Speculation: Ryzen 4000 series/Zen 3

Page 41 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146
Vermeer/Genesis isn't in Family 17h, so it isn't Zen anymore.
Milan/Renoir however are Family 17h, so they are still Zen.

Automatically, that should tell you Vermeer and Genesis aren't Zen3.

Oh dear lord no, if anything it implies the opposite.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Automatically, that should tell you Vermeer and Genesis aren't Zen3.

"You keep using that word, I do not think it means what you think it means...."

Seriously though, you seem to be basing this assumption on information not on any roadmap, nor from what I can discern even a pre emptive Linux kernel patch.

You might have a little more luck hammering that square peg into the round hole if you can be a bit more specific about you sources of information - because all the information publicly disclosed thus far points to Zen3 Vermeer.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
You might have a little more luck hammering that square peg into the round hole if you can be a bit more specific about you sources of information - because all the information publicly disclosed thus far points to Zen3 Vermeer.
Zen3 is contained with Family 17h, it doesn't spill over into Family 19h. With that VMR-A0 is Family 19h 20h-2Fh and GN-A0 is Family 19h 00h-0Fh. (I still don't get why people think GN-A0 equals GPK-A0, which would use VMR-A0 binned dies).

The models don't launch chronologically, Renoir is Family 17h 60h-6Fh and Matisse is Family 17h 70h-7Fh. So, yes Vermeer/Genesis could launch before Genoa.
 
Last edited:

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Zen3 is contained with Family 17h, it doesn't spill over into Family 19h. With that VMR-A0 is Family 19h 20h-2Fh and GN-A0 is Family 19h 00h-0Fh. (I still don't get why people think GN-A0 equals GPK-A0, which would use VMR-A0 dies)
S.O.U.R.C.E.S.

They separate idle speculation and overly enthusiastic musings from fact.

I am not arguing part ID numbers or whatever those are, I am arguing over their source and the provenance of them in reference to Znver family members.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146
S.O.U.R.C.E.S.

They separate idle speculation and overly enthusiastic musings from fact.

I am not arguing part ID numbers or whatever those are, I am arguing over their source and the provenance of them in reference to Znver family members.

Not quite the same thing, but here's a similar list that lines up.



Milan and Genoa aren't listed currently in this list though. Whatever he's quoting is more up to date, and idk where it's from personally.
 
  • Like
Reactions: soresu

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Odd to have Genesis in there, I wonder if it refers to biblical Genesis or the prog rock/pop band (going with the artist vibe).

That Amur seems an odd one also, there is another Amur listed from way back, possibly an ARM part.

Reading further down that tweet, the same Komachi guy actually confirms that Vermeer is Zen3 when another guy mentions it:

1571665622614.png
 
Last edited:

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Odd to have Genesis in there, I wonder if it refers to biblical Genesis or the prog rock/pop band (going with the artist vibe).
Genesis Peak is near Castle Peak and Colfax, fyi.
That Amur seems an odd one also, there is another Amur listed from way back, possibly an ARM part.
Some of it is legacy info.
Zen2 Rome used to be Family 18h 10h-1Fh and Zen3 Milan used to be Family 18h 20h-2Fh and packed AVX512F, as well. This was before HARC/Hygon though so pretty old plans.

They probably don't realize Zen3 is specifically in Family 17h. If going based on the above, Milan will be models 40h-4Fh, because Rome is 30h-3Fh. However, there is 80h-9Fh and the only designs that went that high before is Family 10h A0h: Thuban and Family 10h 91h: Magny Cours
 
Last edited:

juergbi

Junior Member
Apr 27, 2019
12
14
41
Is it possible that this rumor is only about APUs? APUs currently lag behind by a Zen generation (or at least half a year). AMD might want to change that. They could skip Zen 3 for APUs and directly move from 7nm Zen 2 (Renoir) in 2020 to 5nm Zen 4 (Rembrandt?) in 2021, thus skipping 7nm+ and aligning both Zen 4 CPUs and APUs for the launch of the next desktop socket (AM5?).

Stuff to watch out for: RRD. Three new CCDs at TSMC. Of which one is for the new Renoir(7nm) and its successor Rembrandt(5nm). The D one is for a semi-custom project on 6nm(not AMD mainline, however it might get pushed to 5nm).

I've also found something interesting on LinkedIn:
...delivering several products 5/6/7 nm technology for AMD Renoir, Rembrandt, Durango CCD projects.

This would be in line with 'RRD'.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
My money is on AVX-512 implemented on 256 bit vector units. Still gives a nice boost in capability (better ISA for vectorizer), while not blowing up die area or power consumption.
 
  • Like
Reactions: Ajay

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
I've also found something interesting on LinkedIn:
...delivering several products 5/6/7 nm technology for AMD Renoir, Rembrandt, Durango CCD projects.
This would be in line with 'RRD'.
Durango could very likely be another Microsoft joint venture, it's also the codename for the XB1 dev kit.

Though Dali is still in stealth somewhere out there....
 
