Rumor: Zen release date, clock speeds, and pricing revealed

The rumor mill can't seem to exhibit any mindful self-control when it comes to AMD's upcoming Zen CPUs. The latest accounts suggest the specifics enthusiasts have been waiting for: release dates, clock speeds, and pricing. The trail of evidence is pretty nebulous here, so try to follow along: Guru3D reports that a user of Chinese internet search company Baidu claims to have access to an internal "micro-newsletter" from hardware retailer Maxsun.

Maxsun may be an unfamilar name to Western readers, but the company is a division of Palit Microsystems. Palit is one of the world's largest manufacturer of graphics cards by volume and the parent company of the more familiar GALAX and Gainward brands, as well as the OEM for PNY graphics cards.

The leaker claims that AMD will release the first round of Zen-based chips on January 17. Guru3D mentions that this would fall in line with a CES 2017 product launch the week of January 5. All chips in the first round of releases are said to be two-module, eight-core, 16-thread parts with 95W TDPs. The chips will slip into the same unified AM4 socket as AMD's latest Bristol Ridge APUs. The source claims the first Zen chips will be clocked between 3.15GHz and 3.3GHz, with 3.5GHz boost clocks.

The last part of the leak concerns pricing. The leaker claims the most expensive of the first Zen chips will be priced at 1500¥ to 2000¥, or a range between $220 and $290. These amounts seem very competitive for what is supposedly a top-end CPU when compared to the prices of Intel's current range of desktop CPUs. Look to our previous coverage of Zen rumors for performance data.

Comments closed
    • sophisticles
    • 3 years ago

    Only AMD knows how much their products are worth, if they are really pricing their top of the line consumer class dual module 8 core 16 thread cpu at $290, which is about $30 more than you can get a 6700k from Microcenter for.

    This tells me one of following:

    1) AMD has no idea how to run a business much less make money.

    2) The previous video that supposedly showed an 8 core 16 thread Zen beating an 8 core 16 thread Broadwell-E is a fake and Zen will be another Bulldozer.

    3) The claims related to pricing are way off.

    I don’t see any way that AMD releases a sub $300 cpu that is capable of beating a near top of the line Intel E class 8 core 16 thread cpu that Intel sells for $1000.

    Of course, if this is the case, maybe AMD, rightfully in my opinion, believes that it would lead many OEM’s, big business customers and DIYers to switch to using AMD chips by the droves.

    I know I would jump at such a chip.

      • Zizy
      • 3 years ago

      On the other hand, this is 10-15% more than what 6700k costs, so it should be at least 25% better, which is very hopeful.

      As for beating BD-E -> maybe in some workloads, while it is behind in others. Say i7 6800K (6C/12T) competitor on average would be more expected. And as those cost 400 or something, having comparable AMD stuff costing 300 would be quite expected.

      • blastdoor
      • 3 years ago

      I think it’s possible that 8 core Zen performs more like 6 core Broadwell, and they’ve priced accordingly.

      • ronch
      • 3 years ago

      1. No. AMD knows exactly where to position their products and how to price them. In fact, they’ve historically tried to price higher than what their products are worth (e.g. FX-9590) and they know how to price accordingly when they know their products are worth it (early Athlon 64 years). They’re not a bunch of morons. This business is extremely grueling. Out-engineering a company that’s 10x larger than yours with oceans of cash is no easy thing, to greatly understate things, and the fact that they’re able to design very sophisticated cores like Bulldozer and Zen with far less people, far less money, relying on external foundries, and having to design state-of-the-art graphics on the side which Intel themselves couldn’t get a grip on, speaks volumes about the quality of AMD’s engineers, never mind their marketers. Their past management teams have had some bunglers but they’ve largely played to the best of their abilities what hand they’ve been dealt. In an oligopoly where you’re David against Goliath and there’s no room for the slightest error, AMD was bound to lose their balance.

      2. And how, pray tell, would you know that?

      3. These are rumors. I don’t believe them one bit. Take these rumors with a big handful of salt and drink sea water if you get thirsty.

        • RAGEPRO
        • 3 years ago

        He said “one of”.

    • chuckula
    • 3 years ago

    About that 17th of January business… while I’m sure AMD could make a big announcement on that date and that X370 motherboards could be available on that date, I’m not so sure about Zen itself being widely available on that date.

    That’s simply because for all the “leaks” we’ve seen about Zen, nobody claims to have seen a production-ready Zen part floating around. It takes time to go from the official commencement of commercial-scale production to actually having products on the market.

    Case in point 1: There have been plenty of leaks of 100% production-ready Kaby Lake parts on the interwebs even though you also aren’t buying a desktop Kaby Lake until January.

    Case in point 2: Remembering back to Llano, we had this article about how Llano was “shipping” in April of 2011: [url<]https://techreport.com/news/20716/amd-kicks-off-llano-shipments[/url<] [Note: that photo in the story is [b<]not[/b<] from the fab where the lithography takes place, it's from the packaging plant that receives the finished silicon, packages it, and performs the final testing before the products go out to OEMs or the consumer channel.] Then there's this article, posted by TR on the earliest Llano laptop (the desktop review came a week later): [url<]https://techreport.com/review/21099/amd-a8-3500m-fusion-apu[/url<] That's over 2 months of delay from "shipping" parts to actual products being available.

    • kuttan
    • 3 years ago

    …..

      • tipoo
      • 3 years ago

      It’s what a lot of us have wanted for years rather than spending money on wasted die space if we use a dGPU all the time, but we got a lot of “no, it wouldn’t work like that because ????”

    • ronch
    • 3 years ago

    A lot of folks, sorry, MOST folks don’t expect Zen IPC to be as strong as Skylake, but if you look at the [url=http://cdn5.thinkcomputers.org/wp-content/uploads/2016/08/amd-zen-hotchips-4.jpg<]architecture[/url<] you'll see that Zen is a very wide machine, apparently even wider than Skylake. It merely matching Broadwell is probably due to it being the first iteration of a completely new architecture. Yes, a wide architecture on paper doesn't mean high IPC (do Apple's wide mobile CPU cores match Intel's wide big cores?), but it does look like a very ambitious design. AMD seems to be pulling all the stops here. The only thing I'm worried about is the process node it's built on. Those non-Intel process node monikers have become very misleading so I think GF's 14nm is more like 18nm relative to Intel's 14nm. And with those nodes tuned for low power mobile, one has to wonder why AMD decided to give you 8 cores at relatively low clocks and a relatively low TDP first, with seemingly no mention of 6700K equivalents announced yet. Don't get me wrong, 8C/16T at those purported clocks is all good but I just hope Zen can clock beyond 4GHz as well at <100w.

      • synthtel2
      • 3 years ago

      6-wide dispatch and 4 integer ALU ports are in line with Skylake. Skylake has a bit more AGU, and FPUs are obviously tough to compare. In Zen’s favor, it breaks int/float functionality into separate ports better, so SMT with an int-heavy thread and an float-heavy thread might work better. In Skylake’s favor, it’s got twice the L1D bandwidth of Zen. That was doubled with Haswell to keep AVX2 better fed, and Zen has IVB’s numbers there.

