Speculation: Ryzen 4000 series/Zen 3

[A///]

Do you have a link to this report?

Run it through Google translate.

ソニー、PS5の生産台数を400万台削減、チップ生産に苦慮－関係者

ソニーは、年内に発売する次世代ゲーム機「プレイステーション（ＰＳ）５」の今期（2021年３月期）の生産台数を400万台下方修正し、約1100万台と想定していることが明らかになった。ＰＳ５の心臓部分に該当する専用設計の集積回路「システム・オン・チップ（ＳＯＣ）」の生産歩留まりが安定しないことなどが影響している。

www.bloomberg.co.jp

Microsoft isn't facing yield issues. They're disabling CUs on theirs. TSMC has incredibly high yields compared to anyone else. Microsoft has their RDNA2 running at 1.5 Ghz for the S and 1.8 Ghz for the X. Sony has their RDNA2 running at 2.2 Ghz, with a variable rate Zen2 processor at 3.5 Ghz compared to 3.6 Ghz for the S and 3.8 Ghz for the X.

 

[DiogoDX]

For me is a indication that sony bumped the gpu clocks after de design was completed. So basically is a overclocked gpu to close the gap with microsoft decreasing the yields.

Old rumors was around 9tflops (36Cu at 2ghz).

 

[leoneazzurro]

Well the presentation about the PS5 seems to show that PS5 CUs are different from desktop RDNA2 CUs (which were indicated as "larger" or "beefier"). So one should think why desktop RDNA2 Cus are bigger, a possibility is that they have more resources avaiable and/or they are designed for higher clocks and for this reason they need additional circuitry...
But this is the thread about Zen3, so this is completely OT.

 

[Hitman928]

Run it through Google translate.

ソニー、PS5の生産台数を400万台削減、チップ生産に苦慮－関係者

ソニーは、年内に発売する次世代ゲーム機「プレイステーション（ＰＳ）５」の今期（2021年３月期）の生産台数を400万台下方修正し、約1100万台と想定していることが明らかになった。ＰＳ５の心臓部分に該当する専用設計の集積回路「システム・オン・チップ（ＳＯＣ）」の生産歩留まりが安定しないことなどが影響している。

www.bloomberg.co.jp

Microsoft isn't facing yield issues. They're disabling CUs on theirs. TSMC has incredibly high yields compared to anyone else. Microsoft has their RDNA2 running at 1.5 Ghz for the S and 1.8 Ghz for the X. Sony has their RDNA2 running at 2.2 Ghz, with a variable rate Zen2 processor at 3.5 Ghz compared to 3.6 Ghz for the S and 3.8 Ghz for the X.

Thanks.

 

[A///]

For me is a indication that sony bumped the gpu clocks after de design was completed. So basically is a overclocked gpu to close the gap with microsoft decreasing the yields.

Old rumors was around 9tflops (36Cu at 2ghz).

Yes. Apparently the author of the Japanese article is a known fibber. However, the bump in clocks is fairly high and it isn't completely unlike Sony to pull something stupid. Yield issues can mean just about anything. Sony pushing clocks sky high while looking for a set range in energy use can easily screw up their yields. Sony. Always going against the grain. Just because.

 

[obidamnkenobi]

I've been out of the loop a while. I was looking at a R7 3700x, but will now wait for 4000. How does the tiers usually launch? How long after until the R7, lower range models come out?

 

[bauerbrazil2014]

So Apple is going to launch a 5nm processor next month close to Zen3, maybe NostaSeronx is right about Zen3 being 5nm.

 

[Hitman928]

So Apple is going to launch a 5nm processor next month close to Zen3, maybe NostaSeronx is right about Zen3 being 5nm.

Nope.

 

[NostaSeronx]

Nope.

There is only two credible rumors from Asia.
- Zen3 ES is N7P, thus Zen3 MP will be N7P. (Matisse 3, basically)
- Zen3 was on N7+, but has long been shifted to N5. (Sept-Oct launch, with general availability end of Q4 to 1Q21)

Only EUV-node that AMD has processor blocks(FE, LSU, Core, FPU, Bus, L2, L3) + IO interface is on 5nm.

Zen3 on N7P is pretty much in line with a ~1:1 refresh of Matisse/Rome.
Zen3 on N5 is also pretty much in line with a new CPU architecture.

00h-2Fh basically being Zen2 goes well with N7P.
30h and greater being true Zen3 goes well with N5.

