Question AMD's next-gen Ryzen 6000 APU: Zen 3+ CPU, RDNA 2 GPU expected in 2022 - TweakTown

[JPB]

AMD's next-gen Ryzen 6000 APU: Zen 3+ CPU, RDNA 2 GPU expected in 2022

AMD's next-gen 'Rembrandt' APU packs DDR5, USB 4.0, PCIe 4.0, AM5 socket, Zen 3+ CPU cores, RDNA 2 GPU core

AMD has a gigantic pipeline of products coming out over the coming years, but there is a metric boat load of new information that has just leaked through Expreview.

The new Ryzen 6000 series APUs are known as "Rembrandt" and is a massive upgrade over current APUs, where it will pack the new Zen 3+ CPU architecture, and upcoming RDNA 2 GPU architecture. We're expecting AMD to make the new Rembrandt APU on TSMC's new 6nm node (which is an optimized version of TSMC's N7).

AMD's next-gen Zen 3+ architecture will have a bunch of improvements over the current Zen 2 chips that are here, but it is the GPU side that is interesting. AMD will replace Vega GPU cores with RDNA 2 GPU cores, which will surely provide a gigantic leap in performance -- especially considering this is an APU.

This means we should expect the Rembrandt APU to have ray tracing technology, meaning we could have near next-gen console levels of performance which is pretty damn amazing. By then (2022) the Zen 3+ CPU architecture and RDNA 2 GPU architecture could leap frog what is built into the consoles.

Not just that, but AMD is expected to have Rembrandt APUs support the AM5 socket (this is new) which will bring DDR5 support, LPDDR5 support, USB 4.0 support, and of course PCIe 5.0.

AMD will launch its new Rembrandt APUs in 2022, and I'm sure we'll have more details between now and then.

 

[NTMBK]

RDNA2 in 2022? That's disappointing. I was hoping that they might catch up to the dGPU tech, and have RDNA3.

 

[Gideon]

It's funny to see this rumor circulated at about the same time as the 5nm+ Ryzen rumor Someone is surely spreading total BS ...

To my experience unfortunately it's usually the "too good to be true" rumor that is false and that would be the 5nm+ Ryzen in this case, as this rumor at least matches with AMDs usual "slightly slower than people hope" product cadence.

Remember Picasso for instance? There were leaks of that CPU 6+ month before release, when everybody was already eagerly awating for zen 2. As it was only a super-minor change from Raven Ridge, nobody thought it would take a year to release (after Raven), yet it did.

 

[uzzi38]

AMD's next-gen Ryzen 6000 APU: Zen 3+ CPU, RDNA 2 GPU expected in 2022

AMD's next-gen 'Rembrandt' APU packs DDR5, USB 4.0, PCIe 4.0, AM5 socket, Zen 3+ CPU cores, RDNA 2 GPU core

AMD has a gigantic pipeline of products coming out over the coming years, but there is a metric boat load of new information that has just leaked through Expreview.

The new Ryzen 6000 series APUs are known as "Rembrandt" and is a massive upgrade over current APUs, where it will pack the new Zen 3+ CPU architecture, and upcoming RDNA 2 GPU architecture. We're expecting AMD to make the new Rembrandt APU on TSMC's new 6nm node (which is an optimized version of TSMC's N7).

AMD's next-gen Zen 3+ architecture will have a bunch of improvements over the current Zen 2 chips that are here, but it is the GPU side that is interesting. AMD will replace Vega GPU cores with RDNA 2 GPU cores, which will surely provide a gigantic leap in performance -- especially considering this is an APU.

This means we should expect the Rembrandt APU to have ray tracing technology, meaning we could have near next-gen console levels of performance which is pretty damn amazing. By then (2022) the Zen 3+ CPU architecture and RDNA 2 GPU architecture could leap frog what is built into the consoles.

Not just that, but AMD is expected to have Rembrandt APUs support the AM5 socket (this is new) which will bring DDR5 support, LPDDR5 support, USB 4.0 support, and of course PCIe 5.0.

AMD will launch its new Rembrandt APUs in 2022, and I'm sure we'll have more details between now and then.

With the news about N5P ramping up in Q4 2020, I can't help but believe this rumour is BS.

 

[chrisjames61]

Gotta call bs on this "leak".

 

[turtile]

RDNA2 in 2022? That's disappointing. I was hoping that they might catch up to the dGPU tech, and have RDNA3.

