Zen 2 Review Summary

Discussion in 'AMD Processors' started by DuronBurgerMan, Jul 7, 2019.

  1. Derangel

    Derangel [H]ard as it Gets

    Messages:
    17,701
    Joined:
    Jan 31, 2008
    There is going to be very little to no difference between 3200 and 3600. Heck even 3000 might not provide a real tangible performance loss in the vast majority of applications.
     
  2. NKD

    NKD [H]ardness Supreme

    Messages:
    7,704
    Joined:
    Aug 26, 2007
    lol. You can get by with x470, you can get by with 3200mhz ram. Second sucks to be in canada I guess, not everyone lives there. I don't mean to be a jackass but people aren't lying atleast the ones who don't live in canada.

    + how about those times where intel ass raped people forcing a new socket each and every time. Atleast AMD gives you the option to let it work in older gen boars without losing any features.
     
    Last edited: Jul 8, 2019
  3. 1_rick

    1_rick Gawd

    Messages:
    520
    Joined:
    Feb 7, 2017
    Then Newegg (or someone) is ripping you off, man. Micro Center is selling the 3000s at list price (if they have them in stock) and is selling x570 motherboards for prices ranging from $155 to $700 (although the lowest-priced board listed as in-stock on their website is $299.)

    $550 for an x570 and a 3600x, orderable right now (if you were in the US).
     
  4. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    Zen/Zen+ was more sensitive than Skylake/CFL with regard to RAM speeds, yes. It is unclear if Zen 2 is likewise. We have not seen enough testing to conclude this one way or the other, although I've heard speculation that Zen 2 is less sensitive to RAM speed than the original Zen was. Even with original Zen, though, this typically affected latency-sensitive tasks more than throughput tasks. It came down to gaming more than anything.
     
  5. Bladestorm

    Bladestorm [H]ard|Gawd

    Messages:
    1,232
    Joined:
    Dec 10, 2006
    After a 2nd attempt I was able to get them to cancel.
     
    GoldenTiger likes this.
  6. Opus131

    Opus131 n00b

    Messages:
    47
    Joined:
    Jul 29, 2016
    Is the cooler of the 3600 noisy? That would worry me more than the temps, which i suppose would be sufficient for stock.
     
  7. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    This is why I like to read/watch all the reviews from non-shitty reviewers, then kind of aggregate them in my head for a more general picture. For instance, average all 8 of those results out. 9900k = ~7% faster than the 3900X in SoTR at 1080p.

    This falls in line with my general impression of a ~5% gaming gap at low resolution.
     
  8. cyklondx

    cyklondx Limp Gawd

    Messages:
    256
    Joined:
    Mar 19, 2018
    Sarcasm aside,

    It does give quite the uplift if you run KVM's
    (not much for a gamer, as there's no point - you can wait those extra 5-10sec for game to load.)
    In most cases i'd say 60-80% of VM performance is in storage.

    Mainly the typical config with Zen1 is 1x GPU, and 1x pci-e 2x nvme expansion card, + your local mobo m.2 slots filled in and used as your system. The zen1 x370/x470 gives you upper limit around 10GB/s if you play around with it in zfs if you map the local m.2 slots with your expansion card/s.
    But its not in any speed comparing to even lower end memory modules, not even ddr3. You'll find yourself making datatiered storage anyway, with memory used as cache whatever you can spare from 64GB max that zen1 limits.
    In zen2 we have option to not use memory at all, as 30GB/s is comparable to slower memory modules. Thus I could save far more on assigning it as a memory itself to my kvm's rather to boost performance on my storage for the vm's.

