, here's the sum total of my labor:

How we got to this point is the mystery that I will hopefully reveal the answer to later today. Thanks for your patience and understanding.

Comments closed
    • Pancake
    • 2 years ago

    Based on that graph:

    i9-7900X > 1950X for the same money

    i7-7820X=1920X for much less money

    Being completely objective, Intel FTW.

    • anotherengineer
    • 2 years ago

    I think the big question now is, how long will it take software Devs to start making programs/games, etc. able to take advantage of all these cores/threads??

      • Krogoth
      • 2 years ago

      It is already here though for the markets that have a demand for it.

      The real problem is that there’s no killer mainstream application a.k.a normies that demands anything more than four threads. You are hard-pressed to find gaming titles that use more than two-threads effectively.

      It is the main reason why HEDT market has been reducing itself to a small niche.

      • ronch
      • 2 years ago

      That question is asked every time CPUs with more cores come out.

    • Krogoth
    • 2 years ago

    Threadripper lands where it needs to be. It is the first serious challenge to Intel’s massive dominance in the HEDT field since the Nehalem family.

    It is a platform not meant for average joe users either. It is too bad that HEDT market is continuing to be marginalized into a small but profitable niche. It makes Intel’s short-term ham-fisted answers (7540, 7740, 7960 and 7980) to be more baffling.

    • Klimax
    • 2 years ago

    I saw already some reviews.

    Bad news for AMD fans: AMD needs 14 cores to compete with Intel’s 10 cores. (especially when AVX happens)
    i9 will have fun with Threadripper…

    • Cuhulin
    • 2 years ago

    Jeff,

    Thanks for the hard work! I am hanging in there until your review is done.

    Cuh

    • hiki
    • 2 years ago

    I forgive you FOR NOW, just because your articles are soo good, and your XY resume charts are unbeatable (Tomshardware has just copied it, but they don’t get it right).

    • Mr Bill
    • 2 years ago

    We need a new poll…
    (1)Actually went out to look at all the youtube videos with threadrippers?
    (2) I only went to sites with a written review
    (3) I liked the written review sites that also had a video summary
    (4) cheese I can’t make up my mind

      • cygnus1
      • 2 years ago

      2, video reviews of a CPU is either going to be someone reading me a written review or a commercial for something, maybe the reviewed CPU, maybe something else

        • MOSFET
        • 2 years ago

        2 is the only viable option for me.

      • Mr Bill
      • 2 years ago

      I made a poll in the forums.

      • derFunkenstein
      • 2 years ago

      2

      • HERETIC
      • 2 years ago

      22222222222222222222222222222222222222

    • willyolioleo
    • 2 years ago

    Despite being late, TR can still stand out by doing more thorough workstation benchmarks. Most other sites only gave productivity benches a minor glance.

      • Ethyriel
      • 2 years ago

      Man, I would love someone to figure out a good virtualization benchmark. I wonder if VMmark is any good.

    • Ethyriel
    • 2 years ago

    Please please please tell me you’re testing with both NUMA and SMT enabled. Of what I’ve read so far, very few are even mentioning NUMA, and I haven’t seen anyone but Anandtech test multiple combinations. But they tested UMA+SMT and NUMA-SMT.

    • brucek2
    • 2 years ago

    Anyone know of a simple Windows utility that can easily record the amount of time a system spends at CPU frequency cap (1+ cores) vs. core capped vs. neither?

    I know this information is available via parsing the logs of general purpose monitoring utilities, I’m just wondering if it’s possible to access more conveniently?

    While I bet most of us have guesses as to how our systems spend their days and therefore which of the upcoming options may be most attractive, it’d be interesting to actually see the hard data.

    • Redocbew
    • 2 years ago

    All of you that are using TR as an abbreviation for Threadripper are confusing the hell out of me.

    Oh, and Jeff is my new favorite person on the Internet.

      • Wirko
      • 2 years ago

      “Totally Recommended!” will be the Terminal Revelation of this Thorough Review.

    • bfar
    • 2 years ago

    Thanks for all the hard work TR!

    Had high hopes for Zen, but wouldn’t have believed we’d ever see a chart like that in the ultra high-end so soon. Let the prices fall!!

    • soccergenius
    • 2 years ago

    Checking out some other sites’ reviews, I’ve come to realize that Tech Report has truly spoiled me with their concise and easy to read charts and graphs.

    I can’t believe some sites’ editors think it’s okay to make vertical charts so tall and poorly labeled that you have to constantly scroll up and down to identify each result. Ugh.

    • rechicero
    • 2 years ago

    I’d rather have a good review a little late than a hack job on time. Thanks for the efforts and don’t worry!

    • deruberhanyok
    • 2 years ago

    I know it’s still early times, but damn, AMD has really shaken things up with Zen.

    It’s about time!

