Speculation: Ryzen 3000 series

Page 104 - 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
    192

StinkyPinky

Diamond Member
Jul 6, 2002
6,386
70
126
12c with a boost to 4.7ghz would be huge. I certainly think it's possible. I really don't think AMD will release a 16c initially. That will probably be either around xmas or they will hold off entirely until Intel releases something to compete with the 12c. Releasing a 16c right away would seriously hurt their threadripper range and why bother to cannibalize your own products when the competition is stuck on 8c for the same price.
 

B-Riz

Senior member
Feb 15, 2011
965
58
91
My bold predictions, based on AMD targeting Intel 10nm as the competition for their 7nm and Zen / Zen+ being the warm up chips as they knew what needed fixed and what could be improved with Zen2.

(If you think about it, Zen / Zen+ are pretty boss for being built on a process that was not really intended to compete with what Intel has done with 14+ / 14++)

AMD will match Intel on performance at every price point / core configuration.

They have something that matches the i9 9900K and will be cheaper.

They have something that beats the i9 9900K, but, it will cost the same or more.

Intel will lower prices to compete with AMD at Zen2 physically release.

It has taken longer to launch Zen2 because AMD does not want the motherboard fiasco repeated from the Zen launch.

I'm pretty sure this will happen, as AMD started the core war, and Dr. Su is just that dang good at her job.
 
Apr 27, 2000
11,512
843
126
What I don't know and nobody seems to bring up the question is, since the new ryzen is IO die + chiplets, how will the IO die be clocked? I assume all the talk is about chiplets clocks.
My guess is the I/O die will have a clockspeed based on IF/RAM speed. It will probably sync to IF directly, with IF running at half the DDR4 rating (or 1/4 the DDR4 rating if the user selects that option, which will apparently be a new feature of Zen2).

12c with a boost to 4.7ghz would be huge. I certainly think it's possible.
I would settle for an 8c with a 4.7 GHz boost. Higher if I can get it.
 
Feb 4, 2009
20,070
1,081
126
My guess is the I/O die will have a clockspeed based on IF/RAM speed. It will probably sync to IF directly, with IF running at half the DDR4 rating (or 1/4 the DDR4 rating if the user selects that option, which will apparently be a new feature of Zen2).



I would settle for an 8c with a 4.7 GHz boost. Higher if I can get it.
Im no expert but coincidentally I came here to post this about memory support rumors

https://www.google.com/amp/s/www.te...en-2-a-memory-oc-beast-ddr4-5000-possible?amp
 
Apr 27, 2000
11,512
843
126
My x370 Taichi has DDR4-4000 as an option, but there's no way my 1800x can actually run RAM that fast. I'm more interested in seeing how fast IF will run on Zen2. As much as I want to see high memory speeds, I want to see high IF speeds as well. Higher than what we see on Summit Ridge or Pinnacle Ridge anyway.
 
Feb 4, 2009
20,070
1,081
126
My x370 Taichi has DDR4-4000 as an option, but there's no way my 1800x can actually run RAM that fast. I'm more interested in seeing how fast IF will run on Zen2. As much as I want to see high memory speeds, I want to see high IF speeds as well. Higher than what we see on Summit Ridge or Pinnacle Ridge anyway.
I’m just hoping it’s less bitchy about what speed it runs with 2 dims vs 4 dims or what type of memory Samsung vs anything else.

This is my #1 concern and the main thing that could really piss me off
 
Apr 27, 2000
11,512
843
126
I’m just hoping it’s less bitchy about what speed it runs with 2 dims vs 4 dims or what type of memory Samsung vs anything else.

This is my #1 concern and the main thing that could really piss me off
The real question is whether AMD has been testing early samples of 10y-based DDR4 (10y ICs are already in some LPDDR4 I think). That stuff could get stupid fast, and it'll be important for Zen2 to support as much of it as possible.
 
Feb 23, 2017
507
424
106
What would be the benefit of running IF at 1/4 of RAM clocks?
Seems contrary to early rumours we had about IF clocks having been doubled this time around.
 