Zen3/Zen4 architects are 5nm only though. ¯\_(ツ)_/¯

Lets keep the doom and gloom up: 7nm designs in 2018. 5nm designs in 2022. For AMD.
While for Apple w/ more complex SoCs; 7nm in 2018, 5nm in 2020, 3nm in 2022.

 

[Hitman928]

There is only two credible rumors from Asia.
- Zen3 ES is N7P, thus Zen3 MP will be N7P. (Matisse 3, basically)
- Zen3 was on N7+, but has long been shifted to N5. (Sept-Oct launch, with general availability end of Q4 to 1Q21)

Only EUV-node that AMD has processor blocks(FE, LSU, Core, FPU, Bus, L2, L3) + IO interface is on 5nm.

Zen3 on N7P is pretty much in line with a ~1:1 refresh of Matisse/Rome.
Zen3 on N5 is also pretty much in line with a new CPU architecture.

00h-2Fh basically being Zen2 goes well with N7P.
30h and greater being true Zen3 goes well with N5.

Zen3/Zen4 architects are 5nm only though. ¯\_(ツ)_/¯

Lets keep the doom and gloom up: 7nm designs in 2018. 5nm designs in 2022. For AMD.
While for Apple w/ more complex SoCs; 7nm in 2018, 5nm in 2020, 3nm in 2022.

I'll take AMD's word for it.

 

[A///]

I've been out of the loop a while. I was looking at a R7 3700x, but will now wait for 4000. How does the tiers usually launch? How long after until the R7, lower range models come out?

Flagships first, then later models. Except with Zen 2 they launched the entire lineup on the same day. They also launched the video cards on the same day. I spoke to some well known reviewers about it in private. The general consensus was they were all pissed TF off with AMD for doing that. A lot of testing with short notice.

 

[DrMrLordX]

I've been out of the loop a while. I was looking at a R7 3700x, but will now wait for 4000. How does the tiers usually launch? How long after until the R7, lower range models come out?

Rumour has it that the 12c and 16c parts will launch in October. 6c, 8c, and maybe even 10c parts will show up in Nov/Dec. Probably.

 

[A///]

Rumour has it that the 12c and 16c parts will launch in October. 6c, 8c, and maybe even 10c parts will show up in Nov/Dec. Probably.

Oh nice dude. Any idea when TR may show up? March-May?

 

[obidamnkenobi]

Rumour has it that the 12c and 16c parts will launch in October. 6c, 8c, and maybe even 10c parts will show up in Nov/Dec. Probably.

Do we have any idea which parts/tiers are which core counts? Is R7 still 8c? Or 10-12?

 

[Kenmitch]

Sony production rumor was a fake

Sony denies making millions of fewer PS5 consoles than planned

Sony is denying reports that it is making millions fewer PlayStation 5 gaming consoles than initially expected.

www.foxbusiness.com

Sony denies report it’s making millions fewer PS5 consoles than expected

Will there be enough PS5 stock at launch?

www.theverge.com

 

[A///]

@Kenmitch Thanks for that. I posted it in the RDNA thread but couldn't remember where else I'd posted the link. Someone drew up the older articles with the numbers fluxing, and it seems as if it's the same Bloomberg journalists. I wonder if there's some pump and dump going on. I'll let someone forward their suspicions after January to the SEC.

 

[DrMrLordX]

Do we have any idea which parts/tiers are which core counts? Is R7 still 8c? Or 10-12?

No idea, though I suspect R7 will encompass 8c if core counts stay the same as Zen2 (which they probably will).

Oh nice dude. Any idea when TR may show up? March-May?

God only knows. AMD is always funny about HEDT. It took 7 months for the 3990X to hit the market after the launch of Matisse, so earliest I would expect it would be May 2021.

 

[trexfromouterspace]

Do we have any idea which parts/tiers are which core counts? Is R7 still 8c? Or 10-12?

Realistically speaking, the core counts will probably be more or less be the same across the stack. A 10-core could slot into where the 3800X was if yields/binning/volume alloow. So Ryzen 7 3700X 8-core (probably the best pure gaming chip with unified L3) and Ryzen 7 3800X as a 10-core part. Would make that sku stand out more.

 

[A///]

God only knows. AMD is always funny about HEDT. It took 7 months for the 3990X to hit the market after the launch of Matisse, so earliest I would expect it would be May 2021.

Wasn't that delayed for some reason? I don't think I'll be getting a 4990X. TR is a fantasy for me. I don't think I'd be able to use it for what it is. I'd be paying a lot of money I wouldn't be grinding down with my workload.