    • gerryg
    • 2 years ago

    To go along with charts like this it would be interesting to see a spider chart (aka radar chart) showing features, as objectively as possible of course. E.g. pcie lanes, memory speed support, memory capacity, standards support, etc. In other words, the things not found in the pure performance and price numbers that might swing a choice one way or another. Especially when there is apparent parity between two choices.

    • TravelMug
    • 2 years ago

    Shame on AMD that they didn’t send you a sample on time and also that they didn’t send you that personalized package they’ve send to everyone and their dog!

    I have to admit this is somehow irrationally upsetting me, maybe because of some loyalty to the site as a reader since it’s inception? I don’t know, but it shouldn’t have happened for sure!

    End of rant.

      • Shobai
      • 2 years ago

      Can confirm that neither myself nor my dog have received a personalised package from AMD, so…

    • ronch
    • 2 years ago

    So it doesn’t look like Threadripper is a repeat of the Grand Slams that Ryzen 7 and Ryzen 5 were?

      • ermo
      • 2 years ago

      Depends on your workload it seems. It won an editor’s choice award at PCPer for instance.

        • ronch
        • 2 years ago

        PCPer loves AMD quite a lot, I gather. I remember them rooting for AMD back in the FX days, releasing a video that pretty much says FX isn’t so bad.

          • derFunkenstein
          • 2 years ago

          Has something to do with Ryan Shrout’s old website, [url=https://www.extremetech.com/extreme/48166-amdmbcom<]AMDMB[/url<]. Anecdote: in the Slot A era, he put out a call for anyone who wanted to donate their Quake CD so he could benchmark Athlon's with it. He offered up a free AMDMB shirt in return. I had like 3 copies somehow, so I sent him one. Never did get the shirt.

    • ermo
    • 2 years ago

    For reference, a Ryzen R5 1400 is faster in all metrics than an FX-8350 (!).

    The TR 1950X is maybe around 5x faster than an FX-8350 (in render benchmarks) and costs around 10x as much.

    However, if I’m not mistaken, the FX-8350 has better perf/$ at current prices in render tasks (as long as electricity prices are low)? That’s … a bit crazy tbh.

    And now to the point: If you’re looking to build a system for rendering/compilation/virtualization on the cheap, the second hand Xeon market might not be a bad place to look with respect to perf/$?

    (downvote away).

      • just brew it!
      • 2 years ago

      [quote<]The TR 1950X is maybe around 5x faster than an FX-8350 (in render benchmarks) and costs around 10x as much. However, if I'm not mistaken, the FX-8350 has better perf/$ at current prices in render tasks (as long as electricity prices are low)? That's ... a bit crazy tbh.[/quote<] It's not crazy, it's just a meaningless comparison. Even if you ignore electricity costs, the perf/$ edge completely evaporates once you factor in the expense of building 5x as many systems to get equivalent throughput.

        • ermo
        • 2 years ago

        The FX-8350 comparison was meant as a curious aside though I still think it’s awesome that the lower-binned 4-core/8-thread R5 1400 running in the low 3GHz range beats the crap out of the FX-8350 across the board. This more than anything shows what a mistake AMD made with the BD/PD CMT approach.

        */me clings to his FX-8350 like a relic of a bygone era*

        I find it curious that no-one commented on my used Xeon comment. Oh well.

      • MOSFET
      • 2 years ago

      I just finished the PCPer review (mainly out of boredom and a little curiosity – I enjoy their podcasts when I’m bored, surprisingly) and I must gloat a little. I’ve used and loved many processors in the last few years (including several FX and lately a Ryzen 5 and Kaby Lake), and their results really show what a powerhouse the Core i5-7600K is (especially perf/$). Of course, that’s why I bought one the second day they were available!

      • Laykun
      • 2 years ago

      This ignores time cost of doing work. While there is electricity cost there is a much greater time cost on tasks that can’t be farmed out to a cluster of machines. Considering that you’d primarily buy these CPUs to make money, money will be lost on single CPU systems if the CPU impedes the progress of your work (waiting for renders/simulations, etc). Over the life time of the CPU these costs build up.

      In terms of having a cluster of machines, considering physical restrictions and IT cost restrictions if you’re going to build a farm of FX-8350 machines why not just build a farm of Thread ripper machines? Generally space and networking capacity problems are going to plague you before the base cost of the CPU does.

      TL;DR these CPUs are not priced based on regular consumer concerns of price/performance, and for good reason. The biggest cost in computing work environments is rarely capital expenditure, it’s usually people’s time that costs the most.

        • ermo
        • 2 years ago

        The point I was *actually* making is that for someone doing rendering/virtualization/compiling on the side, the used Xeon market is probably not a bad place to look if you’re on a tight budget.

        The FX-8350 comment was meant as a curious aside. It’s obvious that it doesn’t scale out at all.

    • ludi
    • 2 years ago

    So 10 cores and 20 threads in the 7900X put it pretty much even with the 1950X sporting 16 cores and 32 threads? Be interesting to see the individual test numbers feeding into that.

