Speculation: Ryzen 3000 series

[amd6502]

Your analogy makes no sense and is a strawman. We aren't talking about Intel and their recent monopoly of workstation cpu's. Please answer why you would sell a next gen 8 core at the price of the previous gen 6 core? AMD aren't going to do that. They are looking to make money and have decent profit margins. Not give cpu's away.

I agree.

Besides, imho it's clear that 12nm/11nm is going to remain the staple and mainstream node for the next one or two years (if not more), and that most likely the 4c/8t APUs will be the bulk of sales for this category. So the ryzen 5 category will continue to contain quadcores for the next ~2+ years.

Ryzen 3 (and athlon) will continue to be dumping grounds for bottom die salvage, plus the addition of a small die big core to replace BR+Stoney; imho it's very doubtful any 7nm product will make it to that brand in the near future. (unless yields are so bad that there is significant number of 2c and 3c die salvage)

 

[PotatoWithEarsOnSide]

The reason that few/no 7nm dies will end up in Ryzen 3 is that the demand for them in Epyc 2, Threadripper, and R9(?)/7/5 will be huge. Even if yields are really good, there won't be enough to go around.
However, that is not to say that AMD will charge Intel prices on marginally superior (core for core) performance; a 16c might command a new price bracket, but expect the improved 8c to be at current (or lower) pricing, especially if there are 12c and 16c variants. AMD can still make very handsome profits from such 8c CPUs, whilst also practically obsoleting Intel's mid-tier offerings.

 

[AtenRa]

Your analogy makes no sense and is a strawman. We aren't talking about Intel and their recent monopoly of workstation cpu's. Please answer why you would sell a next gen 8 core at the price of the previous gen 6 core? AMD aren't going to do that. They are looking to make money and have decent profit margins. Not give cpu's away.

Because you will sell another product at the place of your last year 8-Core cpu. That is called progression, unlike what Intel did from 2011 to 2016.
Ohh and they will most definitely make more money, they could have once again a $499 Mainstream CPU in AM4 platform since the first Ryzen launch.

 

[tamz_msc]

Your analogy makes no sense and is a strawman. We aren't talking about Intel and their recent monopoly of workstation cpu's. Please answer why you would sell a next gen 8 core at the price of the previous gen 6 core? AMD aren't going to do that. They are looking to make money and have decent profit margins. Not give cpu's away.

Why is it a strawman? In March 2017 the Core i7 6900K was a $1000 CPU. Why didn't AMD price the Ryzen 7 1800X close to 1000$ if margins were their primary concern? Six months later when the i7 7820X was selling for $600 why was AMD giving away the 1700X for less than $350?

AMD sets those prices simply because they can. Their margins have never been a problem since the launch of Zen. They started with low-to-mid 30% margins at the beginning of last year and touched 40% in last quarter.

Their margins are fine.

 

[PotatoWithEarsOnSide]

Their goal will be to make as much profit as possible, and no doubt they've made the calculation that higher volume at reduced margins is going to achieve that for them. It has the added bonus of market penetration and increased market share.
AMD clearly view the benefits of getting people to buy into the platform as being better than a one-off profit surge down to increased margins on similar volume. They seem to be taking the view that 'as much profit as possible' is not a single year metric, unlike what greedy shareholders might view it as.

 

[moinmoin]

Besides, imho it's clear that 12nm/11nm is going to remain the staple and mainstream node for the next one or two years (if not more), and that most likely the 4c/8t APUs will be the bulk of sales for this category. So the ryzen 5 category will continue to contain quadcores for the next ~2+ years.

Ryzen 3 (and athlon) will continue to be dumping grounds for bottom die salvage, plus the addition of a small die big core to replace BR+Stoney; imho it's very doubtful any 7nm product will make it to that brand in the near future. (unless yields are so bad that there is significant number of 2c and 3c die salvage)

How is this clear? Most if not all people expect the 7nm Ryzen within 2019, and by the usual pattern a 7nm APU should arrive early 2020.

Or what "mainstream" are you talking about?

