Speculation: Ryzen 3000 series

Page 14 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

What will Ryzen 3000 for AM4 look like?


  • Total voters
    230

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
I also added a spoiler tweet to temper mine and others enthusiasm. Cobalt caps are used by Intel 14nm, TSMC 16nm and Samsung/GF 14nm. But my understanding is Scotten was talking of cobalt being used elsewhere such as contacts, which is the most likely place in TSMC 7nm, as they use copper for metal interconnect. I think the recent leaks of Ryzen 3k CPUs maxing out at 5 Ghz are possible if TSMC 7nm uses cobalt contacts. If that happens AMD will take the ST perf crown in H2 2019 most likely or atleast match Intel in ST perf while crushing them in MT perf with 16c/32t Ryzen 9, 12c/24t Ryzen 7, 8c/16t Ryzen 5 and a likely 8c/8t Ryzen 3 imo. I think AMD is going to celebrate its 50th anniversary by taking the PC and x86 server performance leadership. Go AMD.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Looking forward to where AMD takes their platform in 2019 and beyond. I'm happy to have gotten on board early and thus be intimately familiar and comfortable w/ their family of products. I hope to consolidate a number of nodes into one platform in 2019 and I am quite excited about the I/O potential that has been detailed so far :D
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
AMD will take the ST perf crown in H2 2019 most likely or atleast match Intel in ST perf while crushing them in MT perf with 16c/32t Ryzen 9, 12c/24t Ryzen 7, 8c/16t Ryzen 5 and a likely 8c/8t Ryzen 3 imo

If the 5 ghz numbers and higher IPC are true eg. they can match intel in ST or are much closer than now, why would the even need a 16 core part? They could sell the 8-core for $400 or even $450 and still make the 9900k obsolete. The >8 core parts are much more likely for the refresh eg. ryzen 4000-series. Threadripper3 can then fill the niche of 16-32 cores and TR4 refresh can then up that to 64.

If they up cores now, what an they do with 4000-series?
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
If the 5 ghz numbers and higher IPC are true eg. they can match intel in ST or are much closer than now, why would the even need a 16 core part? They could sell the 8-core for $400 or even $450 and still make the 9900k obsolete. The >8 core parts are much more likely for the refresh eg. ryzen 4000-series. Threadripper3 can then fill the niche of 16-32 cores and TR4 refresh can then up that to 64.

If they up cores now, what an they do with 4000-series?

AMD's value proposition needs to much stronger than Intel because of Intel's brand and strong industry clout. AMD Ryzen was successful because it offered vastly superior MT perf and good ST perf even though Intel had a significant ST perf lead. With Zen 2 AMD can look to cement their position by taking ST perf crown and crushing Intel in MT perf. Intel is already readying a Comet Lake with 10c/20t so AMD needs to deliver something which will not be easy to beat. 16c/32t Ryzen 9, 12c/24t Ryzen 7 and 8c/16t Ryzen 5 would give them a unassailable position for atleast 2 years. AMD's goal will be to maximize share gains and sell every CPU they can make at the highest possible prices. Striking the balance between market share gain, revenue and profits is what AMD will be looking to do.
 

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
So we're potentially talking about matching ST and doubling the core count, all at the same price point, which essentially translates to a higher margin per sale; fixed costs spread over a large volume of sales.
That sounds like perfection.
The other aspect of it is that they likely have higher production capacity and yields based upon their process and design choices, so the goal of extending market share is entirely achievable.
Of course, Intel is a behemoth that isn't going to just disappear like a fart in the wind. It'll need to reassess its model, but it'll continue to be highly profitable...even into the medium term without significant improvements of its own.
Good times ahead.
 

Joe Braga

Member
Dec 31, 2017
25
10
51
I wanna know if you believe that's possible a Ryzen 9 16c/32t, Ryzen 7 12c/24t, Ryzen 5 8c/16t, Ryzen 3 6c/12t e then threadripper3 lineup starts at 24c/48t e up to 48c/96t
 

Joe Braga

Member
Dec 31, 2017
25
10
51
I wanna know if you believe that's possible a Ryzen 9 16c/32t, Ryzen 7 12c/24t, Ryzen 5 8c/16t, Ryzen 3 6c/12t e then threadripper3 lineup starts at 24c/48t e up to 48c/96t
Completing my question and about Ryzen 3000 series Clock Is it possible being up to 5.1ghz?
 

PotatoWithEarsOnSide

Senior member
Feb 23, 2017
664
701
106
I have more questions over the clocks than the core count TBH.
Either way, a 9900k runs at 4.2GHz 24/7 within TDP limits, so if Ryzen 3000s can do the same, then that core count advantage really drives home.
I reckon an 8c APU wins the war regardless of how individual battles pan out.
 

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
Completing my question and about Ryzen 3000 series Clock Is it possible being up to 5.1ghz?
I'd just wait until the parts are out and tested. Speculation is fun, but it often drives up expectations that may or may not be warranted. This goes for either AMD or Intel btw.
 