    The 1920X is sure sitting pretty, though.

      • drfish
      • 2 years ago

      I don’t know, if the chip is paying your bills then the 1950X or 7900X are going to be worth the extra coin. If it’s just for kicks, then the 7820X looks like smarter money than the 1920X. Plus, I don’t think AMD has a platform cost advantage this time around to help… Of course, we need to see the rest of the review to know for sure.

    • DragonDaddyBear
    • 2 years ago

    I think TR vs i9 is really going to depend on your needs.

      • cygnus1
      • 2 years ago

      I think that’s always the case. But judging by initial benchmarks popping up, they’re close enough in general performance that my default recommendation to people wanting HEDT, is going to be TR. The underlying platform is just better with ECC RAM and way more PCIe lanes, especially if you’re looking at the 1920x price range.

        • Sahrin
        • 2 years ago

        Yep.

        It’s pretty clear that Kaby Lake-X and SL-X were going to win single threaded benches, but if you’re buying a HEDT/SHED then this is an implicit statement that there are other things that matter more than single thread.

        The question was always going to be, does TR get close enough in single thread that I can afford to trade off a few percent loss there to get access to *double* the total execution resources.

        For Ryzen, this question was difficult – single thread matters more to those users.

        But for TR, it’s pretty unequivocal at this point: unless you are so phenomenally highly dependent on one of a handful of workloads where SL-X has a lead, the answer is threadripper.

        A single Threadripper 1950X has 96 ALU’s and 64 128-bit FPU’s. Typing that makes me smile. I don’t know what I am going to do with that much math hardware, but I do know that whatever it is will be done fucking quickly.

          • cygnus1
          • 2 years ago

          [quote<] I don't know what I am going to do with that much math hardware, but I do know that whatever it is will be done **** quickly. [/quote<] Lol, it's a great time to be alive, isn't it?

            • Sahrin
            • 2 years ago

            Just so.

          • Klimax
          • 2 years ago

          I doubt answer is actually Threadripper. It needs more cores at higher frequency to just match 10 cores by Intel. Never forget cent cheaper, pound foolish. Especially with i9 still incoming.

          Actually my recommendation would be not SK-X or TR, but waiting for Coffee Lake-X. 14nm+ will do a number on both current HEDT platforms. (Note: Obviously if one can wait a bit)

        • Beahmont
        • 2 years ago

        TR Platform is also more expensive. ECC tracings and 16 more PCIe lanes don’t come free you know. PCIe lanes increases add to the price at a worse than linear rate. If you see a 44 lane vs 60 lane motherboard for the same cost to the consumer, someone is subsidizing that 60 lane board.

          • cygnus1
          • 2 years ago

          Or charging a premium for those 44 lanes from Intel…

      • gerryg
      • 2 years ago

      Tech Report (TR) is my *only* need. πŸ™‚

      • Klimax
      • 2 years ago

      Correct. SSEx only workloads will be fine on either chip (but given AMD needs 4 extra cores to just match 10 cores by Intel, it might pose problem for users)
      As soon as AVX is an option, Skylake-X wins. Once you get AVX-512 (x264 BTW supports it in recent builds – beware of reviews using older version or not stating it outright!) it gest funny fast.

    • Jeff Kampman
    • 2 years ago

    For the folks who are going to lose their minds because TR is not 1000% better than the Core i9-7900X in this chart in the interim: [url<]https://twitter.com/jkampman_tr/status/895645729972080640[/url<]

      • just brew it!
      • 2 years ago

      Interesting. The DAW issue could be a memory latency/prefetch thing. I figure if you’ve got a workload with a lot of channels and effects plugins, the memory access patterns will be close to random.

      • chuckula
      • 2 years ago

      The DAW stuff doesn’t surprise me because unlike rendering it’s got more work involved than just having in-cache data that gets crunched on a lot of cores.

      As for rendering & media encoding, those workloads are ripe for AVX-512 [downthumbs welcome for mentioning it I know]. While AVX-512 might be a “niche” so are workloads like encoding and rendering that scale well with a crapload of cores too, so they are both targeting the same niche.

        • Klimax
        • 2 years ago

        Latest x264 build support AVX-512…

      • jihadjoe
      • 2 years ago

      Anandtech looked into Creation vs Gaming modes, which actually does something with Threadripper. If it’s latency-sensitive then DAW stuff might be better off being run in gaming mode.

        • just brew it!
        • 2 years ago

        Yes. If you have multiple latency-sensitive threads, and your OS does a decent job of NUMA scheduling, Gaming mode is a potential win.

          • JustAnEngineer
          • 2 years ago

          Interestingly, there are two switches being toggled with the change from the default Creator mode to Gaming mode: symmetric multi-threading is turned off and non-uniform memory architecture is exposed to the OS rather than appearing as uniform memory architecture. [url=http://www.anandtech.com/show/11697/the-amd-ryzen-threadripper-1950x-and-1920x-review/4<]Anandtech[/url<] found that each switch was independently controllable in their motherboard's BIOS.