 

[exquisitechar]

https://www.semiwiki.com/forum/content/7923-iedm-2018-imec-interconnect-metals-beyond-copper.html
Contacts – Co filled contacts have been introduced at 10nm (similar to foundry 7nm) by Intel and 7nm by TSMC.

 

[Fanatical Meat]

https://www.semiwiki.com/forum/content/7923-iedm-2018-imec-interconnect-metals-beyond-copper.html
Contacts – Co filled contacts have been introduced at 10nm (similar to foundry 7nm) by Intel and 7nm by TSMC.

Good for them, what is the context of this regarding the thread?

 

[tamz_msc]

Good for them, what is the context of this regarding the thread?

Cobalt allows them to decrease resistance which helps them to attain higher clock speeds.

 

[exquisitechar]

Cobalt allows them to decrease resistance which helps them to attain higher clock speeds.

Exactly. Huge reduction in resistance.

 

[Zapetu]

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.

https://www.semiwiki.com/forum/content/7923-iedm-2018-imec-interconnect-metals-beyond-copper.html
Contacts – Co filled contacts have been introduced at 10nm (similar to foundry 7nm) by Intel and 7nm by TSMC.

TSMC 7nm using cobalt for contacts is exactly what raghu78 assumed two weeks ago. So that article is just confirming what Scotten Jones meant by "TSMC 7nm adds cobalt". To say the least, it looks very promising for Zen2 clock speeds.

 

[Zapetu]

AdoredTV posted a new video on what we should expect from CES 2019:

He still stands by his leak that Ryzen 3000 CPU's (desktop, 7nm, 6-16 cores, AM4) will be announced at CES 2019. The highest end 16 core SKU and all G variants with Navi graphics would be announced later. Previous leaks also suggest that Picasso (12nm, 4 cores, Vega w/ 11 CUs, "tweaked Raven Ridge") mobile parts would also be announced (edit: or released).

Jim (AdoredTV) also believes that AMD was planning to release Vega II (possibly Vega 20 (7nm) for destop) but has since cancelled that plan and focuses on Navi only. So there might be (three) Navi parts announced at CES 2019, remains to be seen. If Vega II actually materializes (for destop) then it would likely be a limited volume part (20k-60k).

It's only about a week before we hopefully know a lot more how AMD's 7nm consumer lineup will look like. Happy new year to everyone and soon we'll have more exciting information to base our future speculations on. Obviously what most of us are really looking for is to actually be able to get these parts on our hands but that will likely have to wait a little longer.

 

[Shivansps]

I havent seen the video yet, but im going to tell you this, IF AMD dosent release the 3000 series in the next 3 months im going to say the entire "leak" was BS. AMD dosent give products an official name, specs or prices until they are about to come out. This is the same reason of why all the APUs on that leak are BS.

 

[darkswordsman17]

That's more or less what I'd guess at. I was thinking they might only do the 8 core (single module) Ryzen parts for now, and then wait to launch the 12 and 16 core ones alongside the 500 series motherboards at CES (since early boards are likely to be more premium, but 400 series ones should be able to manage 16 cores that are within the Ryzen 2000 series power envelope with probably the highest clocked 16 core chips pushing beyond that and needing higher spec). Maybe even brand them as a version of Threadripper (maybe name the TR4 ones Threadripper XL or the Ryzen ones Ryzen TR). Or maybe even just wait til Intel announces their 10 core parts and/or other CPUs to steal some thunder (which maybe Intel is going to do that at CES).