      I like a lot of Zen architecture decisions, but it doesn’t have a solid win on width.

        • ronch
        • 3 years ago

        Some of Zen’s numbers seem to be less than Intel’s (in-flight instructions, out of order windows, those things) but I reckon AMD had to do an extremely sophisticated core design here at the shortest amount of time possible with a tiny budget. That’s frickin amazing. Heck, I reckon Bulldozer took 6 years, had an easier target to hit, and they still had far more cash back then. Zen took just 4 years, roughly. Wow.

      • tipoo
      • 3 years ago

      “Yes, a wide architecture on paper doesn’t mean high IPC (do Apple’s wide mobile CPU cores match Intel’s wide big cores?)”

      They’re limited in clocking near as high, but actually, in instructions per clock per core they do get shockingly close. Core M vs A10 usually still sees Core M pulling more than half of the low level test wins, but it also has a much much higher turbo clock and sometimes base clock.

      Better example for your point would be the even wider Nvidia Denver.

      • the
      • 3 years ago

      That slide is interesting but it does omitted a few things and shows a few weaknesses. The idea of individual integer execution unit schedulers is unique. I’m really curious how this will work out. The integer rename block in that diagram likely employed some dispatching logic and queue before each instruction stream is scheduled. I’m also curious if this arrangement has some weird quirks like the micro-ops using upper 8 registers are always fed to a particular ALU while work with the lower 8 registers goes to another. The integer rename block could hide some of these quirks so uncovering them in the real world maybe difficult if they exist (i.e. hand coded assembler specifically written to uncover these). The brand and decode units likely inject some metadata into decoded micro-ops for the instruction schedulers. Otherwise incorporating six full schedulers would consumer too much die space. The floating point unit is still 128 bit wide and contains two units. I think it is safe to say that Intel will still dominate in floating point heavy workloads, especially compared to SkyLake-E with its AVX-512 support.

      I’m not optimistic about extremely high clock speeds on Zen, or really anywhere outside of IBM. The power consumption trade off is rather dire to go beyond 4 Ghz on modern processes, even with architectures designs to scale that high. At some point, it may become wiser to create a dedicated core design around high single threaded performance/high clock speed target for the desktop/workstation/server market without concern for the energy bill and sit it in a 200W rated socket. Due to the expense of designing a new core, Intel and AMD have been recycling a mobile focused core for server usage and used the energy savings to quickly scale up core count. Not a bad strategy for the past several years but when you can purchase 24 core/48 thread chips, Amdahl’s Law is readily apparent.

      AMD’s choice for 8 core first is likely a strike upon Intel’s continued pace of four core designs in the mainstream. Even with an IPC edge, it is unlikely that Intel’s quad cores will be able to clearly out run AMD’s Zen eight core designs in parallel workloads. AMD needs a win and this is a simply way to get there.

      As for the rhetorical question of Apple’s wide cores matching Intel’s, Apple is indeed nipping at Intel’s heels. Apple is floating around Sandy Bridge level of IPC in the 2.3 Ghz clock speed range with less than 5 watts of power consumption for the SoC. That is incredibly impressive considering that Sandy Bridge is a six year old part and consumed 95 W.

        • ronch
        • 3 years ago

        The idea of separate schedulers for Integer and floating point has been an AMD design constant since K7 (not sure about earlier designs but I don’t recall AMD ever doing central schedulers). The schedulers for each individual ALU are probably a bit similar in concept to the K7/K8/K10, where each ALU/AGU pair had its own reservation station and scheduler. The FPU, at least on paper, is a bit curious. AMD has historically doubled their FPU width from one generation to the next (K8 64-bit, K10 2 x 64-bit, BD 2 x 128-bit). Given how Intel chips crunch 512-bit AVX 2.0 chunks already, it seems AMD again decides to wait before fully embracing the latest FP extensions.

          • the
          • 3 years ago

          SSE and AVX permit the actual execution of wide instructions to be cracked in smaller units for processing. This is actually how AMD implemented 256 bit wide AVX instructions on the dual 128 bit hardware inside of Bulldozer. Intel being Intel nerfed the FPU on the Atom chips in a similar matter to provide only 64 bit throughput on 128 bit SSE instructions.

          The majority of consumer parts Zen will be up against will only have AVX2, 256 bit wide support so they won’t be too far behind. It will only be the big Skylake-E parts that will utterly crush Zen in floating point heavy workloads with its AVX-512 units.

            • ronch
            • 3 years ago

            Yep. K8 had a 64-bit wide FPU and it had to do SSE operations in two passes. K10 saw a doubling of its FPU to 128-bit so it can crunch SSE operations in one go. BD is a repeat of this stunt, with both 128-bit FPUs able to work together to crunch 256-bit AVX. But AFAIK starting with Haswell (or was it Sandy or Ivy?), Intel can do two 256-bit AVX operations concurrently, one add and one mul, which effectively makes Intel’s cores twice as strong in AVX compared to BD. Skylake (or was it Haswell too?) also supports AVX 2.0 512-bit ops. Not sure if this was turned off as per usual Intel fashion in consumer CPU models.

            Correct me if I’m wrong on any of the things I’ve said here. It’s been a while since I read up on these.

            • the
            • 3 years ago

            AVX-512 operations were first introduced with Knight’s Landing and then next year in SkyLake-EP.

            AVX1 and AVX2 both have a maximum width of 256 bit. The main difference is that integer vector operations are generally 128 bit on AVX1 but have expanded to 256 bit on AVX2. The other major component with AVX2 are gather instructions. Technically speaking, fused multiply-add (FMA) operations are strictly not part of AVX2 but a separate extension that was also added with Haswell (Bulldozer had FMA support with AVX1).

            Bulldozer and Haswell introduced FMA support for a combine operation that effectively doubles the FLOPs figure. Haswell also has a separate multiply unit so that it can actually perform two 256 bit multiplies simultaneously. That puts a theoretical limit of 12 double precision FLOP per clock for Haswell. Bulldozer has two 128 FMA units so it tops out at 8 double precision FLOP per clock.

            Also the way Bulldozer splits 256 bit operations has a bit of overhead. Bulldozer is actually faster executing two 128 bit operations than a single 256 bit unit by a few percent.

            • jts888
            • 3 years ago

            To what extent is AMD really limiting itself by not pursuing AVX-512, given that they want to have hybrid Zen-Vega MCMs out next year anyway?

            It takes rather branch-y code to get higher FLOPS per watt and per mm^2 on a traditional slave SIMD unit than on a less latency concerned SIMT complex.

            • the
            • 3 years ago

            ISA parity is important as newer software takes advantage of the new hardware. AVX-512 occupies the same niche as AXV1 and AVX2 so it is odd that there isn’t support for the new extensions. Even if AMD didn’t implement a full 512 bit wide hardware unit, using the existing 128 bit units should have been to difficult to implement. Granted there are a few changes like additional registers to AVX-512 that would impact things like how register renaming OoO logic works but those are relatively minor in the grand scheme of things.