    • southrncomfortjm
    • 2 years ago

    I think someone needs to do a time lapse documentary of exactly what Jeff does in order to test two CPUs in such a short time period. I’d be interested to see just how much time is spent switching out hardware v. running the benchmarks v. analyzing the data v. writing up the review v. sleeping.

      • Wirko
      • 2 years ago

      The results of profiling could be presented in a nice pie chart, “Inside the second”.

      • kuraegomon
      • 2 years ago

      I’m just gonna go ahead and guess that there was no sleeping. Except maybe while standing up over the test bench… with eyes open.

      Funk already said it, but it bears repeating: None of the _thorough_ reviews (and there won’t be many of those) popping up today could have been produced in two days. Or anything close to it.

    • boomshine
    • 2 years ago

    I have watched Hardware Unboxed review and the price to performance ratio compared to TR has a somewhat significant difference hmmm…

      • Jeff Kampman
      • 2 years ago

      Hardware Unboxed isn’t running my test suite.

        • southrncomfortjm
        • 2 years ago

        This is why you actually have to read the reviews people. Numbers alone don’t tell the whole story.

    • thx1138r
    • 2 years ago

    Minor note, but do we really need the different memory flavors of the i7-7820X and the i9-7900X in the graph. The results are almost exactly the same so it just ends up cluttering up the graph. I would much prefer if the graph had some additional older-gen chips for comparison/upgrade purposes.

      • Jeff Kampman
      • 2 years ago

      You are correct, it’s a waste of time that I won’t be repeating. But the numbers were there, so…

        • derFunkenstein
        • 2 years ago

        But it was worth it for this round just to put the notion to bed. πŸ˜€

        • SomeOtherGeek
        • 2 years ago

        Yes and you would not have known that unless you tested it. So, I’m glad you did and we now know the you did.

      • Beahmont
      • 2 years ago

      Well you know sir, while the numbers don’t change for Intel based on RAM speeds, it’s hardly a waste of time this time because now we have proof that the numbers don’t change. That’s really important going forward for Tech Report’s future testing and very important for Tech Report gerbils wallets. Considering I’m in the market for an upgrade to my 2008 built C2Q, I personally really appreciate the fact that I know that DRAM faster than 3200 is just money spent poorly.

      TL;DR: Getting a null result in this case was just as important as getting a getting any result because we just didn’t know the answer.

    • maxxcool
    • 2 years ago

    Since every other site will be running different ram, different ram timing, and different ram speeds i’d caution everyone to read multiple reviews and not twist your undies on this one single slide.

      • chuckula
      • 2 years ago

      True, there’s plenty of video blogs with laser-engraved and numbered Threadrippers for the people who are butthurt over these more in-depth numbers.

    • southrncomfortjm
    • 2 years ago

    We see what kind of RAM was required to get the results for the big Intel chips, but not for the two Threadrippers. Can you shed some early light on that Jeff?

      • Jeff Kampman
      • 2 years ago

      DDR4-3200.

      • DPete27
      • 2 years ago

      According to [url=https://techreport.com/news/32343/updated-wanted-for-review-amd-ryzen-threadripper-cpus<]this article[/url<] they tested with a 3600MHz kit of RAM on the ThreadRipper platform. Although IIRC Ryzen can't go higher than 3200MHz.

    • BehemothJackal
    • 2 years ago

    Interesting results.

    Techspot published their findings and found the 1950X to be significantly faster than the 7900X.

    “Because of pricing, core count becomes somewhat irrelevant: the Core i9-7900X and Threadripper 1950X both have a lot of cores and they both cost a grand. The key difference being that the 1950X often delivers 20-30% more performance while consuming almost 10% less power under full load.”

      • chuckula
      • 2 years ago

      10% less under full load?

      That’s not what PC Perspective found: [url<]https://www.pcper.com/reviews/Processors/AMD-Ryzen-Threadripper-1950X-and-1920X-Review/Power-Consumption-and-Overclocking[/url<] Or what a very detailed analysis of multiple workloads at THG found: [url<]http://www.tomshardware.com/reviews/amd-ryzen-threadripper-1950x-cpu,5167-14.html[/url<] Oh and Anandtech: Yup, didn't find that either: [url<]http://www.anandtech.com/show/11697/the-amd-ryzen-threadripper-1950x-and-1920x-review/19[/url<]

        • BehemothJackal
        • 2 years ago

        Yup, check it out:

        [url<]https://techspot-static-xjzaqowzxaoif5.stackpathdns.com/articles-info/1465/bench/Power.png[/url<]

        • credible
        • 2 years ago

        Chuck is getting desperate posting a Toms link lol.

          • chuckula
          • 2 years ago

          Fascinating, I don’t remember you complaining when that exact same website posted complaints about the 7900X using lots of power when overclocked.

