AMD Zen 3 (Ryzen 4000) already in the works

[Woodsta888]

So what you're saying then is that 4.2 is not achievable for many in real world conditions.

So it's essentially unachievable by many. Ie, somebody who buys an off-the-shelf PC might never seen that 4.2 clock as stated on the packaging.

That doesn't play well with me no matter how you spin it.

Looks like in their attempt to kill overclocking (and deny people "free" performance), they've gone too far the other way; now people can't even reach the advertised "up to" speeeds.

Yes I konw "up to" is supposed to be a get-out clause. But it's sleazy to behave this way. AMD do not endear themselves to us by taking notes from Intel's playbook.

I dont have to turn anything off or slim down what windows is running to get 5/8 cores to do 4.6Ghz+ on my X570 board with my 3800X. Most people complaining that performance isn't there is often on older chipset designs. They wouldn't have designed X570 if it wasn't required.

It is well documented that X570 boards have much more powerful VRM's etc requiring additional cooling right up to 200A. Older chipsets wont usually get into the realms of 150A and many require extra cooling at 100A. You only have to look at how many phases are on X570 compared to older chipsets.

Many of the people complaining about not reaching the advertised clocks are doing so on older boards. Can you really expect the last few Mhz from older boards when the new chipset is designed much more powerfully? Why wouldn't a MOBO manufacturer spend more time with the BIOS's on X570 that folk have just bought new getting that right, more so than older boards? Whilst Zen2 IS backward compatible on older boards you may not get the full performance just due to the power requirements seen under X570 boards.

Bet you don't find the worlds OCers going for max scores on anything less than an X570 board as other chipsets just wont have the power delivery configuration X570 has.

I bought mine as an enthusiast knowing that the BIOS's were immature etc. The new AGESA 1.0.0.4 which has to come from AMD themselves fixes a lot of things and actually adds new functionalities. So finally we are getting a BIOS somewhat worthy, that maybe should have been available on release day. Any 'enthusiast' who read up would know upon buying Zen2 that you'd be entering early adoption phase.

Can anyone seriously expect maximum performance from anything other than X570? Benchers and OCer's are they at the top of scoreboards on older chipsets?

Lets hope the roll out of AGESA 1.0.0.4 isn't plagued with bugs etc with the addition of extra functionality.

 

[Grim5]

Rumor : AMD Zen 3 Architecture to Support up to 4 Threads Per Core With SMT4 Feature

https://wccftech.com/rumor-amd-zen-...t-up-to-4-threads-per-core-with-smt4-feature/

AMD shot this one down - Ryzen 4000 will continue the 2 x SMT like the current gen

 

[LePhuronn]

AMD shot this one down - Ryzen 4000 will continue the 2 x SMT like the current gen

Ryzen 4000 perhaps, but that's not EPYC or Threadripper. Just because desktop doesn't get SMT4 doesn't mean server won't. But that poses the question how do you discern between SMT2 and SMT4 in a common pool of chiplets? Unless the I/O die can be responsible for the different modes.

 

[4K8KW10]

Zen 3 IPC Gains Are ‘Greater’ Than 8 Percent – EXCLUSIVE
http://www.redgamingtech.com/zen-3-...-than-8-percent-exclusive/?utm_source=ixbtcom

 

[Grim5]

That would put it well above Intels best 14nm but under Intels 10nm Ice Lake (just speaking IPC here)

 

[Panos]

That would put it well above Intels best 14nm but under Intels 10nm Ice Lake (just speaking IPC here)

The article refers to IPC only, there is a speed bump which already shows +200mhz over Zen2 for the Engineering sample.

Now, the "estimate" IPC performance of the IceLake is +18% over the CFL (9900K), which would bring it in par of the worst case estimate for Zen3. The problem is Intel 10nm Ice Lake has issues with the clocks.
Also there are issues with the size of the chips. 6-8 cores, are very hard to made because of the failure rate. After all those years, Intel barely manages to make quad core low power parts.
Up to last year couldn't make anything bigger than dual core at 10nm.

 

[Doobedoo]

Ryzen 4000 perhaps, but that's not EPYC or Threadripper. Just because desktop doesn't get SMT4 doesn't mean server won't. But that poses the question how do you discern between SMT2 and SMT4 in a common pool of chiplets? Unless the I/O die can be responsible for the different modes.

Haven’t they already confirmed that the next Threadripper/Epic will be smt2 in a recent roadmap?

It shows it here https://www.overclock3d.net/news/cp...require_a_new_socket_-_they_told_us_in_2017/1

 

[4K8KW10]

That would put it well above Intels best 14nm but under Intels 10nm Ice Lake (just speaking IPC here)

The article refers to IPC only, there is a speed bump which already shows +200mhz over Zen2 for the Engineering sample.