Feb 4, 2009
20,070
1,081
126
The real question is whether AMD has been testing early samples of 10y-based DDR4 (10y ICs are already in some LPDDR4 I think). That stuff could get stupid fast, and it'll be important for Zen2 to support as much of it as possible.
You need to dumb this down for me please
I get the low power part but that’s about it
 
Jan 26, 2004
40
20
81
What would be the benefit of running IF at 1/4 of RAM clocks?
Seems contrary to early rumours we had about IF clocks having been doubled this time around.
Well, because DDR speeds with ryzen are limited to what IF can do, IF runs at half the DDR speed now.So when you are pushing ram to 4000MT/s, IF has to be able to do 2000Mhz, and 90% of them cant do it, BUT, if you introduce 1/4 divider even ram at 5000mhz speed runs IF at just 1250Mhz so every chip should be able to cope.
Whole another thing is, what will be actually faster, a CPU with 5000MT/s ram and 1250Mhz IF or a chip with 3733MT/s ram and 1866 IF...
As for IF multis, its nothing new i think, if i remember correctly stilt just after zen1 or even before that mentioned that there are IF dividers/multis not open to general public and that was zen1.
 
Feb 23, 2017
507
424
106
My instinct tells me that it'd be increasing bandwidth at the expense of latency, and since latency was the biggest issue with Zen(+) then it seems counter-intuitive.
 
Jan 26, 2004
40
20
81
My instinct tells me that it'd be increasing bandwidth at the expense of latency, and since latency was the biggest issue with Zen(+) then it seems counter-intuitive.
Well its a complicated issue, alot depends on bandwith of IF, if ist going to be sufficent for that ram speed or not.Also latency decreases with ram clock and timings so thats another set of variables.
My bet is on 3733ram with lower timings and higher IF tho for gaming.
 

DarthKyrie

Senior member
Jul 11, 2016
544
385
116
I may have heard wrong but didn't someone from AMD say during the event when they showed off Epyc2 that IF is now tied to PCIe 4.
 
Jan 26, 2004
40
20
81
Guys, i know its kinda lame, but these 3700X 5ghz boost speeds seems more and more likely.
There are many sites which list them at that speed as a fact, and not only some weird taiwanese shops.
Here, a polish shop in EU.
Its in polish but u get the clocks.
https://www.morele.net/wiadomosc/amd-ryzen-7-3700x-wydajnosc-oraz-specyfikacja/14012/
techpowerup it lists it now with these also.
https://www.techpowerup.com/cpudb/2129/ryzen-7-3700x
Some photography site:
https://www.fullexposure.photography/amd-ryzen-3700x-vs-2700x/4/

Redgamingtech, the same guy which correctly got the vega20 news early also talks about those clocks today.

I mean.MSI kinda leaked their boards for computex launch in may so they are coming soon.
And they are making x570 with the highest tiers of their motherboards,godlike and creation.
Im kinda getting on the hype train now i guess :p , theres too much chatter about this.
16 core Am4 chips and high end boards also makes Threadripper less useful, so the cancelation or postponing of them makes sense.
 
Jan 26, 2004
40
20
81

Shivansps

Platinum Member
Sep 11, 2013
2,476
204
126
I find hard to belive that AMD archived core and freq increase on a new node, i belive this never happened before?

16C... even at 4ghz its amazing, similar for 12C at 4.2Ghz, always talking of ACT turbo... Now sure why everyone wants to beat 9900K freqs at all costs, it is a 8C.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,359
148
126
I find hard to belive that AMD archived core and freq increase on a new node, i belive this never happened before?
45nm PDSOI Quad-core Bulldozer had a F_nom of 3.2 GHz, 32nm PDSOI Octo-core Bulldozer had a F_nom of 3.5 GHz.

=> HKMG
=> Next-gen Strain/Stress
=> etc, however.

TSMC's FinFETs are a severe downgrade from GlobalFoundries' FinFETs. So, the frequency guaranteed for 9T-HPC @ GlobalFoundries, will definitely not be achieved at 7.5T-HPC @ TSMC. 6T vs 6T edges for GlobalFoundries being able to go 1.5x power for 1.7x performance, whereas TSMC went 1.25x power for ~1.475x performance. GlobalFoundries also reduced various Resistance, Capacitance, leakage metrics to allow for really absurd speeds; IBM z, power, etc.
 

Topweasel

Diamond Member
Oct 19, 2000
4,768
363
136
Well, because DDR speeds with ryzen are limited to what IF can do, IF runs at half the DDR speed now.So when you are pushing ram to 4000MT/s, IF has to be able to do 2000Mhz, and 90% of them cant do it, BUT, if you introduce 1/4 divider even ram at 5000mhz speed runs IF at just 1250Mhz so every chip should be able to cope.
Whole another thing is, what will be actually faster, a CPU with 5000MT/s ram and 1250Mhz IF or a chip with 3733MT/s ram and 1866 IF...
As for IF multis, its nothing new i think, if i remember correctly stilt just after zen1 or even before that mentioned that there are IF dividers/multis not open to general public and that was zen1.
Small technical point, IF runs at the actual clock rate of DDR. Not half.
 

alexruiz

Platinum Member
Sep 21, 2001
2,488
14
106
Personally, I do NOT think the 6C/12T will be Ryzen 3 3300 series.