            HSA with CPU + GPU is indeed the way to go for HPC long term but you don’t want to fall behind on CPU feature support.

      • Mad_Dane
      • 3 years ago

      AMD has said it will hit 4 on air and over 5 on LN2.

        • Waco
        • 3 years ago

        Which isn’t much of a bragging point…

    • LocalCitizen
    • 3 years ago

    Zen does scratch an itch of many enthusiasts here: iGPU-less CPU

    how many of 6600k and 7600k owners here don’t use a discrete GPU? if you look at the skylake die shot : [url<]https://newsroom.intel.com/news-releases/introducing-6th-generation-intel-core-intels-best-processor-ever/[/url<] if you just remove the graphics, there's room for 4 more cores. if intel brings 8 core to z170, even with memory limitations, many many here would love to get it.

      • jts888
      • 3 years ago

      Broadwell-E does exist, but Zen feels almost more like clusterable Xeon-D in a proper socket instead of BGA.

    • rudimentary_lathe
    • 3 years ago

    If AMD marketing was smart, they’d start under-promising and over-delivering. Imagine the positive press on launch day if these CPUs typically hit 4GHz on average? Reviewers would drool and they’d sell like hotcakes.

    I’m hoping this is why the rumour of 3.5GHz turbo is out there, to keep expectations on the low side.

      • ronch
      • 3 years ago

      The same can be said regarding the purported price points. Folks might expect them then get shocked when official Zen pricing comes out.

    • ronch
    • 3 years ago

    If those prices are true, it’s gonna be interesting choosing between a 6700K and a 16-thread CPU with presumably lower per-thread performance. Time to start a poll, TR. Me, maybe I’d wait for a direct 6700K competitor from AMD, perhaps if a 4-core model can reach 4GHz+ at a very compelling price, then I just might bite. *MIGHT*, because I still can’t justify replacing my 8350. I don’t really need anything faster at this point. Heck, I seldom even game and even when I do, I play old games anyway.

    • ronch
    • 3 years ago

    At those prices, it seems AMD wants to steal Trump’s thunder 3 days earlier.

    • synthtel2
    • 3 years ago

    3.3 base / 3.5 turbo doesn’t bode well for overclocking, if true. If it were just about power, I’d expect to see a bigger gap there. Small gaps like that are what I expect to see when someone’s already maxing out a process. Bad power management of various sorts could also account for that, but that’s worse.

    If it can properly get Ivy Bridge IPC, I’d still buy it.

    • Unknown-Error
    • 3 years ago

    [url<]http://www.legitreviews.com/wp-content/uploads/2016/11/amd-zen-cpu-pricing.jpg[/url<] So SR7, SR5, SR3 are all in the RMB 1500 or above category. So we can assume the cheapest Zen will be USD 199? Compared to previous gen AMD CPU, it is pricey. That is not necessarily a bad thing. Hopefully it means they are confident about the overall performance.

    • albundy
    • 3 years ago

    i’ll wait till it’s in the bargain bin by summer for 75% off.

    • klagermkii
    • 3 years ago

    This worries me a bit when combined with the claim of +40% IPC over Excavator.

    So 3.3GHz could presumably perform like a 3.3GHz + 40% = 4.6 GHz Excavator? The FX-9590 was a 4.7GHz Piledriver core, and there was some gain between Piledriver and Excavator but nothing amazing.

    I feel like a FX-9590 at lower power and improved multi-threaded performance just isn’t enough.

      • jorelldye
      • 3 years ago

      It apparently has slightly better IPC than broadwell. Overclocking headroom is the question.

      • Antimatter
      • 3 years ago

      I don’t think its fair to compare a 220W pre-overclocked CPU to a 95W base clocked CPU. It’s possible that Zen could reach similar overclocks.

      • Pholostan
      • 3 years ago

      You’re forgetting Steamroller. Excavator is the fourth revision, not the third.

      So the math should be something like:
      Zen 3.3 GHz + 40% = 4.6 GHz Excavator + 5% = 4.8 GHz Steamroller + 10% = 5.3 GHz Piledriver.
      Using conservative gains between the revisions.

      It has been rumored that the 8-core Zen parts OC to about 4-4.2 GHz. That would mean ~ 6.4-6.8 GHz Piledriver.

    • bfar
    • 3 years ago

    Sounds like they’re going after HEDT rather than Intel’s mainstream offerings. Although I expect a true 8 core part would be very tempting over similarly priced 4 core skylake or kabylake products, even with a lower clock speed. If these things overclock well, they’ll be awesome.

      • ikjadoon
      • 3 years ago

      Hmm…their pricing is perfect for Intel’s 4-core i5 and 4-core i7, though.

      Intel’s HEDT 8-core (Core i7-6900K) costs $1100.

      Why would AMD (a company struggling financially) undercut their competition by $700?

        • Klimax
        • 3 years ago

        Only brain dead corporation would do that.

    • Kelbor
    • 3 years ago

    I believe another angle to look at this pricing to performance equation is that the leaked prices, if true, may be for bulk pricing? Could it be possible AMD is giving generous pricing to distributors that put in very large orders?

    • rechicero
    • 3 years ago

    I dont’ really understand people who expect something similar to Intel. The difference in the processes is simply abysmal. If they are competitive in the middle end, with acceptable TDPs, while being profitable that would be a huge win (with the actual difference of fab tech). We cannot expect for design to overcome such a difference in fab proccess. Intel is 1-2 steps ahead and has the fabs perfectly tuned for their designs. AMD has to settle with worse nodes, worse tech and processes optimized for mobile.

      • mesyn191
      • 3 years ago

      Intel has been pushing to improve performance per watt and doing very little to improve performance since Ivy Bridge. Its not too shocking to expect AMD to get within 10% of Skylake per clock performance but with lower peak clocks and somewhat higher TDP’s.

      It’d be great if they overclocked well but I wouldn’t get too optimistic. Even if they do OC well they’ll probably need really good cooling. But for a budget chip that still gives a good showing against much more expensive chips that is all OK.

    • ColeLT1
    • 3 years ago

    I hope this can beat a 2600k in gaming, but I am not optimistic. Clock for clock it may match, but 2600k’s 3.80 GHz turbo may be enough to win over 3.5ghz. In January I will be getting a Zen or 7700k, whichever one gives the most (and more importantly most consistent) frames in today’s existing games.

    If this only matches a chip from 6 years ago, then it can’t expect to compete with a 4.5ghz chip with better ipc in the gaming world. If the chip will hit close to 5ghz with water, then it could be my next gaming PC CPU.

    • maxxcool
    • 3 years ago

    I’m getting a super weird vibe about his launch.

    *IF* accurate, pricing it at 290$ tells me they view sub 350$ Intel CPU’s as their target.. the it’s ‘good enough’ for me crowd who wants 8 threads or more and save 70$.