          Where is your immature snarky post attacking Tom’s Hardware in the comments of this story?
          [url<]https://techreport.com/news/32216/report-skylake-x-cpus-prove-hard-to-cool-thanks-to-tim-change[/url<] I just can't seem to find it... funny. I mean, I know you didn't miss the story since you went out of your way to comment in the thread in a way that seems to imply you think Tom's was right. [url<]https://techreport.com/news/32216/report-skylake-x-cpus-prove-hard-to-cool-thanks-to-tim-change?post=1043924[/url<]

            • raddude9
            • 2 years ago

            So you’re still here, where’s my apology?????

            • chuckula
            • 2 years ago

            OK: I’m sorry that I have to deal with a person like you who apparently has major problems with this website but still feels the need to come here anyway.

            Question: Why the hell isn’t the 1950X 60% faster across the board compared to the “failed” 7900X?

            • raddude9
            • 2 years ago

            I have a problem?

            You said:

            [quote<]Incidentally, where's your apology for insulting people who spotted the RyZen bug AMD just acknowledged?[/quote<] Now I've looked at the page: [url<]https://techreport.com/discussion/32125/amd-epyc-7000-series-cpus-revealed[/url<] I can't see any post where I insulted anyone. Who did I insult? or are you just making things up. Case in point, you put the word "failed" in quotes. Who said the 7900X failed? I didn't. Also, I have no major problems with this website, again, that's your wild assumption.

            • Redocbew
            • 2 years ago

            NERD FIGHT!

      • raddude9
      • 2 years ago

      Ars went even further and said “Better than Intel in almost every way
      Cheaper, faster, and more feature-rich than Skylake-Xβ€”what’s not to love”

      [url<]https://arstechnica.co.uk/gadgets/2017/08/amd-threadripper-review-1950x-1920x/[/url<] [edit]wow, that's a lot of downvotes just for reporting some facts[edit]

        • chuckula
        • 2 years ago

        Oh year, Arstechnica… where you go for Geekbench.

        Incidentally, where’s your apology for insulting people who spotted the RyZen bug AMD just acknowledged?

          • raddude9
          • 2 years ago

          Ah-ha, the source of the downvotes is the same person who complains at being down-voted for reporting facts. Oh, the irony.

          And who did I insult, and in exactly what post. Please send me a link and evidence of insult. Until then I’ll assume this is yet another one of you strawmen.

            • raddude9
            • 2 years ago

            Still no details on my alleged insulting behavior I see.So how about an apology for the accusation? Clearly you have been mixing me up with someone else.

        • Redocbew
        • 2 years ago

        [quote<]Threadripper embraces the enthusiasts, the system builders, and the content creators that shout loud and complain often, but evangelise products like no other.[/quote<] Obviously this dude thinks that's his job also when writing a review on new hardware. Plus, they're including 3DMark and Geekbench as "productivity" tests? Blender makes an appearance as well which we all know AMD has themselves used heavily leading up to all this, but at least that's a real application you can use to actually get something done.

      • maxxcool
      • 2 years ago

      That slide literally means nothing.. it could game like crap and double the workload in SQL or Maya and the chart would look the same.

        • demani
        • 2 years ago

        But it says it’s “non-gaming”. So that isn’t it.

          • maxxcool
          • 2 years ago

          hmmm doesn’t say either way..

          I do have all of the gaming and productivity results I need to finish our review; it’s just putting them all in context that’s the obstacle at this stage. As a consolation, here’s the sum total of my labor

            • Anovoca
            • 2 years ago

            “Price versus performance – non-gaming applications”

            -Written right on the chart itself, not in the article.

      • Jeff Kampman
      • 2 years ago

      Techspot isn’t running my test suite.

        • maxxcool
        • 2 years ago

        Well played sir, and accurate πŸ™‚

        edit: played*

        • DoomGuy64
        • 2 years ago

        Hopefully the article explains why your test suite is better than everyone else who has been showing 20-30% more performance, and also discusses the differences between UMA/NUMA, because apparently NUMA mode mitigates some of the issues associated with UMA mode depending on the scenario.

        Even if the performance is completely the same, both platforms are radically different and offer different perks. AMD has more PCIe lanes, Intel has AVX, etc, and the review should point out what platform is better suited for what usage case. Performance is [i<]not[/i<] the whole picture with these platforms.

          • NeoForever
          • 2 years ago

          You made good points. Not sure why the downvotes without explanations. It’s simple, if a test suite causes the score to show outside of a standard deviation. It’s worth explaining why that happened, or why the test suite is a valid one (or even better) to use.

      • Klimax
      • 2 years ago

      Yet it needs extra 4 cores to actually achieve that. That’s why AMD priced relatively low Threadripper. And not because they are charity…

        • Amiga500+
        • 2 years ago

        So it needs 60% more cores to deliver 20-30% more performance?