Joe Braga

Member
Dec 31, 2017
25
10
51
I have more questions over the clocks than the core count TBH.
Either way, a 9900k runs at 4.2GHz 24/7 within TDP limits, so if Ryzen 3000s can do the same, then that core count advantage really drives home.
I reckon an 8c APU wins the war regardless of how individual battles pan out.
Could AMD beats Intel this scenario within or not the TDP limits e could the Overclock limit ends Zen2 and the Overclock limit beats 5.3Ghz on Coolerbox?
 

Joe Braga

Member
Dec 31, 2017
25
10
51
If the 5 ghz numbers and higher IPC are true eg. they can match intel in ST or are much closer than now, why would the even need a 16 core part? They could sell the 8-core for $400 or even $450 and still make the 9900k obsolete. The >8 core parts are much more likely for the refresh eg. ryzen 4000-series. Threadripper3 can then fill the niche of 16-32 cores and TR4 refresh can then up that to 64.

If they up cores now, what an they do with 4000-series?
I don't understand why in 2018 we still discuss with more focus in ST Performance and not in MT performance?
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
I don't understand why in 2018 we still discuss with more focus in ST Performance and not in MT performance?

mainly because you need ST performance for games, its a matter of getting stuff done faster, every ms matters, no matter how much MT you code is, ST will always matter. 1ms CPU frame render time IS A LOT for a game.

Its a different story for content creation where beign 1ms faster or slower per thread dosent matter and you actually need a lot of CPU power.
 

Joe Braga

Member
Dec 31, 2017
25
10
51
mainly because you need ST performance for games, its a matter of getting stuff done faster, every ms matters, no matter how much MT you code is, ST will always matter. 1ms CPU frame render time IS A LOT for a game.

Its a different story for content creation where beign 1ms faster or slower per thread dosent matter and you actually need a lot of CPU power.
The games developer should migrate from OpenGL /DirectX11 to Vulkan/DIrectX12 for better multithread usage
 
  • Like
Reactions: DarthKyrie

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
The games developer should migrate from OpenGL /DirectX11 to Vulkan/DIrectX12 for better multithread usage

Dosent matters, first off, Vulkan/DX12 graphic rendering is just a part of the overall time to render 1 frame, this is why DX11 is still used and it still gives good results at zero extra dev time, compared to DX12/Vulkan.

Second, it dosent matter how much you can split the work intro multiple threads, faster ST perf means that work will be done faster and even 1ms diference is a big thing for a game. Also there is a limit on how much you can split the work before it start giving negative results.

Unless the games start needing to process huge amount of data compared to today this will stay this way. And this havent changed much in the last 15 years, Denuvo is the reason of why 4 core / threads arent enoght these days.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
The games developer should migrate from OpenGL /DirectX11 to Vulkan/DIrectX12 for better multithread usage

They already have, but at the end of the day there will always be one thread that has a higher load and ends up being the limiting factor. As soon as you figure out how to add parallelism to whatever that part of the game may be, you just end up with something else being the new bottleneck.

I don’t think it’s a real issue though as most games are GPU bound at higher resolutions and in the cases where people put the settings low enough to where the CPU becomes the limiting factor, the frame rates are higher than the display can output anyways.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
They already have, but at the end of the day there will always be one thread that has a higher load and ends up being the limiting factor. As soon as you figure out how to add parallelism to whatever that part of the game may be, you just end up with something else being the new bottleneck.

I don’t think it’s a real issue though as most games are GPU bound at higher resolutions and in the cases where people put the settings low enough to where the CPU becomes the limiting factor, the frame rates are higher than the display can output anyways.

Even with a perfect MT scaling it will still be a matter on how fast every one of these threads can finish the work for 1 frame. Unless the CPU cant process the volume of data fast enoght. And games offload most of the heavy stuff to the GPU (shader compute) because GPU are just faster for that.

Again if it wast for Denuvo, 4/4 would still be enoght for every game.
 
Last edited:
  • Like
Reactions: Joe Braga

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
mainly because you need ST performance for games, its a matter of getting stuff done faster, every ms matters, no matter how much MT you code is, ST will always matter. 1ms CPU frame render time IS A LOT for a game.

VM response times are also important.
 

Glo.

Diamond Member
Apr 25, 2015
5,657
4,409
136
I have more questions over the clocks than the core count TBH.
Either way, a 9900k runs at 4.2GHz 24/7 within TDP limits, so if Ryzen 3000s can do the same, then that core count advantage really drives home.
I reckon an 8c APU wins the war regardless of how individual battles pan out.
AMD's 8Core/16T CPU will run @4.2 GHz all core Turbo within 65W TDP, if the clocks and rumors are correct.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
The games developer should migrate from OpenGL /DirectX11 to Vulkan/DIrectX12 for better multithread usage

Experience in the last couple years have shown that dx12 for some reason doesn't work or is too hard if even dice /BF-series can't get it to work right. Besides that not everything can be split into threads.