    Thats how i see it,
    I had 2x e5-2670 build with 768GB DDR3 1600, and booting KVM's from it was a dream. It was fast, but also contained additional problems saving, and initial image ramdisk extraction. It would boot win7 within a sec, and games that normally load a bit - loaded almost instantaneously.
    That build was too expensive and i had to sell it. But my dream of running systems from memory stayed. Today it gotten to a point where I don't have to buy memory for couple k's, i can spend little over $800 to get 4-8TB of persistent m.2 storage that may be around the speed of memory.
     
    dgz likes this.
  9. Dan_D

    Dan_D [H]ard as it Gets

    Messages:
    54,528
    Joined:
    Feb 9, 2002
    Yeah, there are a number of reasons why there are variances between the different sites. The short answer is that we didn't all do everything the same way.

    As to your specific question that's very hard to say. I'd really have to look at each review and their methods to try and answer that. Naturally, everyone is going to do their reviews a bit differently. We were all crunched for time and had to make fast decisions about how to proceed with the tests. The fact is, you have a limited amount of time to test this stuff and decisions have to be made. Anandtech is getting flak for the RAM speeds they used. Evidently, they were basically low sort of "stock" spec'ed RAM speeds. Your sort of damned if you do and damned if you don't. If you keep the speeds the same, then people argue about which processor needs what RAM to shine. If you cut loose and let each system have the best RAM it can, then people cry that it isn't fair because one platform can outclock the other. If you went over 3600MHz on the 3rd generation Ryzens, then you'll have people point out that your mclk, uclk, and fclk ratios went from 1:1:1 to 2:1 impacting performance whether it does or not. On the Intel side, I can use whatever the hell I want and not have to deal with that.

    I used identical RAM timings and memory sizes on both. The systems were indeed stock. Everything was set to automatic. I didn't use ASUS' overclocking features or MSI's on their respective platforms. Here's the thing though. I kept the memory as "apples to apples" as I could, but the fact of the matter is I've got RAM that can do DDR4 3600MHz at those same timings for the Ryzen 9 3900X. I've got RAM for the Intel side that can do at least DDR4 4000MHz. The timings are garbage, but that RAM can clock. Intel's CPUs are generally not that dependent on timings but like raw clock speed. AMD's often been different, but Zen 2 is a bit of a different animal. At DDR4 3600MHz it might close the gap vs. an Intel at DDR4 3200MHz. At the same time, I might gain more on the Intel side at DDR4 4000MHz. I'd have clocked both at DDR4 3600MHz, but I could only take the Ryzen 7 2700X to DDR4 3200MHz, and as a result I wanted to showcase just how far Zen 2 had come vs. Zen+.

    I chose DDR4 3200MHz to keep it "apples to apples" and to make the comparisons between Threadripper and the Ryzen 7 2700X more meaningful. Beyond that, I think most people buy RAM at DDR4 3200MHz speeds and given prices and what I see available at stores and online, I think that's probably the case. Lastly, I've never seen RAM speeds impact gaming all that much beyond DDR4 3200MHz. However, that was on previous architectures from Skylake onward. With Ryzen, using 3600MHz memory might close that gap. AMD certainly wanted us to do that. AMD's own data showed a smaller gap than I found, but they installed most of the security mitigations for Intel and hobbled their i9 9900K as much as they could.

    I had actually intended to do some gaming tests at higher RAM speeds and include them, but I literally finished the testing on the last day as I wrote the article, and totally wrapped up the review four hours before the embargo lifted. For these launches, you never get the time you want to do things 100% like you wanted to. I think the spread you'll see comes from variances in graphics cards used as well as potential game settings. Did they manually set everything to low or just use a low preset? In some games like the Division 2, the low preset isn't as low as it goes. I don't know exactly what Gamer's Nexus or anyone else did that I didn't or vice versa.