        I’d take that.

        With the depressed clock speeds that the higher core’d i9s will have to fit within the power budget, assuming that the i9s will automatically be significantly quicker is quite the leap.

          • Klimax
          • 2 years ago

          Only 200MHz delta between 7900x and 7940x. Not big enough to eliminate IPC delta between Intel and AMD cores.

    • chuckula
    • 2 years ago

    One interesting extrapolation from that slide is that will be interesting to test when the higher core count Skylake X chips come out is the relative performance boost from the 7820X –> 7900X compared to the 1920X –> 1950X.

    The 7820X –> 7900X is a 25% core count boost and basically identical clock speeds across the board* including identical turbo boosts.

    The 1920X –> 1950X is a 33% core count boost although the 1950X has slightly slower base clock speeds.

    It’s interesting that the relative scaling from the 7820X –> 7900X is actually greater than for the 1920X –> 1950X in that chart.

    * See boost clocks here: [url<]http://www.anandtech.com/show/11698/intel-finalizes-skylakex-processor-specifications-18cores-44-ghz-165w-on-september-25th[/url<]

      • derFunkenstein
      • 2 years ago

      You can’t tell anything without the details. The 7820X isn’t double the speed of the 7740X even though they’re really close in clock speeds. It’s more like 60% where it should be 90+. So we should all know by now that scaling with many cores isn’t a linear thing. Scaling from more cores to even more cores is (apparently) diminishing returns for both. Since AMD is starting at a higher core count in the first place, having even worse scaling is expected, right?

        • chuckula
        • 2 years ago

        I just said it was interesting and something to look out for as the second wave of Skylake X comes onto the market. Since there is no second wave of Threadripper we’ve seen what it does in this chart.

    • Lianna
    • 2 years ago

    Thank you for the great work. We’re waiting patiently, because we know it will be worth the wait.

    • chuckula
    • 2 years ago

    Looking at the prices and looking at the performance, it’s pretty clear that for all the talk of every Intel product being some “reaction” to AMD that the pricing on ThreadRipper was very much intentionally done to allow AMD to sell it for the highest price they possibly can without looking like it’s a bad deal even compared to Intel’s “overpriced” offerings.

      • DPete27
      • 2 years ago

      Nothing wrong with that. It would be silly not to price your product within the established price/performance curve.

        • chuckula
        • 2 years ago

        I never said anything was wrong with it.

        I was just noting (correctly) that AMD didn’t set these prices in a vacuum and isn’t trying to go out of its way to sell these things at a loss.

        In other words, AMD is acting like Intel.

          • DPete27
          • 2 years ago

          Even at that pricing structure AMDs margins are assumedly lower than Intel’s.

            • chuckula
            • 2 years ago

            I’m sure they are.

            And there’s nothing wrong with Intel getting margins.

            • Sahrin
            • 2 years ago

            How do you figure? AMD’s yields are better than Intel’s are. They also have less development overhead (simpler interface, less money invested in compiler cheat development).

            • chuckula
            • 2 years ago

            [quote<]How do you figure? AMD's yields are better than Intel's are. [/quote<] Please put the internal documents from both GloFo and Intel's fabs into a public Drop Box account so we can verify that.

            • just brew it!
            • 2 years ago

            [quote<]How do you figure? AMD's yields are better than Intel's are.[/quote<] And you know this because...? [quote<]They also have less development overhead (simpler interface,[/quote<] Not even sure what you're referring to here. [quote<]less money invested in compiler cheat development).[/quote<] I'd be willing to bet the compiler team is pretty small in the grand scheme of things.

      • ronch
      • 2 years ago

      Some people believe Intel really is deliberately keeping AMD alive, even if barely sometimes.

        • chuckula
        • 2 years ago

        Zombie AMD!

      • Sahrin
      • 2 years ago

      I’m about as big a conspiracy theorist as there is and even I can’t figure out what you’re accusing people of here. AMD…priced their processors to the competition? *squints*

      • just brew it!
      • 2 years ago

      …and your point is what? That AMD is trying to make a profit and return value to their shareholders? That’s what businesses are supposed to do. We can quibble over whether they’d actually make more money by dropping the price a bit (and thereby increasing volume), but that’s all speculation.

      TBH I’m more interested to see where the relative [i<]street[/i<] prices settle in after TR has been on the market for a month or two. That's a lot more meaningful than launch MSRP.

        • ptsant
        • 2 years ago

        I expect $800 for the top model and $600 for the 1920X. There will also be cheaper motherboard models (instead of the ultra-deluxe that are currently available).

        Ryzen has already been discounted quite a bit where I live. It already hovers steadily 10-15% below MSRP with the occasional special offer/rebate even lower than that.

        If RAM were less expensive I would have been seriously tempted…

    • just brew it!
    • 2 years ago

    Based on that, I’d say AMD is being too optimistic with their list pricing. They need to undercut Intel at least a little, to win back market share.