A lot of my build is fluid. I don't have a clue what AMD prices will be like for Zen 3. They may be significantly higher. As much as I'd love to spend $5,000 on an ultimate build and do hard tubing, I need to ground myself with what I know I'll use and make use of.

Thanks though. I must have missed the alert for this post earlier today.

Realistically speaking, the core counts will probably be more or less be the same across the stack. A 10-core could slot into where the 3800X was if yields/binning/volume alloow. So Ryzen 7 3700X 8-core (probably the best pure gaming chip with unified L3) and Ryzen 7 3800X as a 10-core part. Would make that sku stand out more.

Some price shifting at the low end methinks.

 

[DooKey]

So Apple is going to launch a 5nm processor next month close to Zen3, maybe NostaSeronx is right about Zen3 being 5nm.

LOL, NO, just no.

 

[dnavas]

LOL, NO, just no.

So you're saying there's a chance? I mean, one of the hot rumors is they're renaming to 5xxx. Clearly this means they've hit 5Ghz on 5nm and throwing in ddr5 and pcie5 and avx512. Also, SMT5.

j/k [Obviously it just means marketing has lost their collective minds.... snafu]

Maybe we need a new rumor? Launch is delayed two days until 10/10 at 10:10.

I'm actually wondering whether it's worth building a new TR rig on Zen3 -- I can live without new memory, but pcie5/cxl seems like a shift worth getting ahead of?

 

[NostaSeronx]

Zen3 N5 = Tapeout in Q3 2019(Same period as A14-ES tapeout) => Potentially, 6T-57Cx/38Mx
Zen4 N5P = Tapeout in Q3 2020 => Potentially, 6T-51Cx/34Mx

Zen3 and Zen4 having DDR5 or PCIe 5.0 is on the IOD, not the CCDs. New IOD can lead to a new model series though.

Native 512-bit is unlikely, as I have been constantly been bombarded that they are returning to 128-bit native for F/V concerns. There is also a shift from the split Integer-FPU after retire to a fully unified execution pipe.

Zen3 on 5nm is the full new architecture, while Zen3 on 7nm is basically just Zen2 w/ extras.
Zen3 on 5nm is in-line with the "next-gen Zen" architecture(which through investigative is actual Zen3, aka first meditative post-Zen core) that popped on July 2019+. With 5nm being there since 2017~2018 timeframe on timesaved archives.

Zen/Zen2 => Derivative from Apple's A7
Zen3/Zen4 => Derivative from a post-Vulc v8.3-A+SVE architecture from Broadcom that had a target of beating Intel.
True Zen3 is thus, faster than any Intel arch at Games, Office, Web, Server, Cloud, HPC.

 

[A///]

I'm actually wondering whether it's worth building a new TR rig on Zen3 -- I can live without new memory, but pcie5/cxl seems like a shift worth getting ahead of?

High cost, buggy new platform with new never before seen tech, high price of parts especially DDR5. I'd probably look into Zen3 Threadripper 4000 and see if it'll work for you. I'd skip TR5 and go for TR6. They'll probably sport more cores and much better performance with a mature DDR5 by then.

Or YOLO it and buy a TR system for Zen 3, 4 and 5.

 

[Hans Gruber]

High cost, buggy new platform with new never before seen tech, high price of parts especially DDR5. I'd probably look into Zen3 Threadripper 4000 and see if it'll work for you. I'd skip TR5 and go for TR6. They'll probably sport more cores and much better performance with a mature DDR5 by then.

Or YOLO it and buy a TR system for Zen 3, 4 and 5.

Remember when a new memory standard is introduced. That means the previous generation (DDR4) will be faster out of the gate than DDR5. The new memory standard gets faster with time.

 

[dnavas]

High cost, buggy new platform with new never before seen tech, high price of parts especially DDR5.

Yeah, high price of DDR5 is specifically why I'm a huge meh on that, but buggy new platform sounds like just enough chaos to make it interesting. That said, there's going to be a lot less opportunity-filled chaos with nvidia buying out every other industry out there :|

And I know better than this, but what the heck -- if AMD backs out of vector processing, it would be a colossally stupid thing to do. It's not just multimedia, it's straight-on data stream handling. Consider FD.io/VPP as a random example. Even with CXL, you're not going to context switch between a stream processor and your CPU. Memory coherency is one thing, but that's a whole nother ball of wax. Not that the (apparent) performance of FD.io on TR is anything to celebrate, but do you really think AMD would go super-wide on core-count, while narrowing the core?
[wait, why does that sound familiar?]