    You could even have some differences regarding motherboard setup. On the Intel side, did they adjust the cache ratios? I didn't. But, on the Maximus XI APEX, the fastest road to 5.0GHz is to load De8auer's 24/7 5.0GHz overclocking profile. If you do that, it sets up all kinds of values you don't need. So doing that could impact your testing. Generally the motherboard doesn't make much of a difference, but adjusting your cache ratios vs. automatic might skew the results. That's in the profile but chances are if you manually overclocked the system yourself, you didn't do that. Lastly, variances you see could be regarding boost clocks. Boost clocks from either camp aren't guaranteed. Some of the other sites might have gotten luckier or worse with their boost clocks on either side. Also keep in mind that the GPU's used in these reviews were different. Some sites used GTX 1080 Ti's. I used a factory overclocked GIGABYTE RTX 2080 Ti Aorus Xtreme 11G. Even though 1920x1080 is still very much CPU bound, you will get different results with the different GPU. I started my testing with an old Titan X and switched it out once I realized that I had done some testing with the RTX 2080 Ti and that my results were skewed because of it. Some of the applications like After Effects get GPU acceleration too, so that comes into play.

    I think the idea of looking at a lot of reviews and seeing what the sort of aggregate opinion is makes a lot of sense. If everyone does things a bit differently and reaches the same conclusion, you have your answer. I think you can't really argue that Intel is faster for gaming. It just is. Not in every game, not all the time but overall, Intel leads on that front. The question you have to ask is whether or not that lead justifies sticking with Intel and losing out on performance in other applications. You have to decide whether or not the difference is worth it at the resolutions you use. For 1080P gamers, Intel might absolutely be the way to go. At that resolution, AMD lags far more than it does at higher resolutions which are more GPU bound.
     
  10. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    Sometimes I think we all rehash this conversation with every major CPU release since original Zen.

    "AMD is faster at X."

    "Intel is faster at Y."

    "Well I think Y is more important than X!"

    "Well I think [some demographic] thinks X is more important than Y!"


    "My use case favors AMD/Intel."

    Gone are the days where one company leads in all the things. You don't get an easy "only buy [brand]" decision on the CPU front. I laugh at anyone who thinks this is a bad development. And TBH, I don't think it was ever fully that way even in the old days. You have to carefully weigh your needs, wants, and budget to determine a winner. IMHO - and it's just that, an opinion - Zen 2 fixes a lot of the weaknesses of Zen. Where it doesn't win, it's much more competitive. Where Zen/Zen+ won, Zen 2 extends the lead. Overall I think it's better positioned than Zen was at its launch. But even so, not all decision roads will lead to Zen 2. More of them, perhaps, but not all.

    Example: multithreaded performance is very important to me because of my work, and with respect to gaming, I did not buy a 1080 Ti to game in 720p or 1080p. 4k is my preference. I don't have a high refresh rate monitor. I went 42" 4k monitor, but with a traditional 60hz refresh rate. I would get no value from a 9900k's strengths, and a boatload of value from the 3900X's strengths. My decision is clear!
     
    Azrak, tom_ozahoski, mikeo and 9 others like this.
  11. VIC-20

    VIC-20 Gawd

    Messages:
    901
    Joined:
    Mar 24, 2006
    If we exclude older components for fairness, a brand new equivalent by manufacturer specification PC is going to be about the same price on both platforms.

    3600x/x570 MSI A Pro/3200
    9600k/Z390 MSI A Pro/2666

    Intel is actually a bit cheaper here, but whatever.

    3900x/x570 Aurus Pro Wifi/3200
    9900k/Z390 Aurus Pro Wifi/2666

    About the same price.

    Making adjustments by buying older components, upgrades, B and H series boards, to reduce costs, history is another conversation. Very worthwhile talking about, but misses the point.

    We have reached parity. Finally.
     
  12. KazeoHin

    KazeoHin [H]ardness Supreme

    Messages:
    7,843
    Joined:
    Sep 7, 2011

    In my mind, the arguments are that AMD is 50-100% faster per dollar in literally everything except gaming, where it is 5% behind on average FPS and margin of error +/- 5% on minimum FPS, and AMD has better socket longevity, better (by nature of existing) bundled cooling, better chipsets and memory and CPU overclocking freedom (remember XMP is only allowed on Z series on Intel).

    But people like to boil it down to "AMD SUCKS AT GAMING".
     