I'm curious how long ago this Vega II stuff was, as I got the strong impression that was decided quite some time ago (before Vega 20 ever entered production, basically not long after Vega 64 came out and it was quite a bit of a dud in gaming, which led to them just focusing on enterprise focused stuff and not tweak the rest of the chip for graphics for Vega 20, and putting their graphics development focused into Navi). I thought they outright said quite some time ago though that Vega 20 was enterprise and Navi would be their consumer chip. Think it started with them no longer listing 7nm Vega consumer chips on their roadmap (and I think they might've also removed Navi 10x2 or whatever dual Navi thing they had as I believe they're just going to do a small - consumer - Navi 10, and then a larger - pro - Navi 20 similar to how they changed Vega. Which hopefully maybe we'll get 3 moving forward. A large-ish consumer chip (that has the max rendering/rasterization capability) for the $400-600 market, a smaller one for the mainstream segment ($100-300), and then the Enterprise one. And if they go with an I/O chip, that could give them more flexibility over memory for different products. As well as they could maybe do a specialized ray-tracing module (not sure how much it relies on the raster pipeline though, so not sure how feasible it would be to split them, but Nvidia seems to be using a pretty specialized block that isn't integrated right into the traditional graphics pipeline on RTX so that makes me think its possible). Which that might be how they segment the premium consumer stuff, as it just being the normal typical GPU paired with a ray tracing module and more memory/bandwidth. Which I wouldn't hate that if it meant that the mainstream chips got a very strong graphics focused chip. I personally wouldn't mind if they integrated display stuff into the motherboard (and the CPUs getting the video processing blocks, meaning its no longer on the GPU, but rather be part of say the I/O module, which I guess they could put that in the I/O module of the graphics card, but I'd like them to be able to move the display connectors so that the video card would just be able to vent as much of the heat as possible). They can include a card that has connectors (and make it so you can pick and choose what connectors you want and place them where you want so maybe you put them in the bottom slot of the board). On motherboards they're probably gonna head to being largely USB-C Thunderbolt, since that can takeover for all the other connectors (obviously be some transition period, or maybe they include adapters for HDMI/DP/older USB/etc).

Hopefully Navi brings some good improvements. Will be interested to see how it turns out, if they maybe pushed for it to be more graphics focused since they'll have stronger CPU in the consoles. Which by that I mean, it'd offer compute capability of One X/PS4 Pro level, but not a ton more, instead choosing to use transistor count to maximize the rasterization capability to try and offer native 4K at 60FPS (with some other ways of getting there, like lowered precision where they can get away with it). Which I wonder if they wouldn't develop a compute specialized module even, especially for something like AI or other specialized hardware. I know there's a decent amount of the modern graphics pipeline that is suited for some compute and I don't expect that to change (if for no other reason than compatibility, but moving forward they look to specialized co-processors more).

 

[DrMrLordX]

I havent seen the video yet, but im going to tell you this, IF AMD dosent release the 3000 series in the next 3 months im going to say the entire "leak" was BS. AMD dosent give products an official name, specs or prices until they are about to come out. This is the same reason of why all the APUs on that leak are BS.

Release as in announce them, or release as in sell them? Even optimists don't see Matisse hitting the shelves before April.

 

[amd6502]

we can probably count on 16c 7nm threatrippers being announced too.

 

[TheGiant]

I hope for 9900K level of performance at 4.7GHz as achievable. With 60% of power consumption.
My personal guess they don't hit 5 GHz. It takes more than manufacturing process to hit that frequency.
But anyway, I don't think we will see current AMD prices with their 7nm parts. Te strategy of low prices/high volume never worked well.

 

[Shivansps]

Release as in announce them, or release as in sell them? Even optimists don't see Matisse hitting the shelves before April.

I expect them to do the exact same thing they did with the first Ryzens. So at least we need to know the models names and configurations.

 

[DrMrLordX]

I expect them to do the exact same thing they did with the first Ryzens. So at least we need to know the models names and configurations.

So, announce. Got it.

 

[Spartak]

I just made my vote above to "something else":
8 core monolithic + optional 8 core chiplet (same chiplet as Epyc)
Only one IF connection on the Ryzen 3000 die instead of now 3.

edit: because it makes no sense to use two dies for a low budget CPU, and an 16 core monolithic seems too costly, this above allows chiplets for high end CPU and 1 die for the low end. With only 1 desktop die and reuse of the 8 core chiplet of epyc for >8 cores.
(or connect a GPU die for consoles)

This. They wont make two seperate integrated designs but combining an 8CCX with either an 8core chiplet or a GPU seems to me the most straightforward path.

So R3 and R5 will just be one 8CCX
The G versions will be 8CCX+GPU
The R7 and R12 will be 8CCX +8C chiplet

Funny how everyone took adoredTV's speculation for fact . Chiplet for consumer never made much sense to me, but i can see it for the higher end R7 and R9.