    Guess well see where street prices end up a few weeks in.

    Edit: The 1920X seems to be reasonably well-positioned. I was referring to the flagship 1950X with my above comment.

      • derFunkenstein
      • 2 years ago

      The 1950X very well could be poorly priced, but the 1920X stuck out like a sore thumb. Maybe the scatter plot is just too low vertically, but it looks like it’s barely any faster than the 720X while being $200 more expensive.

        • just brew it!
        • 2 years ago

        If Intel doesn’t have anything in that price range some people will be willing to pay the extra $200 to get the performance bump.

          • chuckula
          • 2 years ago

          AMD is lucky that Intel is averse to price cuts in the same way that vampires are averse to garlic.

          That way AMD doesn’t have to worry about a reslotted 7900X at $800 and 7920X at $1000.

            • derFunkenstein
            • 2 years ago

            That would kill Threadripper and Intel knows it. Intel needs AMD to at least exist to keep anti-trust probes off their back.

            • cynan
            • 2 years ago

            That’s one theory.

            Another is that Intel could be worried about loosing mind share if they were seen to cut prices on their products to match those of AMD’s. That is, Intel could very well believe that it’s a slippery slope between matching (for lack of a better word) AMD on price and loosing the general consensus among consumers that Intel’s CPUs are, generally, a superior product to AMD’s.

            In other words, once Intel starts chasing AMD on price, they sort of admit – and tell their loyal customers – that AMD’s products are worthy competitors – as good as theirs. From that point, the loss of market share from the majority of consumers who buy these products (or computers containing these products) more casually may be a legitimate threat.

            • derFunkenstein
            • 2 years ago

            That would be a very Apple-y way of thinking. That could also be true, though.

            • cygnus1
            • 2 years ago

            Losing…

            Not loosing… there is a big difference

            • Klimax
            • 2 years ago

            It’s more likely Intel will use upgraded 14nm (either + or ++) to upgrade frequency like it is seen on Coffee Lake chips and rebase their segments. Won’t be losing much on ASP while keeping AMD barely above water.

            BTW: Linus had some interesting observations about Intel’s core and die sizes:
            [url<]http://www.realworldtech.com/forum/?threadid=170530&curpostid=170562[/url<]

          • derFunkenstein
          • 2 years ago

          Yeah, that does seem to be [url=https://techreport.com/news/32360/intel-reveals-full-core-i9-specs-and-launch-dates<]a gap[/url<] in Intel's lineup. They'd have to make a lower-clocked 10-core CPU, because I have a feeling the 8-core CPUs are already hitting their top end in their given thermal range.

        • cynan
        • 2 years ago

        I’d be inclined to agree that AMD”s TR seem a tad overpriced compared to their usual tactic of undercutting the competitor. And from this plot, could use a 10% price cut right off the bat. Though, you could almost say something similar – that the 7820x is overpriced vs the 1800x and the 7740x.

        The higher priced/core offerings seem to be poorer bang per buck in general in this average performance/price plot. Heck, a Ryzen 1700 at street prices would probably smoke everything on this chart from a pure price/performance standpoint.

        The 7820x does look like a pretty nice all around sweet spot for HEDT. AMD’s 1900x will have to perform within la few % delta of the 1920x to be remotely competitive at $550.

        That all said, we need to dive down into what performance numbers are actually being averaged here before any concrete opinions can be made. Bring on the review!

          • derFunkenstein
          • 2 years ago

          You’re right about the details, because some tests are more important for certain prospective buyers than others.

          And you also make a good point about the 1800X – we’re not in “value” territory here, and it’s easy for me to forget that. If you need more than eight cores and 16 threads, you probably are shopping for absolute performance in a given workload. Threadrippa might be better for some people than it is for others.

          • just brew it!
          • 2 years ago

          [quote<]The higher priced/core offerings seem to be poorer bang per buck in general in this average performance/price plot.[/quote<] This has [i<]always[/i<] been true. Flagship models carry a price premium out of proportion to the actual performance numbers.

            • cynan
            • 2 years ago

            Sure. I was merely responding to the comment that the 1920x, a 12-core part, “stuck out like a sore thumb” as poor value compared to the 8-core 7820x.

            Another minor point is that the Threadrippers offer 64 lanes of PCIe while the 7820x only offer 28 – and features like this, which can be pretty crucial for certain applications aren’t factored in in this chart.

            • Beahmont
            • 2 years ago

            As far as I’ve been told, ThreadRipper is60 lanes in a 2×30 manor*. There is no cross talk bar on package. It’s 30 lanes to one chip and 30 lanes to the other chip with 2 lanes from each chip going to the chipset. When one core/CCX/chip needs to talk to a device attached to the other chip’s PCIe lanes, it has to spend extra time talking to the other chip to access the other lanes. I’ve been told that this extra latency can in some circumstances be as high as a main memory fetch, though it’s usually shorter though still slower than an L3 fetch. That also has major implications to people who want to use all 60 lanes of PCIe because you’re going to have cross CCX and chip-to-chip latency that won’t be found on Intel chips.