    FlawleZ and n=1 like this.
  13. Brackle

    Brackle Old Timer

    Messages:
    7,226
    Joined:
    Jun 19, 2003
    The thing is. You dont need X570 to run a 3600/3600x/3700x/3800x.....You can use a B450 motherboard, which is MUCH cheaper then Intel.

    If you need PCI-E 4.0 then yes get a X570.....but you are paying for the latest and greatest....you ARE paying for something Intel doesn't even offer? Why not charge a premium for it? Intel would charge a premium for it
     
    Last edited: Jul 8, 2019
  14. cybereality

    cybereality [H]ardness Supreme

    Messages:
    4,602
    Joined:
    Mar 22, 2008
    I have a B350 mobo I purchased over 2 years ago and it's still supported. Just gonna drop the 3600X in there (replacing a 1600X), easy upgrade.
     
  15. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    Yeah, but there folks who want every last frame, and don't do hardly anything else on their machines - 9700k or 9900k all the way, if they have the cash. And then there is e-peen value to consider. Never underestimate that, either.
     
  16. NKD

    NKD [H]ardness Supreme

    Messages:
    7,704
    Joined:
    Aug 26, 2007
    [​IMG]

    4-4-OC3900x.jpg

    4.4 ghz all core OC 3900x 90 mins in to the stress test. 1.375v. This is what I started with, if its fully stable for a 4-5 hours I will then reduce the voltage to 1.35 to see and lower it until it crashes.

    Also all core voltage stock was around 1.35v according to ryzen master. So not too bad.

    Here is a link to the high res picture not sure why the tumbnail is reduced if you need to see the settings.

    https://photos.app.goo.gl/8AmBoYbiEh2uH78eA
     
  17. Imhotep

    Imhotep Gawd

    Messages:
    777
    Joined:
    Feb 12, 2014
    I wonder if anyone still wants to argue that the 9900K is superior...:D
    Well done !
     
  18. Gideon

    Gideon 2[H]4U

    Messages:
    2,285
    Joined:
    Apr 13, 2006
    Very nice, I will be curious to see how my x370 board does with the 3900x when I get a chance to get one. I noticed no reviewer decided to give the x370 series a try with the new chips, which is kind of a shame.
     
  19. Imhotep

    Imhotep Gawd

    Messages:
    777
    Joined:
    Feb 12, 2014
    I skimmed through some x470 reviews. It will perform exactly the same if your mobo supports it.
     
    defaultluser likes this.
  20. NKD

    NKD [H]ardness Supreme

    Messages:
    7,704
    Joined:
    Aug 26, 2007
    2 hours 21 mins in now. lol. if I can pull 4.5ghz out of this bitch, I am golden. But I was only expecting 4.2 all core lol, so still excited at 4.4. I am wondering how long I should stress it before call it stable at this. 4, 5 hours?
     
    Nightfire and DuronBurgerMan like this.
  21. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    Let me know. I'm in the same boat as you.
     
    Gideon likes this.
  22. n=1

    n=1 2[H]4U

    Messages:
    2,388
    Joined:
    Sep 2, 2014
    Not really: https://www.techpowerup.com/review/amd-zen-2-memory-performance-scaling-benchmark/

    Going from 2400 to 3600, 13/15 applications TPU tested see 1-7% improvement, the only two exceptions are:

    Machine Learning/AI: 14%
    File Compression: 10%

    Likewise for gaming, the only two games that showed >5% improvement at 1080p are:

    Far Cry 5: 12.5%
    Sekiro: 11%

    I mean yes sure if your workflow hits the above a majority of the time, then yes you should invest in better ram. Otherwise even a basic 2400 kit will afford 95% of the performance 95% of the time.
     
  23. Gideon

    Gideon 2[H]4U

    Messages:
    2,285
    Joined:
    Apr 13, 2006
    I typically find at 4 hours stable then it's should be good to go. 4.5 was what I was hoping for as well but either way it's going to be faster then my 1700x by quite a bit.
     