    That makes me think there is a issue were not aware of. Because if *I* were CASH DEPRIVED AMD and I actually had very close *PER CORE* IPC parity with intel I would sell 1 module 4c/8t cpus ALL DAY and get near double the output on my wafer investments for a smaller CPU build\DIE. That’s winning all day long. Hell id sell it for 250$ instead of a “possible” 290$ and still make great margins.

    For them to bang the drum and sell a 8c/16t CPU for such a price makes me cringe and think that YET AGAIN they have bet on ‘specialized’ multi-threading performance to make up for real world single thread performance. This smells like BD 2.0..

    we will have to wait and see …

      • Beahmont
      • 3 years ago

      Actually I think that the word ‘Module’ is the tip off here. I’m betting that those 8c/16t chips are 2 4c/8t chips in the same package much like the original Core2 Quads. The Package may have some way for them to share information, but they likely aren’t on the same ring bus.

      If they can do that very cheaply, then it is still possible for AMD to be making cheaper higher yield wafers and still selling high end chips on the cheap.

        • smilingcrow
        • 3 years ago

        I think it’s been confirmed that the 8 core CPUs are a single chip and the server versions use up to 4 of these in an MCM with 4 dies on the same package.

          • Waco
          • 3 years ago

          It has been confirmed. I spoke with an AMD rep this week and 4 modules in a package is what the new top-end Naples CPU is. 32 cores, 64 threads, 128 PCIe lanes.

            • jts888
            • 3 years ago

            Any word on the number of active memory channels on Naples?

            8 would require pretty thick motherboard PCBs to route all the traces, and the DDR controllers/transceivers might in principle be used for chip interconnects.

            • Waco
            • 3 years ago

            They wouldn’t comment officially, but when I asked about eight, they nodded.

            I’m hoping it’s true!

            • jts888
            • 3 years ago

            Nice. Any hints about the 10GbE MACs too? Have heard that there are 2 per die, so 2x40GbE per Naples socket seems plausible.

            (Also hoping for 2x10GbE on SR but not holding my breath)

            • Waco
            • 3 years ago

            They wouldn’t talk much about CPUs unfortunately. No dirt on integrated NICs/HBAs.

            • the
            • 3 years ago

            If anything, it would be the PCIe transceivers that would be utilized for inter die communication. Adding that layer complexity to the memory controllers is an easy way to add latency there, something to avoid for performance reasons.

            If AMD wanted to go the high-end route, they could incorporate memory buffers like Intel and IBM have been doing in their big iron chips. That is a more economical way of scaling memory capacity and bandwidth. However, a 512 bit wide memory bus per socket is coming to IBM’s POWER9 scale-out variant so AMD wouldn’t be alone with such a configuration.

            • Waco
            • 3 years ago

            PCIe is woefully terrible for inter-die communication. I surely hope not.

            • the
            • 3 years ago

            That is actually the basis for what IBM has done with POWER8 and its [url=https://www-304.ibm.com/webapp/set2/sas/f/capi/CAPI_POWER8.pdf<]CAPI (PDF)[/url<] and [url=https://www.nextplatform.com/2016/05/04/nvlink-takes-gpu-acceleration-next-level/<]nvLink[/url<] interfaces. The next iteration of these will be arriving next year with full cache coherency support. Interestingly enough, AMD has signed on to openCAPI so there could be a few surprised coming in future Opterons.

            • Waco
            • 3 years ago

            Hmm. Sunday reading for me. Thanks for the links!

            • chuckula
            • 3 years ago

            That should mean 32 lanes of PCIe on consumer single-die parts. Hopefully they will all be available on consumer-level platforms.

            • Waco
            • 3 years ago

            Yep, 32 lanes per module. I hope they maintain that across the line!

        • the
        • 3 years ago

        The Zen architecture is based around the idea of four cores per cluster. This is due to how the cache hierarchy has changed in terms of intra and inter socket communication. Each core in a cluster has immediate access to all the L3 cache in that cluster. Multiple clusters maybe placed on-die but they use a different cache coherency protocol. The protocols between sockets are again slightly different to account for even more latency and lower bandwidth vs. on-die.

        This cluster design is similar at a high level to what Intel did with the Xeon only Dunnington die which was composed of three Core 2 Duo clusters plus L3 cache.

      • blastdoor
      • 3 years ago

      I might be misunderstanding your point, but I’m not sure we can preclude the possibility that they will sell a higher clocked 4c/8t version at a competitive price.

      A 3.2 GHz quadcore Skylake is about $200. If AMD comes in at 80% IPC, then maybe it would be reasonable for them to sell a 4 GHz quad core Zen for about $180.

      But maybe it just depends on what you mean by “very close” IPC. My expectations for AMD are low enough that 80% seems “very close”.

        • maxxcool
        • 3 years ago

        Yeah, hence me hedging wit the closing statement of well have to wait and see in 60 days or so. there is something iffy here imo.. but I could be wrong.

      • ptsant
      • 3 years ago

      It has been known for quite a long time that Zen (at least v1.0) will not surpass Intel IPC or single-threaded performance. Then again, it doesn’t absolutely have to. Decent single-threaded performance and excellent multithreaded performance is not a bad mix if it matches your needs.

      Since the Pentium, AMD always had “strong” and “weak” points. Back when it was the K5, then K6 vs the Pentium I-II, AMD had the integer crown (=99% of applications). Yet since the Quake engine used the FPU for calculations (actually, only the FDIV instruction), the Pentium was much better in all ID tech games. Back then, everyone said that FPU was all that counts.

      Don’t fall for the marketing. Look at what you actually do before looking at benches.

      • hansmuff
      • 3 years ago

      I don’t believe we’ll see those chips in Best Buy machines or some such. It’s an enthusiast part, right, so they are looking to break perf/$ for enthusiasts who can use (and know they can use) a 16 thread CPU.

      And damn if I’m not excited to see what this thing can do in a large Visual Studio project compile.

      Games still seem to do very well with Sandy Bridge 4c8t performance, so I don’t expect problems (or really major advances) there.

      I think they want to beat the drum and generate excitement, and a 4c8t part probably won’t do it. It’d be “good”. Gamers are probably (and rightfully so) not going to be too excited about that chip.

      It’s like when Microsoft bets on Developers, Developers, Developers. It can work, it might not. It’s got a better chance, I think, than jumping into mainstream right away.

        • maxxcool
        • 3 years ago

        Fair points. We will see in a couple years and revisit the topic. πŸ™‚

      • Zizy
      • 3 years ago

      They will have 4C and 8C chips, don’t worry. And likely 6C and 2-3C ones too – lower bin parts.

      4C will be also the basis for the future APUs. I guess 4C will be i5 competitor essentially. Won’t be all that much smaller than 8C, as it will have the same memory controller and likely the same other external interfaces too. Only less cache and cores, which is a meaningful space savings, but not half of the chip.

