Speculation: Ryzen 4000 series/Zen 3

[DrMrLordX]

Didn't they say they started dual sourcing 7nm products and at some point dropped the ones on GF for obvious reasons?

Pretty sure they did. Rome and Vega20 were confirmed TSMC 7nm, so the expectation was that AMD was going to use GF for everything else. Nothing else was officially confirmed of course. And as time wore on obviously everything was moved to TSMC.

 

[Gideon]

Pretty sure they did. Rome and Vega20 were confirmed TSMC 7nm, so the expectation was that AMD was going to use GF for everything else. Nothing else was officially confirmed of course. And as time wore on obviously everything was moved to TSMC.

I remember that in the Anandtech interview , where AMD's commented on the GloFo 7nm cancellation, they explicitly said they only had one product there, that was supposed to be released in the end of this year.

Considering that Rome and Matisse use the exact same chiplets, it would sees really odd to abandon all the possible savings by fabbing them in multiple places. Let's also not forget that Matisse wasn't supposed to come out at the year end.

My bet is, that the planned product on GloFo was the 7nm APU

 

[scannall]

My bet is, that the planned product on GloFo was the 7nm APU

That would be my bet as well, considering the APU's are lagging the rest of the stack by about 1 product cycle at the moment.

 

[DrMrLordX]

@Gideon

That's a reasonable conclusion; however, Renoir was not scheduled to come out in 2019 at all. So I'm not 100% sold on the idea that the "one product" was Renoir. It might have been Navi for all we know. For awhile, Navi was on roadmaps for Q4 2019 until AMD pulled it forward.

 

[Vattila]

My bet is, that the planned product on GloFo was the 7nm APU

I think that's probably a good guess. My understanding is that AMD worked closely with both TSMC and GF on the 7nm "big lift" (as well as with EDA tool vendors). TSMC beat GF on time-to-market (and perhaps on process specifications as well), so AMD committed to TSMC. Perhaps they initially committed only the 7nm CPU chiplet (does it have a code name by itself?), leaving the door open for GF to make "Renoir", the 7nm APU, if they could get ready on time. They probably failed to meet milestones on the latter, at which point AMD moved it to TSMC. Faced with no immediate 7nm product to make, GF threw the towel into the ring on 7nm altogether. Reportedly, AMD urged them to stay the course, perhaps dangling opportunities for Milan and beyond, but maybe they did not see any realistic prospect of catching up to TSMC.

 

[Tup3x]

Well I seemingly wrongly assumed you were one of those "ARM will take over the world" guys. About 10 years ago even Anand was talking about x86 in iphones. Clearly that never happened. Meanwhile we have had people talking about ARM taking over x86, and that has yet to happen. There was a similar RISC vs CISC debate in the 90's, and we all know how that turned out.

If after 10+ years we haven't seen x86 on mobile or ARM on PC's, I have to think they are both the best at what they do in those form factors. Maybe we will see an ARM uprising within five years, maybe not. I would bet against it though. Sorry if I misunderstood you as an "ARM everywhere" type.

Except we are definitely starting to see ARM in PCs. Samsung just announced their Snapdragon 8cx based Windows 10 laptop. Now and those (ARM laptops) may or may not take off but it's a start. That being said there has been x86 phone but anymore.

 

[NTMBK]

Except we are definitely starting to see ARM in PCs. Samsung just announced their Snapdragon 8cx based Windows 10 laptop. Now and those (ARM laptops) may or may not take off but it's a start. That being said there has been x86 phone but anymore.

We've seen ARM in PCs before. Remember the Windows 8 machines with Tegra CPUs? It flopped, badly.

 

[Gideon]

We've seen ARM in PCs before. Remember the Windows 8 machines with Tegra CPUs? It flopped, badly.

This is a fair point, but it's certainly getting more traction now (from the devs). Just look at the benchmarks here:
https://www.windowscentral.com/snapdragon-8cx-benchmarks

The key differences are:

• The hardware is a lot more competitive now, it supports emulated x86 apps well enough that the Average Joe would be satisfied
• Devs are finally making Arm-native implementations of relevant software: Chrome, Firefox and most importantly Electron. As more and more desktop apps are now really just Electron wrappers it means that Spotify, Slack, Discord, Skype will all get native implementations.
• The battery life truly is almost twice as good

You certainly won't do any gaming or professional IT/graphics work on it. But for the stuff usually ran on such laptops will run just fine and have a lot more battery life for the price. The same wasn't really true for the Tegra parts IMO.

 

[Gideon]

On another note. Both Toms Hardware and Phoronix keep claiming (based on AMD driver commits) that Renoir is Vega not Navi, but it seems at least 7nm Vega. A bit disappointing but not too bad, when it's the Radeon VII iteration of Vega and has Zen 2 cores. It's just that it could have been considerably better for gaming if it were Navi instead.

EDIT: spelling

 

[moinmoin]

On another note. Both Toms Hardware and Phoronix keep claiming (based on AMD driver commits) that Renoir is Vega not Navi, but it seems at least 7nm Vega. A bit disappointing but not too bad, when it's the Radeon VII iteration of Vega and has Zen 2 cores. It's just that it could have been considerably better for gaming if it were Navi instead.