            Zen is a decent and interesting architecture, but AMD had to make several design compromises with very real drawbacks to bring these chips to market.

            Note*: This could be wrong and it’s a 32 and 28 configuration, but I’ve been told though not been able to independently verify that AMD went with a 2×30 config.

            • Waco
            • 2 years ago

            Parking processes that need to hit resources off of that CPU’s NUMA domain is nothing new, though. Even Windows handles this automagically as best it can.

            • cygnus1
            • 2 years ago

            I think I read it has multiple x16 PCIe roots, 2 per CCX. Depending on the board layout, different device to device communication could have very different latency even without having to go cross CCX.

            • cygnus1
            • 2 years ago

            Found it: [quote<] [url=http://www.anandtech.com/show/11697/the-amd-ryzen-threadripper-1950x-and-1920x-review/2<] AMD allows each of the PCIe root complexes on the CPU, which are x16 each, to be bifurcated down to x1 as needed, for a maximum of 7 devices. [/url<] [/quote<]

      • chuckula
      • 2 years ago

      The 1920X is definitely a better deal although on a pure price-performance metric the 7820X isn’t a slouch by a longshot, not to mention that 7820X has substantial OC capabilities to close the gap.

      When the full review gets published I expect to see a bunch of inconsistencies in individual benchmarks where TR looks better in some and Skylake X looks better in others.

        • Convert
        • 2 years ago

        Not sure about the down-votes, but I definitely agree with the comment about the 7820x.

        I don’t consider overclocking at all, and even then I’d feel much more comfortable buying something in the 7820x price range with similar performance.

        You’re probably also right about the performance of the two being all over the map with some of the tests. The 1920x having so many more cores/threads should definitely give it an advantage in some tests.

        • Klimax
        • 2 years ago

        I predict that the only testy where TR will have any sizeable lead will be SSEx only. (Frequency and boosted SSE units will help, but cannot do a thing about AVX – like Cinebench versus Blender)

      • Duct Tape Dude
      • 2 years ago

      The above chart doesn’t take power consumption into account. AMD may actually be quite efficient this round.

      • DragonDaddyBear
      • 2 years ago

      I don’t think so clear cut. The a lot of people that buy these chips have very specific workloads in mind that give them an ROI. Those that need cores and/or PCI-E lanes are going to go with ThreadRipper and those that need lower memory latency and/or AVX-512 are going to go Intel.

        • just brew it!
        • 2 years ago

        Agreed. Generalizations by definition aren’t going to cover specific use cases.

        • Klimax
        • 2 years ago

        No need for AVX512, just AVX got some boost.

      • Sahrin
      • 2 years ago

      I’m very pleased with the pricing, honestly. I would probably be willing to pay a bit more. Based on the results and the platform, if anything 7900X is overpriced (rather than TR being underpriced).

    • dodozoid
    • 2 years ago

    So it seems to me that for the first time after a long while, you have to do research to find what CPU fits your workload better.

      • cynan
      • 2 years ago

      And whatever other conclusions may be drawn once the NDA lift review dust settles, this alone makes TR a legitimate success for AMD.

      (Though I’m sure AMD wouldn’t mind terribly if consumers were convinced to actually buy a whole a bunch of ’em too).

    • derFunkenstein
    • 2 years ago

    Jeff, AMD handicapped you something fierce. There is no shame in not being done. We’ll read it when it’s ready.

    That said, I know the rest of the internet doesn’t work that way and that TR will take a hit in traffic. It’s a shame that AMD handicapped you something fierce.

      • trieste1s
      • 2 years ago

      What did AMD do?

        • derFunkenstein
        • 2 years ago

        They gave everyone else an extra week, somehow. Jeff didn’t get his Threadrippa until Tuesday.

        • K-L-Waster
        • 2 years ago

        Didn’t send review CPUs until several days after the other review sites already had theirs.

        • DragonDaddyBear
        • 2 years ago

        It’s more what they didn’t. They DIDN’T give them a sample until just a few days ago and everyone else had over a week.

      • wizardz
      • 2 years ago

      Exactly!
      Jeff, its going to be ready when its ready. I am eagerly waiting for your review, not because i am eyeing a purchase of a TR cpu, but simply to read the best hardware reviews that exists.

      Shame on AMD…

      • CScottG
      • 2 years ago

      YUP!

      • Mr Bill
      • 2 years ago

      What derFunk said. Happy to wait for the best review.

      • mcarson09
      • 2 years ago

      Jeff was late with other reviews and AMD figured that sending Jeff stuff was pointless.

    • chuckula
    • 2 years ago

    YOU ARE SUCH A TEASE KAMPMAN!

    Thanks for the early preview.

Pin It on Pinterest

Share This