      8C will probably be more of a workstation-class chip. Should have nice number crunching performance, but lacking memory compared to 2011 platform. But there are 2 MCM and 4 MCM for that.

    • Kretschmer
    • 3 years ago

    3.5 Ghz boost is anemic, and this rumor prices Zen as such.

    I still don’t understand why AMD continues to push MOAR COREZ when the vast majority of software does not scale to 8+ threads.

      • blastdoor
      • 3 years ago

      Maybe it’s because they can’t beat Intel in single threaded performance and so have no choice but to target the smaller market segments where more cores can be put to good use.

      Fine with me — I’m a customer in that market segment. The problem with AMD for the last decade or so, though, is that even with a bunch of cores, they haven’t been able to overcome their abysmal deficit in single thread performance.

      • swaaye
      • 3 years ago

      Because they want to sell in servers. And because they know that MOAR COARZ sounds cool to a whole lot of people out there. And probably also because winning multithreaded benchmarks is easier than trying to win single threaded tests.

      • w76
      • 3 years ago

      AMD pushed AMD64 when the vast majority of software didn’t utilize it. AMD was probably most peoples first multi-core part, before the vast majority of software could utilize it. And my friends sports car has the option of disabling traction control, which he only did once before spinning out just trying to make a U-turn, so there’s another feature not used the vast majority of the time.

      In short, it doesn’t hurt to lead the way, and it’s a decent marketing strategy. I also suspect more software can leverage it a little better if more cores were commonly available, but the vast majority of laptops out there, thanks to Intel and lack of competition, are still today dual core parts.

      There’s also the ‘prosumer’ class out there, the hundreds of thousands of streamers and Youtube wannabe stars, who really can leverage every core thrown at them and do things like cause odd price spikes in generations-old Xeon parts on eBay whenever a new guide about “how to build a billion-core Xeon monster” comes out. To them, this would sound good.

      And to me, too. Once a day to every other day, I do some database-related work that saturates my CPU, and I’m constantly making Excel pause briefly to calculate. I’m a minority of the market, but still. It’s a market.

      Anyway, I’m not a fanboy by any means. The platform will probably suck, and the part will be probably be mediocre, I’d just like to see Intel face some competition to the benefit of us all. Intel’s quite obviously been holding back for a very long time.

        • Kretschmer
        • 3 years ago

        Many (most?) software tasks are not inherently parallel to 16 threads. This is also a feature that requires clock speed, cache size, and other tradeoffs, not a “pure benefit” like the ability to disable traction control. It’d be like giving up legroom to gain a traction control toggle. Always a drawback; sometimes a benefit.

        • Klimax
        • 3 years ago

        Sorry but there is MASSIVE difference between 64bit and large number of cores. 64 bit solved system-wide exhaustion of address space and permitted some fixing of instruction set and mostly required just recompilation + fix up of code.

        Utilization of large number of cores requires task to be parallelizable. Many problems can’t be written in multi-threaded way and there is no replacement for them. And even fi you can rewrite it all to have parallel execution you will quickly become limited by serial code and threads can starve each other of resources. (Cache contention, memory,…)

        It is far more complex thing then you think. And betting AGAIN on multithreading is insanity.

        • the
        • 3 years ago

        AMD did will in making their Athlon 64 and X2 a bit future proof though. I suspect that very few socket 754 Athlon 64’s in the consumer space were upgraded to a 64 bit version of Windows. On the other hand, they were very popular with developers who would bring those 64 bit versions of applications to market.

        Similarly, the dual core Athlon X2’s still showed some benefit even if applications themselves were not multithreaded. We can thank Microsoft for their scheduler for that benefit.

        The real problem for AMD in terms of leading cutting edge technology is that it is all on the GPU side right now. HSA is a great thing on paper but AMD needs applications to take advantage of it. If a developer was looking at a massive amount of parallel CPU threads for a single task, implementing their algorithm on a GPU would make more sense. Massive amounts of CPU threads on a chip are niche for the massive amount of independent tasks.

      • smilingcrow
      • 3 years ago

      Intel’s current 8 Core desktop CPU is 3.2 Base to 3.7 Turbo @140W TDP and lists at $1089 so based purely on those figures Zen would have to have a terrible IPC to not be very good value for highly threaded workloads.

        • blastdoor
        • 3 years ago

        I just did some poking around on NewEgg and I think the best competitor from Intel might be the 6 core 3.4 GHz Broadwell-E for $379.

        Here’s some crude math assuming perfect scaling with cores.

        Intel
        3.4*6 = 20.4

        Zen
        3.2*8 = 25.6

        For Zen to match the 6 core Broadwell-E in performance, its IPC needs to be at least 80% of Broadwell.

        If that works out to be true, you can save about $90 buying Zen rather than Broadwell-E.

        And that feels like about the magnitude of cost savings for a given performance level that the bean counters at AMD would come up with. That is — it’s just enough savings to move a few people over to Zen, but not one penny more.

          • smilingcrow
          • 3 years ago

          Possibly save another $90 on the motherboard as you will be using the mainstream socket/chipset with AMD.

            • blastdoor
            • 3 years ago

            Good point!

            So based on a fuzzy rumor and some fuzzy math, I’m feeling warm and fuzzy about Zen πŸ™‚

          • the
          • 3 years ago

          The other thing is that that low end 6 core Broadwell-E part only has 28 PCIe lanes instead of the usual 40 lanes for the high end LGA 2011-3 socket. If Zen has 32 lanes off of the CPU for graphics, it may have a slight edge (a few %) by providing two full 16x PCIe slots for discrete GPUs.

          The real question will be how robust Zen’s chipset for this market is as features like 10 Gbit Ethernet, multiple PCIe based M.2 slots and USB 3.1 controllers are rolling out on high end LGA 2011-3 boards. Those carry massive premiums compared to regular LGA 2011-3 boards so AMD could make a good value move here.

      • Tumbleweed
      • 3 years ago

      What Intel 8C/16T CPU has higher clocks?

        • smilingcrow
        • 3 years ago

        Intel has an i7 8 Core desktop CPU with a 3.2 Base to 3.7 Turbo.
        They also have Xeons with higher clock speeds for 8 cores and for 10 cores but the prices get very high for those.
        So if Zen is half decent it will be a bargain by comparison.

          • lycium
          • 3 years ago

          Damn right, and I will be so happy to buy AMD again if this pans out.

          It’s been a long, long time since the Athlon XP / Athlon 64…

          • Tumbleweed
          • 3 years ago

          Yeah, that’s what I was getting at. The Intel ones start at about a grand and go up from there. If this is even vaguely in the same ballpark performance-wise, it’s a huge win for AMD.

          I expect most people to be very surprised how little a difference going from 4 cores to 8 will make for many applications, though. πŸ™‚

          Even forcing Intel to lower their prices would be a big win, though.

      • ronch
      • 3 years ago

      Maybe because they’re gunning for Intel’s expensive 8-core chips first? This is no doubt meant to make a ‘big splash’ in the market before the rest of the chips arrive.