  24. Brackle

    Brackle Old Timer

    Messages:
    7,226
    Joined:
    Jun 19, 2003
    I plan to wait for a better stable bios. Specially for my CH7, Either way I do have my eye on a 3900x. I think paired with my cl15 3600mhz is would perform pretty well!
     
  25. n=1

    n=1 2[H]4U

    Messages:
    2,388
    Joined:
    Sep 2, 2014
    Pretty much. And don't forget to add the caveat that you'l only see those double-digit percentage gulfs if you use a $1200+ GPU. If you have a 1080 tier GPU the results look like this:

    ryzen_9_3900x_sotr_1080p_gtx_1080_ultra-100801136-large.jpg

    I emphasize this because 2080 Ti buyers are the "best performance regardless of cost" crowd, so they're likely rolling with binned CPUs and 4000+ ram kits. I'd argue they're not even the target demographic for this release.

    3700X is clearly aimed at the "mainstream" segment, and I'm gonna guess most buyers in that segment have a 2070 equivalent at best. Since 2070 is only about 10% faster than a 1080, they'll see 5% differences at best at 1080p, and shouldn't need to concern themselves with a "CPU bottleneck".

    3900X vs 9900K is where it gets trickier. If you're like me and game exclusively at 1440p, then the 3900X is a better buy. If you're strictly a 1080p gamer and literally don't do anything else besides gaming, a 9900K is probably the better choice.
     
    FlawleZ, mikeo and Chimpee like this.
  26. n=1

    n=1 2[H]4U

    Messages:
    2,388
    Joined:
    Sep 2, 2014
    Even just 2 years ago I would've agreed with Joker that maxing out a CPU from the factory is stupid and basically a middle finger to the enthusiasts.

    These days though I'm much busier and have less time to tinker with hardware, so I actually don't mind buying (almost) maxed out stuff. At least this way I know I've tapped 95% of its performance potential, and don't have to spend 10+ hours trying to dial in the best OC or whatever. Maybe I should hand in my [H] badge...
     
    motqalden and GoldenTiger like this.
  27. Keljian

    Keljian Gawd

    Messages:
    646
    Joined:
    Nov 7, 2006
    Completely depends on what you're doing. The 9900k is superior when it comes to software and hardware compatibility. This is a fact.

    In raw processing power, the 3900x is superior due to processor cores.

    Versus the 3800x/3700x the 9900k is superior if you overclock it, as often it will reach 5ghz, where the competition will hit 4.5ghz (ish)

    If you're doing video editing, it depends on your software/hardware. Some relies heavily on memory bandwidth, and when that is the case the 9900k is superior.
     
  28. RamonGTP

    RamonGTP [H]ardness Supreme

    Messages:
    7,740
    Joined:
    Nov 9, 2005
    Zen 2 memory controller is rated for 3200 memory, not 3600. If you're going to go by "intel spec" when talking about Intel then talk about "AMD spec" when talking about AMD. Faster ram is always recommended if it can be used.
     
  29. RamonGTP

    RamonGTP [H]ardness Supreme

    Messages:
    7,740
    Joined:
    Nov 9, 2005
    I get that it's available on a wider variety of SKUs than intel but "overclocking freedom" doesn't mean a whole lot here. AMD is far pickier then Intel with RAM. In terms of CPU overclocking Ryzen 3000 appears to be pretty damn close to it's limit in stock form. I'm not hating on AMD, I just spend over 2 grand on a 3900x and company, but still feel it's important to point out that OC is not at all an AMD advantage.
     
  30. Derangel

    Derangel [H]ard as it Gets

    Messages:
    17,701
    Joined:
    Jan 31, 2008
    Ryzen 1st gen was incredibly picky on RAM, the memory controller in Zen+ was much less picky and Zen 2 is probably even less picky than that. The only picky part with AMD and memory right now is speeds and over 3200 (which you can still get pretty cheap these days) I wonder if it even makes enough difference to matter in like 95% of applications.
     