6C / 12T will stay Ryzen 5 3600 series
8C / 16T will stay Ryzen 7 3700 series
a potential 12C / 24T will be Ryzen 9 3800 series
a potential 16C / 32T will be Ryzen 9 3900 series
6C / 6T will be Ryzen 3 3500 series.
No 3200, 3300 or 3400 series.

This is a fairly straight forward market segmentation that keeps some halo into the 8C / 16T parts
But as usual, we will see
 
Oct 27, 2006
19,676
168
106
Personally, I do NOT think the 6C/12T will be Ryzen 3 3300 series.

6C / 12T will stay Ryzen 5 3600 series
8C / 16T will stay Ryzen 7 3700 series
a potential 12C / 24T will be Ryzen 9 3800 series
a potential 16C / 32T will be Ryzen 9 3900 series
6C / 6T will be Ryzen 3 3500 series.
No 3200, 3300 or 3400 series.

This is a fairly straight forward market segmentation that keeps some halo into the 8C / 16T parts
But as usual, we will see
I agree with the 6C not being a 'Ryzen 3'. Perhaps not at launch, but I do expect a later 7nm 3000-series Quad Core, and more APUs.

Right now, probably the best to get the highest ASP possible given limited production capacities. I think some form of 'harvesting' is likely, as it's simply too inefficient to not at least consider it.

But particularly with uSFF/NUC style boxes, ultrabooks, etc, minimum 6C on the roadmap doesn't seem ideal to me. We all love cores and performance, but a lot of applications simply don't need it, or it would be a waste to run a fatter/hotter/pricier component where it's overkill. For now, that may be supplied for a length of time by 2xxx products, but I don't think that makes sense to continue indefinitely.
 
Aug 25, 2001
43,608
539
126
But particularly with uSFF/NUC style boxes, ultrabooks, etc, minimum 6C on the roadmap doesn't seem ideal to me. We all love cores and performance, but a lot of applications simply don't need it, or it would be a waste to run a fatter/hotter/pricier component where it's overkill. For now, that may be supplied for a length of time by 2xxx products, but I don't think that makes sense to continue indefinitely.
Well, there's always GF, the WSA, and the possibility of re-badges. Though, that might make the Ryzen brand look bad if they choose to do that.

I assume that they will still have a 2C/4T APU, at some point, whether 12nm or 7nm, I don't know, but in the 3000-series. (Or Athlon 300-series.)
 
Apr 27, 2000
11,512
843
126
You need to dumb this down for me please
I get the low power part but that’s about it
It's okay. I was caught off-guard by the 10x and 10y thing.

Apparently, DRAM manufacture is "stuck" on a 10nm node. So node refinements are all they can achieve. I don't know all the technical details behind it. Anyway, DDR4 as we know it is manufactured on the older 10x node. 10y node should feature higher speeds and lower voltages. There is a vanishingly small amount of 10x DDR4 that can bin for DDR4-5000, but 10y should be a different story.

I myself am sitting on some 10x DDR4-4400 which should be "good enough" for me until DDR5 hits the scene. I hope!

Whole another thing is, what will be actually faster, a CPU with 5000MT/s ram and 1250Mhz IF or a chip with 3733MT/s ram and 1866 IF...
That is something I hope to test. Not sure if my RAM will be up to DDR4-5000 or not, but if Zen2 can handle the XMP settings then I can at least test DDR4-4400. Higher IF speeds will equal lower IF latency, which should in turn improve performance for any workload where intercore latency is a concern.

Small technical point, IF runs at the actual clock rate of DDR. Not half.
Actually, no. Check this out:

https://www.transcend-info.com/Support/FAQ-296

JEDEC members have been employing tricks over the years to increase data transfer rate without (necessarily) increasing memory clocks. For example, DDR4-3200 operates at only 200 MHz internally. On a Ryzen system running DDR4-3200, the IF frequency is 1600 MHz, which is eight times that of the memory clock. IF is actually linked to the memory bus clock in this circumstance. If we want to be technical, then, there you have it.

It appears as though most Summit Ridge chips crapped out at an IF speed of around 1066-1233 1733 MHz. Pinnacle Ridge was a bit better, hitting speeds of 1800 MHz for a non-trivial number of overclockers. I am hoping Zen2 can manage at least 2000 MHz IF.
 
Last edited:


ASK THE COMMUNITY