        • Klimax
        • 3 years ago

        Wrong pricing then.

      • demani
      • 3 years ago

      For my very particular interests though, Zen could (new to saying it will) be exactly right though. Video compression and rendering gain a lot with extra cores/threads since they scale linearly. So even if the IPC is 1/2 intel’s, and the clock is 80%, I would still see a performance boost over 4c/8t intel part at the same cost. So maybe the market for this won’t necessarily be game boxes, but virtualization and rendering might be good targets. I could see getting a set of decent render nodes up and running for under $1k each.

      (I am currently assembling a dual 6-core Xeon as a graphics machine; so that’s the kind of station that makes sense here).

    • Chrispy_
    • 3 years ago

    Pricing looks good and we know that Zen is comparable to Broadwell in terms of multi-threaded performance in AMD-selected tests.

    What the rest of us want to know are the killer answers:

    What is the IPC of a single thread?
    Can it beat Sandy/Ivy/Haswell/Broadwell/Skylake in single or poorly-threaded applications?

      • blastdoor
      • 3 years ago

      Here’s my hope. I hope IPC is about Sandy. If so, then $290 for 8 Sandy cores clocked at 3 GHz strikes me as pretty awesome for people (like me) who have embarrassingly parallel CPU work to do (my work isn’t quite so embarrassingly parallel that it could use a GPU).

      But if somebody knows of an existing Intel option for $290 that gives the multithreaded performance equivalent of 8 3 GHz Sandy cores, please let me know.

        • smilingcrow
        • 3 years ago

        Only if you buy a second hand SB Xeon E5-2690 which are ~$150 for 8 cores at 2.9GHz and will Turbo to 3.3.
        You could buy two for $300 as they are DP enabled but the boards will be ancient.
        But for pure CPU grunt that’s a lot for $300.

        • revcrisis
        • 3 years ago

        Obviously it will be higher IPC than Sandy. If AMD can’t match the IPC of a CPU that came out 5 years ago then what’s the point of even making Zen. It would be a colossal failure on AMD’s part to not catch up from half a decade ago. Sandy Bridge came out in January 2011. January 2016 is 5 years later. FIVE YEARS! If AMD can’t match or exceed the IPC from an ancient architecture 5 years ago, I would be speechless.

          • smilingcrow
          • 3 years ago

          2011 – 2017 = 6 years.

          • blastdoor
          • 3 years ago

          Yeah… except that the progress in IPC over the last 5 or 6 years has been a great example of “diminishing marginal returns”. And of course it’s that very issue that has driven us to multi-core CPUs in the first place. If we were talking about 2006 compared to 2000, then speechlessness would be appropriate. Given where we are now, I’m not so sure.

          • swaaye
          • 3 years ago

          It might not hurt to be prepared for such speechlessness. πŸ™‚

        • Chrispy_
        • 3 years ago

        I’m actually optimistic about AMD for once, and IPC better than Sandy has already been proved, so you’re in luck:

        [url=https://www.youtube.com/watch?v=xzZT2xH3zBk<]8C/16T Broadwell 3GHz vs 8C/16T Zen 3GHz Engineering Sample[/url<] - Performance was practically identical. The thing is, right - IPC on a heavily parallel multicore workload is not the same thing as IPC on a single thread. For the compute and compiling people, Zen looks to be a fantastic product. For most people the performance is still dictated by a single primary thread and the way FPU, scheduler, prediction, cache and internal bandwidth works, you can't just say that Zen's random single-thread performance is going to match Broadwell, even if they match at 8C/16T workloads; CPUs succumb to different internal bottlenecks when running all cores flat out compared to how then run when they're idle enough that the individual subprocessor bottlenecks become a factor. I would hope it's at least Sandy level. That wouldn't be a bad thing, especially if people can get a 3.3GHz Sandy 8-core equivalent for 6700K money - It's a valid tradeoff and there's enough of a market for that the AMD should gain some much-needed ground on Intel.

          • RAGEPRO
          • 3 years ago

          One of the things I find interesting about that Zen performance comparison is that Blender is actually [url=https://wiki.blender.org/index.php/Doc:2.4/Manual/Render/Performance<]sharply limited by system memory bandwidth.[/url<] AMD said the machines were "identically configured", which leads me to believe that they were only using two memory modules in the Broadwell-E machine despite its quad-channel capability, since Summit Ridge only has dual-channel memory. If both machines were butting up against a memory bandwidth bottleneck then it's really not a big surprise that they performed almost identically. Still, that bodes well for AMD's memory performance on Zen, considering that historically their memory controllers have been well behind Intel's.

            • joselillo_25
            • 3 years ago

            3d renders usually are not impacted by memory speed, the CPU calculation speed is always the bottleneck when you do a render in a 3d software.

            • Waco
            • 3 years ago

            You’re trusting a page that hasn’t been updated in over a decade…why?

            “Get more RAM up to your PC’s (motherboard and operating system) limit. Presently, Blender can use up to 8GG (giga-giga) of physical memory (64-bit address space), but most PC’s can only handle 4G of RAM.
            Upgrade your CPU to a multi-core/multiprocessor
            Upgrade your OpenGL video drivers
            Get a faster memory bus
            Get faster memory, up to yor PC’s motherboard limit. 667MHz memory is 30% slower than 800MHz.
            Use or set up a render farm using all available PC’s in your house, or use a render farm such as BURP.”

            Yeah, not gonna trust that. πŸ™‚

            • RAGEPRO
            • 3 years ago

            If you prefer, [url=https://www.blender.org/manual/render/blender_render/optimizations/performance.html<]the modern version of the same page.[/url<] Right there under "Hardware Improvements" is "Get faster memory, up to your PC's motherboard limit." I've used Blender myself a fair bit. It really is sensitive to memory performance. πŸ™‚

            • Waco
            • 3 years ago

            I’ve never seen anyone claiming better performance in rendering from faster memory. Any results to point to?

            EDIT: Seems to matter almost none at all: [url<]http://media.bestofmicro.com/ext/aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9IL00vNTUyNzMwL29yaWdpbmFsL2ltYWdlMDE0LnBuZw==/r_600x450.png[/url<]

            • RAGEPRO
            • 3 years ago

            It’s just not really something people talk about. People kinda have the idea that memory is memory, it seems. I suppose for most purposes that’s true enough; having enough is more important than its performance. I do think people underestimate the effect of main memory speed though.

            • Waco
            • 3 years ago

            I know memory speed affects certain applications, but rendering hasn’t been one in general.

            I’d have to see some proof to counteract this: [url<]http://media.bestofmicro.com/ext/aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9IL00vNTUyNzMwL29yaWdpbmFsL2ltYWdlMDE0LnBuZw==/r_600x450.png[/url<]

            • RAGEPRO
            • 3 years ago

            That’s fair. Keep in mind that it depends on what you’re rendering, too. The contents of the scene can drastically change the performance demands.