Last edited:

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Is it possible that this rumor is only about APUs? APUs currently lag behind by a Zen generation (or at least half a year). AMD might want to change that. They could skip Zen 3 for APUs and directly move from 7nm Zen 2 (Renoir) in 2020 to 5nm Zen 4 (Rembrandt?) in 2021, thus skipping 7nm+ and aligning both Zen 4 CPUs and APUs for the launch of the next desktop socket (AM5?).
Given that AMD is seemingly not even going with RDNA on Renoir, I'm apt to think that they want to milk each bit of IP as much as possible before they drop it.

If Zen3 is worth a major core number increase, it's likely not something they would want to abandon a chance for milking dry.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Is it possible that this rumor is only about APUs? APUs currently lag behind by a Zen generation (or at least half a year). AMD might want to change that.

I imagine they will at some point move the IGP to a chiplet which will fix the issue. But until then they will stick to the current strategy I guess.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146
I imagine they will at some point move the IGP to a chiplet which will fix the issue. But until then they will stick to the current strategy I guess.

Probably not for a while, chiplet power draw is far too high.

Is it possible that this rumor is only about APUs? APUs currently lag behind by a Zen generation (or at least half a year). AMD might want to change that. They could skip Zen 3 for APUs and directly move from 7nm Zen 2 (Renoir) in 2020 to 5nm Zen 4 (Rembrandt?) in 2021, thus skipping 7nm+ and aligning both Zen 4 CPUs and APUs for the launch of the next desktop socket (AM5?).



I've also found something interesting on LinkedIn:

Found the LinkedIn page as well,seems pretty interesting. Just a few extra points. Durango is the name of the XBone. Also, the mention of both 7nm and 6nm leads me to believe that Renoir might have been ported over to 6nm?
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,575
146
I'm an idiot. The D probably doesn't stand for Durango, it stands for Dali - all three of which are APUs, two releasing in the next year, and I'm guessing Rembrandt is releasing in 2021.

Hmm... looking at it again, maybe not. But that makes the use of the same codename twice by AMD seem quite suspect.
 
Last edited:

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
Oddly Piledriver launched with the Trinity APU before the desktop 'enthusiast' part.

Yes, accompanied by this great or very attractive promotional video.No doubt, this is the best AMD-s promotional video or trailer.Green frog was not very good(CPU), but hey at least the video made a great impression.

 
  • Like
Reactions: OTG

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Memory latency testing is hard, all the popular tools out there are just wrong. Core-to-core roundtrip is basically 2x L3 latency, which makes sense as that's essentially how it works.
So what is supposed to be the normal L3 latency of a 3900x? According to the one review I found that actually reported L3 latency, they claimed 9.6 ns. That's low enough that the reported core-to-core scores I got out of Sandra are still significantly longer than what should be reported assuming the review was even remotely correct (should have been around 19ns, not 26ns).

Yeah, Andrie makes a good point. I've been puzzling over this for a few days. It isn't hard to write a program that will keep loading from memory till the caches spill over into L3 and them measure the bandwidth by reading back all the L3 data, or a variety of other cache level performance metrics. I haven't figured out a way to measure latency - but, I don't know enough about low level x86 instructions that may help measure this consistently - so that failing is mine. In the past, I would have looked into latency issues by using a hardware emulator (I worked on embedded firmware development). To find the the bandwidth between a given core and a given L3 slice, you have to understand Zen's L3 cache eviction policy, which I don't. If looncraz and theStilt were still here, they would probably be able to offer some insight.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
Are we sure it refers to meditation and not the philosophy?

If philosophy, Tao would be an awesome name for a successor.
Zen being the American cores w/ Family 17h.
Dhyana being the 14nm Chinese core w/ Family 18h.
Moksha might be the 7nm Chinese core w/ Family 18h????
-%$*---*$%- being the [High-Mobility] 5nm/3nm/etc American cores w/ Family 19h. <- Vermeer and Genesis Peak based on the leaks will be based on this core. While, Milan and Renoir will be based on Zen3(Family 17h). Renoir might not have something Milan has, but what it does have is Zen3-pointed.

Ryzen 4000 will either be MTS-C0(y0)&CPK-C0(y0) w/ Starship2 dies(Milan-derived/Zen3-derived)
or it will be VMR-A0(x0)&GPK-A0(x0) w/ Genoa-derived CCDs(Formally, Zen4-derived)
GPU rebranding-style in CPUs would be nice.
 
Last edited:

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Zen being the American cores w/ Family 17h.
Dhyana being the 14nm Chinese core w/ Family 18h.
Moksha might be the 7nm Chinese core w/ Family 18h????
-%$*---*$%- being the [High-Mobility] 5nm/3nm/etc American cores w/ Family 19h. <- Vermeer and Genesis Peak based on the leaks will be based on this core. While, Milan and Renoir will be based on Zen3(Family 17h). Renoir might not have something Milan has, but what it does have is Zen3-pointed.

Ryzen 4000 will either be MTS-C0(y0)&CPK-C0(y0) w/ Starship2 dies(Milan-derived/Zen3-derived)
or it will be VMR-A0(x0)&GPK-A0(x0) w/ Genoa-derived CCDs(Formally, Zen4-derived)
GPU rebranding-style in CPUs would be nice.
Oooh, I was just reading the Wiki page on Moksha and found a reference to the ultimate follow up name for Zen.

Nirvana.