 

[chrisjames61]

But anyway, I don't think we will see current AMD prices with their 7nm parts. Te strategy of low prices/high volume never worked well.

I agree with you but many people here are saying that the 7nm parts core for core will be cheaper than the current generation which is utter stupidity.

 

[PotatoWithEarsOnSide]

Why is it?
Sure, R&D probably cost more, and the wafers cost marginally more, but then they'll get more dies per wafer, and they'll have higher sales volume over which that upfront R&D costs would be spread.
Nothing stupid here except your inability to comprehend the basic concepts of supply and demand.

@Spartak, I suspect that the major differentiator may be that we might not see many R3s early in the 7nm life cycle, so potentially the R3s could just be rebadged Zen+ chips. It's been stated quite often on these forums that the top and middle tier Zen 2 chips are likely to be in incredibly high demand, so short of a high defect rate no-one is really expecting chiplets in the bottom tier. It is a common belief that any 8/10/12c chips will be two chiplets with defective dies, and the fully functional dies ending up in Epyc 2 and TR.
We already know that AMD is willing to put a previous gen process on a current gen badge.
It's not about blind faith either. The leaks that AdoredTV provide are generally very well presented, including logical analysis. Of course, that's not to say that everything is anywhere near 100% accurate. It is a very compelling argument that the chiplet design allows for higher yields, and better harvesting of defective dies, though clearly, relying on defective dies for a large segment of your product range has some serious drawbacks.

Do I think that all Zen 2 products will be chiplet designs? Yes.
Do I think that all 3xxx Ryzen CPUs will be on 7nm Zen 2? No. I think there's a possibility that the R3s will still be Zen+, though 6c minimum.

 

[Spartak]

The line-up leaked on Reddit had Zen+ for R3 and part of the R5 but that doesnt make much sense and confuses the line-up if you work with two different architectures for the same generation. Looking at that line up the R5 3400X 4C/8T Zen+ would be $120 and the R5 would be 8C/16T Zen2 for just $50 more? Not only faster IPC, but much higher clocks and double the cores for just a bit more?

It looks like there were two line-ups mangled in that list.

It's unfortunate AMD chose to call the APU versions 'next gen' when they were current gen. I hope they correct that this time around but give it little chance.

FWIW I'm expecting the APU's to be 12nm Zen+ as expected and launch directly whereas the 'real Zen2' 3000 series will launch late may/early june around computex. The modular Zen2 APU's from 19Q3 would really confuse the Zen+ APU's, would be much better if those were still called 2000 series.

 

[Spartak]

It is a common belief that any 8/10/12c chips will be two chiplets with defective dies, and the fully functional dies ending up in Epyc 2 and TR.
We already know that AMD is willing to put a previous gen process on a current gen badge.
It's not about blind faith either. The leaks that AdoredTV provide are generally very well presented, including logical analysis. Of course, that's not to say that everything is anywhere near 100% accurate. It is a very compelling argument that the chiplet design allows for higher yields, and better harvesting of defective dies, though clearly, relying on defective dies for a large segment of your product range has some serious drawbacks.

The 'common believe' only gained traction with AdoredTV's video where it seemed to be part of the leak but was his own speculation. He put that chiplet theory in the bin quite adamantly now:

Adored source was very clear the consumer Zen2 would be 'all 7nm'. So an integrated 8CCX or 2x4CCX is all but a given. The question is how they add the second eight cores. A chiplet added to the CCX-module makes the most sense to me.

 

[mattiasnyc]

I don't understand how people can say that the current price strategy didn't work well. AMD is one of the few stocks that have ended up extremely positively this past year. I think it's up over 70% or so.

How is that not a tremendous success?

I also must have missed how people can be so relatively certain yields will be so poor the next Ryzen won't include 7nm parts. I don't really get that. Surely their strategy so far has been to reuse exactly in order to maximize yields. Not fully functioning dies end up in the lower end products. All TSMC has to do is churn out dies and AMD will package accordingly.

What am I missing above?