  31. Verado

    Verado Limp Gawd

    Messages:
    188
    Joined:
    May 16, 2017
    So summarizing this launch as short as possible.
    3000 series zen is fast enough for everyone. Period.
    Make really really really sure you have the newest microcode and drivers.
    570 chipset is for special use cases only. 470 is optimal value.
    Most 3200mhz will give you great performance.
     
    FlawleZ, Brackle and Keljian like this.
  32. n=1

    n=1 2[H]4U

    Messages:
    2,388
    Joined:
    Sep 2, 2014
    FTFY :D

    Did I get shadow banned?!

     
  33. HockeyJon

    HockeyJon [H]ard|Gawd

    Messages:
    1,088
    Joined:
    Dec 14, 2014
    Everyone rips us off in Canada. In Ontario, adjusting for the exchange rate, I’m seeing the 3700x for $20-$30 more than MSRP right now. I can buy it in Michigan, tax and exchange rate included, for almost the same as the list price here in Ontario, not including our 13% sales tax. Getting ripped off is more Canadian than a beaver swimming in maple syrup.
     
  34. Randall Stephens

    Randall Stephens Limp Gawd

    Messages:
    462
    Joined:
    Mar 3, 2017
    But you do have that free health care. In Seattle we have 10% sales tax and have to pay for prostate exams in cash.
     
    Last edited: Jul 9, 2019
    defaultluser and HockeyJon like this.
  35. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
    Some updates:

    Gamers Nexus claims a roughly 2% uplift in gaming performance after fixing the BIOS issue, but he also tells us that his motherboard and BIOS was more or less okay, and that some other reviewers will see bigger changes.

    Browsing Anandtech this morning, their updates show a much greater uplift after their BIOS update. Zen 2 is now winning several (but NOT a majority) benchmarks outright against the 9900k and 9700k. It beats the 8700k enough that I'd call that a tie, or a slight win for Zen 2. 7700k is crushed.

    Pretty confident on a ~5% average gap against the 9900k. The performance uplift over the 2700X is huge.

    And it looks like the single core boost clock issue really was a BIOS issue. But not all reviewers were equally affected.
     
    FlawleZ and Legendary Gamer like this.
  36. n=1

    n=1 2[H]4U

    Messages:
    2,388
    Joined:
    Sep 2, 2014
    Pics or gtfo
     
    HockeyJon likes this.
  37. HockeyJon

    HockeyJon [H]ard|Gawd

    Messages:
    1,088
    Joined:
    Dec 14, 2014
    There are ways to get prostate exams for free...

    .....

    ....wait, what?
     
  38. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,309
    Joined:
    Mar 13, 2017
  39. juanrga

    juanrga Pro-Intel / Anti-AMD Just FYI

    Messages:
    2,551
    Joined:
    Feb 22, 2017
    Some of those reviews aren't stock vs stock because the Ryzen chip is overclocked by using 3600MHz RAM (OC RAM = OC interconnect). Guru3D also used an engineering sample for Intel.

    There was also an alleged mistake at launch and "In practical terms, all reviews which were done on ASUS Crosshair VIII Formula or Hero motherboards using other than 0066 bios build must be considered invalid, at least partially. Reviews using other ASUS motherboard models (not provided by AMD) are under suspicion as well." The issue is that some BIOS allowed the 3900X to run outside spec by making believe the CPU it is consuming less power than it really is. No idea if the above game was affected "Purely single threaded workloads are fine, as well as at least most of the pure gaming tests.
    However, every multithreaded CPU workload / benchmark must be considered invalid, if ASUS Crosshair VIII Hero with any other than 0066 bios version was used as the platform.
    "
     
  40. sover

    sover Limp Gawd

    Messages:
    129
    Joined:
    Jul 19, 2016
    Reminds me of that intel thing, where most of the aftermarket motherboards by default don't constrain the 9900K to its TDP spec, but every reviewer seems to bench that way, anyhow.
     
    FlawleZ likes this.