It's not just games but many applications are still single-threaded. Browser might use multiple processes/threads but rendering of a single page is still single-thread performance dependent.

They already have, but at the end of the day there will always be one thread that has a higher load and ends up being the limiting factor.

true but I want to add my gripe with dx12 and vulkan: There has not been a single-game /game-engine that is dx12 /vulkan only. The paradigms are so different that IMHO dx12 simply can't work correctly as long as the game also supports dx11. dx12 is simply tacked on. Even AotS has dx11 path. You have to limit your game /engine to what can be done with dx11 and then replace the code with dx12. What should be done is pure dx12. Only then if ever will it shine.

Currently it like building a race car on top of a truck platform. It's still a truck deep down and it shows.
 
  • Like
Reactions: spursindonesia
Mar 11, 2004
23,030
5,495
146
Someone else has probably suggested this, but I wonder if AMD didn't look at the Spectre/Meltdown stuff (and IIRC quite a few of the cloud processing setups disable multi-threading) and decide that pushing core counts would be especially beneficial at this time.

As for DX12, it'll take time as companies do new game engines that are built with it in mind (so that it won't take most developers that are using UE or Unity, etc, to implement it). That will probably start to happen this next go around. Doesn't the DX ray-tracing API that they're trying to push require DX12?

I think the biggest issue is that DX12/Vulkan got announced pretty early into this current console generation, and even though their hardware supports it decently well, it wasn't ready soon enough for the major game engines of this gen to be developed with it in mind. They could patch it in but it was never going to usurp DX11 that quickly (which DX11 seems to have taken DX9's place as this weirdly dominant DX version for whatever reason).

I think that's why Microsoft added that DX draw call handler to the GPU in Scorpio, as they knew developers just weren't going do the extra work yet and so it was a way to ease that issue. Plus it has benefits for older games as well (wonder if it might have been necessary in order to do the 4K or higher framerate 1080p rendering on Scorpio of older games as they did their backwards compatibility push).

What would be really interesting is if the next gen engines better utilizing DX12/Vulkan, might actually enable ports of games using them to run better on the current gen than the current gen engines do, making them viable longer. Which for Sony and Microsofts sake, I hope they try and push utilizing the extra CPU grunt the next gen will bring to push for better game AI and physics.
 

Shivansps

Diamond Member
Sep 11, 2013
3,835
1,514
136
Vulkan/DX12 was only marketing and it petty much still is, as graphic engine MT is just ONE PART of the problem, by itself it solves nothing, thats not were the big problem is. And taking advantage of it to even mach DX11 perf takes way too much work.

If amd/ms/sony put 8 zen2 cores in the consoles then it's a self fulfilling that gamers will need more then 8 cores on the pc side.

In the same way that having x86 in consoles was going to help game port performance? -or- having a AMD gpu was going to cause games to run better on AMD hardware? Im not falling for this again.

Also, why? Are they going to mine with the game now? Denuvo, inside denuvo, inside denuvo, inside denuvo copy protection? In what they are going to use all that proccessing power for? Maybe you havent been checking but devs are more worried about sending political messages than in developing a game lately.

Anyway im looking forward to see brain dead AI now needing 8 cores to be even more brain dead. Meanwhile games like X4 that has complex collision avoidance, constant collision checking, real simulated economy and ton of ships in-game runs at the same FPS in my system on 2C/4T than on 8C/16T? You know why? It dosent run the cancer called Denuvo. BTW, it is a Vulkan game.
 
  • Like
Reactions: ZGR and lightmanek

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
In the same way that having x86 in consoles was going to help game port performance? -or- having a AMD gpu was going to cause games to run better on AMD hardware? Im not falling for this again.
But both of those has happened. Game development is a lot closer to CPU development then people give it credit for. Last year the first games that started life developed on the Xbone or PS4 as primary platforms just started to come out. This year would have been the first year that just about every game started on the platform. Even PC only games. Some have ported DX11 to DX12 or Vulkan. But we are still rarely seeing games that started life on DX12/Vulkan. We should see more in the future. But if it wasn't for the 7k series and AMD's success in getting into the consoles, we wouldn't have seen DX12. Vulkan would be even more of an after thought than it is now. But ports are easier, the include a lot more MT work then they ever did before and advanced API's are in use. Over all development is still restricted mediocre GPU performance and out right crappy cpu compute power and until a couple years after then next Gen consoles come out this will still be an issue. But it is getting better and will continue to get better till we get to that point. Things like them finally putting work back into the AI's. A perfect example of this is rewinding a couple years to Alien Isolation. We were stuck with absolute brain dead robot AI (which worked well in the context of the story) mainly because they needed all of compute process they could get for the single Alien AI. Obviously this is made worse by the fact that it was in development for X360 and PS3. But the point still stands. Maybe the next time someone tries. They don't have to nueter all the other ai and physics work to pump up the AI of a particular bad guy.
 
Feb 4, 2009
34,494
15,729
136
I just learned why DX12/Vulcan always sounds cool but never works right or better than DX11.
Thank you guys.