            • Waco
            • 3 years ago

            I’m clinging to the hope that Zen won’t tank AMD forever. πŸ˜›

            A post-AMD world doesn’t spell well for the next few big 10k+ node machines we have to procure.

          • blastdoor
          • 3 years ago

          I’m a little hesitant to interpret a single unreplicated benchmark from AMD as proof that Zen is better than Sandy.

          But still — I’m cautiously somewhat sorta kinda optimistic too (maybe).

        • travbrad
        • 3 years ago

        [quote<]But if somebody knows of an existing Intel option for $290 that gives the multithreaded performance equivalent of 8 3 GHz Sandy cores, please let me know.[/quote<] My video encoding performance more than doubled going from a 4.6ghz 2500K to a 4.6ghz 6700K. Granted part of that difference is hyperthreading on the 6700K, but if you were comparing a 3ghz 8 core Sandy Bridge to a stock clocked (4-4.2ghz) 6700K you'd be getting about the performance you are describing for about the price you are describing, with the added bonus of much better single-threaded performance as well.

      • flip-mode
      • 3 years ago

      We “KNOW” Zen’s preformance? Son, we don’t know shat until we’re actually holding a real product in our hands and seeing independent reviews of the thing. Now, what we do KNOW is that AMD has a long history of overpromising and underdelivering.

        • Chrispy_
        • 3 years ago

        Old news:

        [url=https://www.youtube.com/watch?v=xzZT2xH3zBk<]8C/16T Broadwell 3GHz vs 8C/16T Zen 3GHz Engineering Sample[/url<] - Performance was practically identical. But see my reply to blastdoor above. 8C/16T performance is not something you can just divide linearly. Two products matched at full load can perform very differently at lower loads.

          • flip-mode
          • 3 years ago

          Not only that, but AMD ran that benchmark man. We don’t know shat for sure until we see independent benchmarks.

      • ronch
      • 3 years ago

      Those are questions that we’ve BEEN asking for so long. It’s getting tiresome speculating about those things. Just wait for release.

    • DrDominodog51
    • 3 years ago

    The interesting thing here is the TDP.

    • Tristan
    • 3 years ago

    Only 300$ for competition to Intel 8 cores at 1000$ ?
    Probably 8 core Zen will be significantly slower and more power hungry than Intel’s counterpart.
    AMD know how to price their CPU if they are really competitive – the same price or higher than Intel. But Zen will be price competitive than perf competitive.

    • chuckula
    • 3 years ago

    There’s the optimistic side that thinks an 8 core Zen at $290 will peform like an 8 core 6900K for $1000.

    Then there’s the realistic side that says an 8 cores Zen at $290 will perform pretty close to what Intel is selling for $290.

      • sweatshopking
      • 3 years ago

      It likely won’t meet either.

        • 223 Fan
        • 3 years ago

        At this point the only reason to consider an upgrade from my i3770 is for the ecosystem, i.e. newer USB ports, PCIe standards, DDR4, NVMe, TB3, etc. Under those circumstances the best CPU+Motherboard bang for the buck becomes the dominant decision maker. Whether these Zen parts can hang with Intel’s for that metric will determine whether its a success or not.

      • kmm
      • 3 years ago

      With Zen likely not at IPC parity and lower clocks but 8C/16T, it’s going to beat Intel’s 4C/8T stuff ($300 range) at some things, probably considerably, and lose at others, possibly by a considerable margin. The only thing to speculate about is the matter of degree.

        • Klimax
        • 3 years ago

        So same thing like Bulldozer. Massively more resources on AMD CPU are required to beat small Intel CPU.

        That’s no way to get out of bad place…

      • jts888
      • 3 years ago

      The big factor that you’re glossing over is that Intel hasn’t sold a GPU-less workstation chip in ~$300 price range since Westmere.

      A $300 pure CPU had clearly better provide more value for non-graphics uses than a $300 APU.

      • Unknown-Error
      • 3 years ago

      Definitely won’t reach $1000 Intel CPUs. If it did, they’d price it that way. But it might actually be competitive within its price category. Something Bulldozer couldn’t while eating up more power.

    • colinstu12
    • 3 years ago

    By 8 “cores” do they mean real cores or modules or whatever they are?

    So.. this is really a 4C/16T CPU or what?

      • srg86
      • 3 years ago

      8 Real cores with SMT, so 8C/16T.

      4 cores are grouped into a module, but this isn’t the same kind of module the bulldozer arch had. They’re more just groupings I think.

      • deruberhanyok
      • 3 years ago

      Zen is using “real cores” to use your terminology – each one is a complete core, unlike the streamroller-type cores from the previous generations.

      Each “module” is 4 “real cores” with “hyperthreading”. So a dual-module chip is 8 real cores, 16 threads.

      In theory this means the mainstream parts from AMD could have the same core/thread count as an i7 by simply using a single “module” – which, depending on performance, could really shake things up in the sub $200 segment.

      • Beahmont
      • 3 years ago

      I think it’s 2 4C/8T chips on the same package kinda like the Core2 Quads were 2 2C/2T chips on the same package. Or at least that’s the impression I’m getting from the use of the word ‘Module’ in this context. Unless there is some reason to differentiate subsections of the chip, they wouldn’t differentiate subsections of the chip. That they chose to do so at 4C/8T intervals screams separate ring buses at the very least.

        • colinstu12
        • 3 years ago

        lots of info on AMD’s module/core stuff [url<]http://www.bit-tech.net/hardware/cpus/2011/10/12/amd-fx-8150-review/2[/url<] And no, this is NOT like Intel and them having two dual core dies to be a quad core cpu.

          • RAGEPRO
          • 3 years ago

          Zen doesn’t use the Bulldozer-family “module” design. Zen is a (somewhat distant?) descendant of the cat core family that we last saw with Carrizo-L’s Puma+ cores (e.g. A8-7410.) As a result it has “real” CPU cores.

          AMD has implemented Simultaneous Multi-Threading on the new CPUs, which is the academic name for Intel’s Hyper-Threading. This is not the “clustered multithreading” of Bulldozer.

          • deruberhanyok
          • 3 years ago

          Zen cores are not like the old cores. You’re looking at a five year old article there, and any information about the “modules” they had there does not apply to Zen.

          Read up on the Zen architecture. It isn’t a further iteration of the “bulldozer/steamroller” type cores, it’s a new design.

          From everything that has been released so far, it looks like the high end part will be one die with two modules, each with four real cores and SMT. So 8 cores, 16 threads.

          Not two dies like Intel did back in the day, but they’re calling it a module, so think of it like a GPU design: you come up with the whole “shader block” or whatever they’re being called these days. Your high end GPU has like, 100 of them. Then your mainstream GPU has 60. They’re different dies, but all built on the same “module”.

        • smilingcrow
        • 3 years ago

        All the evidence suggests otherwise so unless you have new evidence I place your idea in the pure speculation tray; which actually shares its cache with the In Tray.

        • just brew it!
        • 3 years ago

        The 8-core Bulldozer/Piledriver chips consisted of 4 dual-core “modules”, all on the same die. Why would they completely change the meaning of the word “module” for Zen?

