Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

AMD Zen 3 is coming: +19% IPC, +200 MHz, +$50 MSRP, Nov. 5

QuizzicalQuizzical Member LegendaryPosts: 23,898
The key points are in the thread title.  The new lineup will be:

Ryzen 9 5950X:  16 cores, 4.9 GHz, $800, 105 W
Ryzen 9 5900X:  12 cores, 4.8 GHz, $550, 105 W
Ryzen 7 5800X:  8 cores, 4.7 GHz, $450, 105 W
Ryzen 5 5600X:  6 cores, 4.6 GHz, $300, 65 W

They use the same chiplet approach as third gen Ryzen.  AMD has apparently decided that fourth gen is laptop only, or maybe will have desktop APUs at some point.  Presumably lower end Zen 3 CPUs are coming later.

AMD is also claiming a 19% IPC increase.  That is, at the same clock speed as before, programs will typically run 19% faster than on a Zen 2 CPU.  Added to the clock speed increase, that would mean Zen 3 is typically about 25% faster than the analogous Zen 2 CPU, and while using about the same power as before.

AMD claims that Zen 3 typically beats a Core i9-10900K at single-threaded performance and is the fastest desktop gaming CPU there is.  They also compared it to Tiger Lake at 4.8 GHz and said that Zen 3 beats it in single-threaded performance at Cinebench.  That has long been a very favorable benchmark for AMD's Ryzen CPUs, but I'd interpret it as a Ryzen 9 5950X being roughly competitive in single-threaded performance to Intel's top Tiger Lake CPU that has been paper launched for laptops but might not actually be available to buy for quite some time.

We'll see what independent benchmarks show.  AMD's history here is one of mild cherry-picking.  That is, the average benchmark AMD chooses tends to be a few percent more favorable than the average benchmark that independent reviewers choose.  Cinebench is, of course, much more cherry-picked than the rest of the benchmarks that generally showed a Ryzen 9 5950X beating a Core i9-10900K.

Zen 3 brings a lot of architectural changes as compared to Zen 2.  The key one that AMD was willing to explain is the unified 8-core complexes.  A Zen 2 chiplet had two complexes of four cores each, with a 16 MB L3 cache for each complex shared by the four cores in that complex.  In order for one core to access the other's L3 cache, it had to hop across chiplets to go through the I/O die.  Zen 3 features 8 cores with a unified 32 MB L3 cache that all cores can access.  That's the same total amount of cache as before, but the cores just have better access to it.

We'll see if Intel has any answer for this.  Rocket Lake is coming, but we don't yet know what it is.  Sky Lake Refresh Refresh Refresh Refresh Refresh would be dead on arrival unless Intel decides to slash prices to compete at the low end.  Backporting Willow Cove cores to 14 nm could be competitive in raw performance with Zen 3 if they can still clock high, but likely at a cost of double the power consumption or more.
TorvalTimukas

Comments

  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    remsleep said:
    "AMD claims that Zen 3 typically beats a Core i9-10900K at single-threaded performance and is the fastest desktop gaming CPU there is. "

    I think they are talking about Ryzen 9 5950X - which at $800 is $200 more than intel 10900k


    It's also 6 more cores.  The clock speeds are so close that if a Ryzen 9 5950X usually beats a Core i9-10900K, then a Ryzen 9 5900X probably will, too--and while being cheaper, having more cores, and using about half the power.  That's a big "if", of course, but if AMD is correct about the 19% IPC improvement, that should get it done.
    Gdemami
  • TorvalTorval Member LegendaryPosts: 21,719
    Awesome post. Thanks. This is exciting news to me.

    Even if AMD does beat Intel at IPC on an i9-10900K it won't matter much to me. IPC isn't the most important factor to me at all. I'm glad they'll have that talking point, but there are other things that overshadow IPC for me.

    For one, I don't have to buy and pay for a 'K' series tax when I choose to build AMD. That alone is about enough to be the deciding factor. My CPU isn't artificially and arbitrarily locked.

    On top of that I like the muliticore (multithreaded) performance AMD brings. I like how memory works on AMD compared to Intel. I like how well the stock cooler works on AMD. Even though my current rig uses a Noctua NH-U12S, when I build my daughter's new rig it will probably just use the Wraith.