EDIT: spelling

This news made the rounds. My guess is that, true to the APUs being a budget line for AMD so far, Renoir is primarily a 7nm shrink with some Zen 2 improvements. If nothing else it possibly being around 100mm² should offer nice battery improvements.

 

[Ajay]

This news made the rounds. My guess is that, true to the APUs being a budget line for AMD so far, Renoir is primarily a 7nm shrink with some Zen 2 improvements. If nothing else it possibly being around 100mm² should offer nice battery improvements.

I’d be really surprised if the CCX complex wasn’t updated to Zen2. The higher throughput per clock would really give AMD a chance to compete with Intel in laptops.

 

[naukkis]

I’d be really surprised if the CCX complex wasn’t updated to Zen2. The higher throughput per clock would really give AMD a chance to compete with Intel in laptops.

Porting designs to 7nm is expensive. Why would AMD spend money porting Zen1 to 7nm when they can just use already ported Zen2? Vega is also already ported to 7nm - Radeon VII so they just need to put those two already made parts to same die.

 

[moinmoin]

I think the question is rather which parts of Zen 2 will they keep. L3$ likely will be cut again. Will they keep the Zen 2 core untouched otherwise?

 

[Ajay]

I think the question is rather which parts of Zen 2 will they keep. L3$ likely will be cut again. Will they keep the Zen 2 core untouched otherwise?

Yes I think they will use the Zen2 cores. Cut down L3$ makes sense.
I think @naukkis misunderstood my post.

 

[DrMrLordX]

On another note. Both Toms Hardware and Phoronix keep claiming (based on AMD driver commits) that Renoir is Vega not Navi, but it seems at least 7nm Vega. A bit disappointing but not too bad, when it's the Radeon VII iteration of Vega and has Zen 2 cores. It's just that it could have been considerably better for gaming if it were Navi instead.

Good news for Radeon VII owners (sort of) wrt longevity of the platform's driver support. Bad news for actual Renoir buyers.

 

[Yotsugi]

Bad news for actual Renoir buyers

Why?
It's Zen2 with it's borderline stupid efficiency at optimal clock ranges on mobile.

 

[moinmoin]

Bad news for actual Renoir buyers.

Regarding the graphics block: depends, it looks like the current Navi is a precursor to what's actually getting used in the upcoming consoles, so an APU using the latter version once available wouldn't be the worst of all worlds.

 

[burninatortech4]

I don't see a problem with a monolithic Zen2 + Vega 2 APU. If that's what indeed is happening. Vega is extremely efficient when the clocks aren't pushed out of the optimal range and has all the modern decode/encode needed (except Playready 3.0? not sure).

 

[krumme]

In a bandwidth limited scenario and power limited scenario navi is extremely more efficient than vega.

Due to the wsa amd needs to continue selling 12nm apu so what they need is something to complement it on the mid and high end. Something that can maintain their lead on gpu perf. Navi can do that. Vega wont. Zen plus api is fine as is and good at offloading 12nm capacity and fulfill the wsa.

From a brand and broad portfolio perspective zen2vega apu is just bad and another defensive old style amd thinking decision. I hope they waited the few extra months to get it right first time.

 

[Tuna-Fish]

Due to the wsa amd needs to continue selling 12nm apu

I don't think they do. Just looking at epyc projections and ryzen sales, the io-dies should be enough.

 

[naukkis]

I don't think they do. Just looking at epyc projections and ryzen sales, the io-dies should be enough.

There's plenty of market for 14nm Raven2 and 12nm Picasso. 7nm Renoir would be step above, probably 8c/16t. Renoir also might be the best gaming cpu from AMD, much better memory latencies than Matisse. Vega is fine, 7nm memory controller might allow very high memory clocks so it might be quite a bit faster than 14nm versions.

 

[DrMrLordX]

Why?
It's Zen2 with it's borderline stupid efficiency at optimal clock ranges on mobile.

It's Vega instead of Navi. Navi is (overall) better.

an APU using the latter version once available wouldn't be the worst of all worlds.

Yes, but that won't be Renoir.

 

[Yotsugi]

It's Vega instead of Navi. Navi is (overall) better.

Who cares about iGPU.

 

[maddie]

Who cares about iGPU.

If not, why APU? If yes, why Vega?

 

[Richie Rich]

1) AMD is developing APUs for Sony and MS consoles based on Zen2 + custom RDNA 2nd gen. And all these to be launched next year. Why would AMD waste workforce to develop something very different?

2) They could do small trims to console chip: to use DDR4 mem ctrl instead GDDR6, cut off half of L3 cache, make proportional vanilla 2ndRDNA to mem bandwidth... and voilá Renoir.

3) I'd like to see some basic GPU integrated in IO chip. Something like AMD Phenom northbridge 780G was. For the 2D desktop and office work would be OK and it would have few transistors on 12nm.