          • RAGEPRO
          • 3 years ago

          “Module” was in use before Bulldozer and Piledriver, hoss, although we’re all using it wrong here. It’s [url=https://en.wikipedia.org/wiki/Multi-chip_module<]this.[/url<] All speculation points to Zen actually being fabricated as quad-core-with-8MB-L3 "modules" which AMD then combines in various ways to create socketed chips. The Naples 16-core machine being four of these "modules".

            • smilingcrow
            • 3 years ago

            That’s not how I interpret the data and when you consider that Zen is their GPU less chip designed primarily for servers a 4 core die makes no sense. Supposedly AMD will be releasing up to 32 core server versions using an MCM with up to four 8 core dies.

            I think the 4 core module thing is a design feature that makes it easier for them to scale the design from a 4 core die to an 8 core die.
            So maybe the APUs will be one 4 core module with the GPU added?

            • the
            • 3 years ago

            The concept of MCM is based around a die as a logical unit. Bulldozer’s definition was the combination of two integer units + single FPU at the core level, a radically different definition. Not that AMD didn’t use MCM either, all the socket G34 Opterons are two dies in a single package.

            With regards to consumer Zen, it will have two quad core clusters on a single die. No MCM packaging necessary. Logically on-die they do the two separate coherency domains but those are still linked by a private on-die bus.

            The 16 core Naples chips will be two dies in a MCM.

            • RAGEPRO
            • 3 years ago

            Yeah, I think you’re right. I was reading up on some stuff last night and came to the same conclusion (that the dies are 8 cores, not 4).

        • jts888
        • 3 years ago

        Current information on intra-cluster dynamics is still unavailable, but it currently looks like each 4-core cluster shares 8 MB L3 divided into 4 slices and connected by a crossbar.

        Each cluster has a single 64+8b DDR4 controller and an external interconnect. The on-die interconnect is presumably a lot fatter and should allow the two clusters to be either pooled (higher latency and hit rates) or remain independent (low latency on most hits but more misses), like Broadwell/Haswell started doing.

      • Tristan
      • 3 years ago

      Real ‘cores’ does not matter. From programming point of view, only hardware threads matter, and thread is defined by registers, stack and ability to run single stream of instructions. HW thread basically fequires just few 100K transistors per thread. Everything other is just ‘accelerating structure’ that exist only for speed, and is irrelevant from programming point of view. So, formally Zen is 16 HW thread with 8 accelerating structures.

        • eofpi
        • 3 years ago

        Nobody’s questioning what a scheduler or compiler sees. It’s a question of how pipelines and execution units are arranged, and if AMD nerfed any of them, like with Bulldozer and descendants. These things matter for performance, especially single-threaded.

          • Waco
          • 3 years ago

          AMD hasn’t shown any indication that execution resources are shared between cores…

            • eofpi
            • 3 years ago

            Until more details come out, the word “module” makes some of us concerned.

            • RAGEPRO
            • 3 years ago

            [url<]https://techreport.com/forums/viewtopic.php?t=118444[/url<]

            • Waco
            • 3 years ago

            Right.

            “Unlike on “Bulldozer,” a “Zen” core does not share any of its number-crunching machinery with neighboring cores.”

    • nico1982
    • 3 years ago

    [quote<]These amounts seem very competitive for what is supposedly a top-end CPU when compared to the prices of Intel's current range of desktop CPUs.[/quote<] To be more blunt: $290 for a 8C/16T means either that the leak is pure garbage or that Zen plain sucks.

      • sweatshopking
      • 3 years ago

      I think the 3.5ghz boost speed speaks volumes. That’s a low clocked boost for a high end chip, being like a full ghz slower than intel. Unless they’ve passed intel in ipc (they haven’t) this thing is going to significantly underperform.

        • gamerk2
        • 3 years ago

        Well, we know the process that AMD chose is optimized for <3Gz clocks. so myself and others have speculated that power draw will increase exponentially as you get above 3 GHz or so, so this isn’t a surprise.

        I’m still predicting Ivy/Haswell IPC, so in single threaded workloads, AMD will likely be 20-30% slower then Intel [though this won’t show in games; AMD will catch up there at least], but likely pull ahead in Software that scales.

        So “better” then BD at launch, in that Zen should be enough to avoid CPU bottlenecks in most games, and will benchmark well [benchmarks tend to scale across CPU cores], but for “normal” workloads, Zen will likely be equivalent to Haswell.

          • mesyn191
          • 3 years ago

          Ivy/Haswell IPC wouldn’t be 20-30% slower than Skylake or Kabylake.

          More like 10-5% slower per clock. Kabylake performs about the same as Skylake and Skylake wasn’t all that much faster than Haswell.

          [url<]https://www.pugetsystems.com/labs/articles/Haswell-vs-Skylake-S-i7-4790K-vs-i7-6700K-641/#Conclusion[/url<] The lack of improvements to peak performance since Sandybridge is why many are still hanging onto systems using that CPU. Intel hasn't been giving anyone much reason to upgrade for a long time. At least not at typical Intel prices, which aren't cheap.

            • sweatshopking
            • 3 years ago

            I think he was suggesting IPC and clock drops to get the 30%

            • mesyn191
            • 3 years ago

            He specified IPC by itself though when talking about performance.

            “I’m still predicting Ivy/Haswell IPC, so in single threaded workloads, AMD will likely be 20-30% slower then Intel [though this won’t show in games; AMD will catch up there at least], but likely pull ahead in Software that scales.”

            A 20-30% IPC performance deficit + ~1Ghz clockspeed deficit would mean Zen would be more like 40-50% slower over all.

        • smilingcrow
        • 3 years ago

        Intel’s current 8 Core desktop CPU is 3.2 Base to 3.7 Turbo @140W TDP.
        The Quad cores do boost a lot higher than that but then they do have half the cores. πŸ˜‰

        • cygnus1
        • 3 years ago

        With that many cores in a 95W power envelope, that clockspeed is very competitive with Intel. We’ll have to wait and see at what speed they can run a single module (quad core) product. That’s the one worth comparing to Intel.

      • just brew it!
      • 3 years ago

      Yup, that was my initial reaction to the (rumored) pricing as well.

      • Liquidiam
      • 3 years ago

      This price is disconcerting as stock owner. One as you pointed out, they are not competitive products. Two, if they are competitive products, they need to price them better so as to increase margins and increase profitability so that they have the necessary R&D cash for 7nm Zen plus etc.

      Let’s hope the actual pricing is higher from a product performance and company cash flow perspective.

      • chrcoluk
      • 3 years ago

      intel dont have any official 4.5ghz turbo mode I am aware off. πŸ™‚

      Shipped turbo clocks and achievable overclocks are two different things.

        • hansmuff
        • 3 years ago

        They do not, but the Kaby 7700K has been reported in a number of places to be 4.2GHz base/4.5GHz turbo.

Pin It on Pinterest

Share This