    It's not that I think Intel CPUs are bad. It's that AMD CPUs and motherboards are excellent at a decent price/performance point and fit my particular set of workloads particularly well - database work, gaming, digital art, and sound. Those are about the 4 things our computers are used for beyond website browsing and social media.
    Gdemami
    traveller, interloper, anomaly, iteration


  • VrikaVrika Member LegendaryPosts: 7,174
    edited October 2020
    Quizzical said:

    Ryzen 5 5600X:  6 cores, 4.6 GHz, $300, 65 W
    I don't like their MSRP for this new 6-core processor. At base frequence it's worse price/performance than Ryzen 3600X was at its MSRP, and even at maximum turbo frequency it's only a couple of percents better price/performance assuming it really has that +19% IPC AMD promised.

    AMD is hiking the price up a bit too much.


    EDIT: With that said, if they really give that +19% IPC increase they're promising I think I'll be buying one of their new CPUs.
    [Deleted User]Asm0deus
     
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    When AMD was selling 6 cores for just over $100 several years ago, that wasn't because they felt like being nice.  It was because they had an inferior product, they knew it, and they priced it accordingly to still be a reasonable value.  Now that they think they own the top end, they're pricing their lineup accordingly.

    If you think Zen 3 is too expensive, then Zen 2 is still on the market, as are a variety of Intel processors.  There will also surely be lower end Zen 3 parts eventually.
  • TorvalTorval Member LegendaryPosts: 21,719
    Intel is competitive with IPC. We'll see if feature improvements like 32MB shared cache actually matter in real life. If not, then AMD will need to adjust their MSRP or suffer.
    traveller, interloper, anomaly, iteration


  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    remsleep said:
    Quizzical said:
    remsleep said:
    "AMD claims that Zen 3 typically beats a Core i9-10900K at single-threaded performance and is the fastest desktop gaming CPU there is. "

    I think they are talking about Ryzen 9 5950X - which at $800 is $200 more than intel 10900k


    It's also 6 more cores.  The clock speeds are so close that if a Ryzen 9 5950X usually beats a Core i9-10900K, then a Ryzen 9 5900X probably will, too--and while being cheaper, having more cores, and using about half the power.  That's a big "if", of course, but if AMD is correct about the 19% IPC improvement, that should get it done.

    From a pure gaming perspective - as long as there's enough cores to not cause a problem - how many more cores it has is irrelevant to me personally.

    I guess for me my bottom line - which one of the new Zen 3s beats 10900K in gaming - the number of cores and power use mean absolutely nothing to me - just curious to see what is the fastest for pure gaming.


    I would love to see AMD make a pure gaming CPU - that has 6-8 cores (plenty for gaming) and is 5Ghz.

    I don't need 16 cores to game, would gladly give up half those cores for higher core clock

    Sucks that to get the highest clock - you also have to pay for all the extra cores that you won't really need for gaming.
    AMD wants for the fastest CPU to be the most expensive one.  Intel does that, too.  If they can make a Core i9-10900K clock at some speed, they could surely make it clock at least as high and likely higher if they selectively disabled four of the cores.  But they want it to be the fastest at everything, so that if you want their best, you have to buy their most expensive.
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    Torval said:
    Intel is competitive with IPC. We'll see if feature improvements like 32MB shared cache actually matter in real life. If not, then AMD will need to adjust their MSRP or suffer.
    Willow Cove cores are probably competitive with Zen 3 in IPC.  Sky Lake {Refresh}* cores are not competitive with either in IPC, though they are about even with Zen 2 cores.

    The 32 MB unified L3 cache is one of the ways that Zen 3 increased IPC as compared to Zen 2.  Intel has had unified L3 caches since 2008, though usually much smaller, currently around 2 MB per core.
    Torval
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    Apparently all of the Zen 3 CPUs have a 142 W max turbo power, regardless of their nominal TDP.  That's the same as the max turbo power for Zen 2 CPUs.  The analogous number on the Core i9-10900K is 250 W.
    Torval
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    Apparently the I/O die in Zen 3 is the same as it was in Zen 2.  Not the just the same features.  It's literally the same die.  That should make motherboard compatibility easier.  It does mean no new I/O features in Zen 3, but they've already got plenty of PCI Express 4.0 lanes coming off the die, so there isn't anything obvious that they should have added.
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    One way to think about the IPC boost is that the IPC improvements of Zen 3 as compared to last year's Zen 2 are the same as Intel has gotten cumulatively over the course of the last five years.  In a sense, that's really just AMD catching up on IPC.  Zen 2 last year basically caught up to where Sky Lake was 5 years ago.  As you know, Intel was way ahead of AMD 5 years ago.  But now Zen 3 has caught up to the paper launched Tiger Lake (which uses the same cores as Ice Lake).