Now, the "estimate" IPC performance of the IceLake is +18% over the CFL (9900K), which would bring it in par of the worst case estimate for Zen3. The problem is Intel 10nm Ice Lake has issues with the clocks.
Also there are issues with the size of the chips. 6-8 cores, are very hard to made because of the failure rate. After all those years, Intel barely manages to make quad core low power parts.
Up to last year couldn't make anything bigger than dual core at 10nm.

It has been previously commented that the "estimates" on the intel slides are over-exaggeration and shouldn't be taken as corresponding exactly to the real-world performance numbers?!

>8% IPC uplift + 100-200 MHz for Zen 3 is not bad at all.

 

[Panos]

It has been previously commented that the "estimates" on the intel slides are over-exaggeration and shouldn't be taken as corresponding exactly to the real-world performance numbers?!

>8% IPC uplift + 100-200 MHz for Zen 3 is not bad at all.

True. But looks like IPC wise, Ice Lake & Zen 3 would be in par.

 

[Troezar]

True. But looks like IPC wise, Ice Lake & Zen 3 would be in par.

Looking good for competition and choice/good prices for the consumer. I'm curious to see what CPU AM4 will top out at.

 

[Panos]

Looking good for competition and choice/good prices for the consumer. I'm curious to see what CPU AM4 will top out at.

We shall see.

 

[LePhuronn]

Haven’t they already confirmed that the next Threadripper/Epic will be smt2 in a recent roadmap?

Note that I posted on October 7th, your article is dated October 8th. So yes, now it's looking likely that Zen 3 is sticking with 2 threads per core.

 

[theforce]

Hopefully, it will provide a nice jump in performance over Zen2. My i5-3570k has started showing it's age.

 

[Doogles]

Hopefully, it will provide a nice jump in performance over Zen2. My i5-3570k has started showing it's age.

If you need to upgrade get onto Zen 2 now, glad I made the move from my 3770k, the lack of cores really hurt in recent games.

 

[Perfect_Chaos]

Lower latency and 8+% IPC plus 100-200Mhz will give quite a boost if it's true, i will only upgrade from 3700x if it is, and being the last chip on AM4 hopefully it's a good boost. Depends on what's forecast for 2021/2022 on Intel's/AMD's side also for me.
I think i will skip the first 6-12 months of DDR5 until it gets up to speed too.

 

[Deleted member 209350]

Rumours that Zen3 is hitting 100-200mhz higher clock speeds with also double digit % IPC boost

https://www.pcgamesn.com/amd/zen-3-ipc-increase

Here I was thinking that Zen3 wouldn't be too much of an upgrade on the current Zen2!

 

[LePhuronn]

Here I was thinking that Zen3 wouldn't be too much of an upgrade on the current Zen2!

Restructured CCD design, SMT4, shrunk I/O with stacked memory, hardware-level core scheduler assistant all say Zen 3 was going to be a big leap over Zen 2. Even with the revised roadmap dropping SMT4 and (possibly) the stacked memory, it's still more than just a respin on 7nm EUV/6nm.

Good news for everybody as there's no socket changes required for 1 more generation. Ryzen 4000 can stay on AM4 since it's not going DDR5, server markets benefit from having another year with their Rome boards as they can just drop in Milan CPUs, which also gives server software vendors time to get up to snuff with SMT4. Threadripper 4 will probably skip Milan unless Intel offers some proper HEDT competition this time next year/early 2021.

That gives everybody time to get proper ROI on their hardware before AMD have to change everything with DDR5, PCIe 5 and SMT4 with Zen 4 (which could now also be 2022 unless Intel wake the fudge up).

 

[4K8KW10]

This post should be here:

Looks like next chips might be a decent increase again and for games.

 

[Deleted member 209350]

Restructured CCD design, SMT4, shrunk I/O with stacked memory, hardware-level core scheduler assistant all say Zen 3 was going to be a big leap over Zen 2. Even with the revised roadmap dropping SMT4 and (possibly) the stacked memory, it's still more than just a respin on 7nm EUV/6nm.

Good news for everybody as there's no socket changes required for 1 more generation. Ryzen 4000 can stay on AM4 since it's not going DDR5, server markets benefit from having another year with their Rome boards as they can just drop in Milan CPUs, which also gives server software vendors time to get up to snuff with SMT4. Threadripper 4 will probably skip Milan unless Intel offers some proper HEDT competition this time next year/early 2021.

That gives everybody time to get proper ROI on their hardware before AMD have to change everything with DDR5, PCIe 5 and SMT4 with Zen 4 (which could now also be 2022 unless Intel wake the fudge up).

Except Intel


Jokes aside, been doing a bit of extra research on the 4000 series and just saw all of these things! The last info update I had on Zen3 was just before Zen2 was about to get released, and most people speculated it was just going to be a small increase in performance, but judging by these new leaks its going to be another big one.

I might bite the bullet and go for a 4000 series chip, and then a few years down the line go for an AM5 build. Don't fancy getting a completely new rig when mine is not even a year old atm

 

[StriderX]

What's up with SMT4 that it couldn't be implemented before now?