    We'll see where AMD and Intel go from here.  It's easier to make big improvements when you're way behind.
    Torval
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    By "the same chiplet approach", I meant that if you open up a Zen 3 package, you'll see the same number of dies as for a Zen 2 package and positioned in about the same places.  The CPU chiplets are different, but the I/O die is the same.

    Each Zen 3 CPU chiplet has a 32 MB L3 cache shared between all of the cores.  It's not 36 MB or 72 MB.  It's 32 MB.  A Zen 2 die had two separate 16 MB caches each shared between four cores.  Both Zen 3 and Zen 2 have a 512 KB L2 cache for each core that is accessible only by that core.

    AMD's numbers larger than 32 MB are adding multiple caches together.  They add their L2 and L3 cache sizes together just because it gives bigger numbers, even though it's stupid.  Furthermore, on the Ryzen 9 CPUs, even though the package has 64 MB of L3 cache, it's not a unified L3 cache directly accessible by all of the cores.  Each chiplet has 32 MB of L3 cache, and to go to the other chiplet's cache, you'll have to go through the I/O die.
    [Deleted User]Torval
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    xD_Gaming said:
    Newegg right now has ryzen 5 2700 / x for $150 That is a great price for anyone a few generation old. Then you can get a AMD 570x mobo and have an upgrade path to ryzen 3. 
    If you want Zen 3, why buy a $150 CPU now and then replace it a month later?  Why not just wait a bit for Zen 3 to launch?
    [Deleted User]
  • CleffyCleffy Member RarePosts: 6,373
    I imagine the reason for the higher clock on the 12 and 16 core has to do with heat. I am assuming those are boost clocks. The 5800 has the highest base clock. That's probably a benefit of boosting clocks of physically separate dies. 
  • VrikaVrika Member LegendaryPosts: 7,174
    Cleffy said:
    I imagine the reason for the higher clock on the 12 and 16 core has to do with heat. I am assuming those are boost clocks. The 5800 has the highest base clock. That's probably a benefit of boosting clocks of physically separate dies. 
    Those boost clock differences are for marketing reasons. AMD goes out of their way to give CPUs with more cores also better boost clock so that people would buy them rather than the cheaper models. Also Intel does same for their own CPUs.
    Quizzical
     
  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    Cleffy said:
    I imagine the reason for the higher clock on the 12 and 16 core has to do with heat. I am assuming those are boost clocks. The 5800 has the highest base clock. That's probably a benefit of boosting clocks of physically separate dies. 
    It's really just binning.  The best dies go into the most expensive CPUs.  AMD and Intel both want to make sure that the most expensive CPUs are the best at everything.

    During the early days of multicore processors, Intel had Core 2 Duo CPUs that could clock higher than their Core 2 Quad ones.  That led some people to say hey, hardly anything will use more than two cores, so the Core 2 Duo is faster than the Core 2 Quad because it clocks higher.  So people who were willing to pay for the best without regard to the price tag ended up getting a cheaper CPU.  Naturally, Intel was unhappy about that.

    That four cores couldn't clock as high as two was really just physics back when all cores had the same clock speed.  Then Intel introduced power gating and turbo with Nehalem.  With Sandy Bridge, they got turbo working well, and ever since then, the CPUs that had the highest stock turbo were always the ones with the most cores and also the most expensive in the lineup, at least among the mainstream consumer models.
  • AmazingAveryAmazingAvery Age of Conan AdvocateMember UncommonPosts: 7,188
    Great looking CPUs shame they did the tests in 1080p, there will be little variance at 4k in games.



  • QuizzicalQuizzical Member LegendaryPosts: 23,898
    Great looking CPUs shame they did the tests in 1080p, there will be little variance at 4k in games.
    Well of course you do the CPU benchmarks at settings that turn the CPU into the bottleneck.
Sign In or Register to comment.