The rumor makes no sense. RDNA2 will be sold to the public in the consoles this year. And if AMD's past strategy remains true for the future, CDNA2 will be the first on 5nm enhanced followed by RDNA3 and Zen 4.

 

[Gideon]

The rumor makes no sense. RDNA2 will be sold to the public in the consoles this year. And if AMD's past strategy remains true for the future, CDNA2 will be the first on 5nm enhanced followed by RDNA3 and Zen 4.

Yeah still usingVega on Cezanne in 2021 would be a huge blunder by AMD. Even if Tiger Lake (and it's successor) wont murder it in performance, they would in features as Vega would be 4 years old tech.

Let's recap:
* Desktop Vega was released in August 2017 (being delayed from may)
* Raven Ridge mobile zen1 APU using it was released October 2017 mere 6 months after the initially planned release date

* RDNA1 was released June 2019
* Renoir was released Q1 2020 and couldn't really use it as it was on a different node (N7 vs N7P), unlike Raven Ridge
* RDNA2 Big Navi is coming "in a couple of months" according to Lisa's last interview, the design should be fixed for a while already (ramping up console chips)
* Cezanne is coming in H1 2021 and is also at least on N7P so there should be no problem using at least RDNA1 (worst case)

 

[DrMrLordX]

they would in features as Vega would be 4 years old tech.

Not if they're updating it independently.

 

[DisEnchantment]

I think embedded (V1000/R1000 series, 2C/4T Vega 3 types) would still use Vega also in the future. Vega CUs are much smaller( ~60% the size of RDNA1. RDNA2 CUs will be bigger, not to mention the ACE, HWS, L1, L2, RB etc )
Ideal for this market. (1 RDNA2 CU vs 3 Vega CUs)

 

[LightningZ71]

As we have seen in Renoir, VEGA CUs are compact and still competitive IN THIS MARKET (embedded iGPUs). Unless there is some amazing, paradigm shattering memory compression breakthrough in the very near future, there won't be any reason to incorporate ray tracing in an iGPU. I submit to you the poor ray tracing performance of the 2060 from Nvidia. It is a large chip with a ton of bandwidth and, in anything but trivial instances, it's ray tracing capabilities are largely just a tease. Yes, it's miles better than any software solution, but it doesn't make it performant enough to be very relevant.

AMD APU s ar laptop chips first and foremost. They need to be cost effective, compact, low power and still have relevant performance in CPU tasks. Trivial amounts of hardware raytracing aren't going to help anywhere. What WILL help is larger caches, wider cores, higher clocks, better power management. What will also help is greater memory bandwidth. We're getting that with DDR5 and lpddr4x. But it's still not the truck loads that dGPUs get.

With limited bandwidth, size rules the day for the iGPU, and there, VEGA currently wins. Now, that's not to say that RDNA2 and potentially 3 don't improve on that. We just don't know. Currently, there isn't a major feature that RDNA supports that VEGA doesn't, it's just faster and more power efficient in certain cases, at the expense of die area, and when VEGA and RDNA are throttled, and on a similar node, the difference is trivial. So it doesn't make sense currently, on N7, to push for RDNA. It may at N5P.

At present, AMD has leadership in the iGPU space, though, that's not as big of a gap as it used to be witnessing Intel's G7 iris plus products. If they have a product that is competing well there now, there's no need to sacrifice other areas of the chip to push it even further.

All of the above is why I don't find it upsetting or bad that it looks like the next APU after Renoir will be ZEN3 with VEGA on either N7 improved or N5P. If it's got barely faster RAM, there's no need to overhaul the GPU side. The only changes to DirectX have been to incorporate ray tracing, which is useless in current APUs given floor plan constraints.

With the following APU, having DDR5, it should bring a decent improvement in bandwidth, though I doubt it will be more than 50%. However, going to RDNA at that point, with it's better memory management and on a more compact node that makes the extra circuit count of RDNA CUs not as much of an issue, could make for a noticeable improvement in overall performance. But, again, I doubt that we're talking about performance that gets past RX550 levels, much less RX560.

 

[Gideon]

At present, AMD has leadership in the iGPU space, though, that's not as big of a gap as it used to be witnessing Intel's G7 iris plus products. If they have a product that is competing well there now, there's no need to sacrifice other areas of the chip to push it even further

You can't make such decisions looking at things "at present". Tiger lake will be out before the year end and will overtake current Vega performance. The chip you are talking about will come out Q1-Q2 2021 and last till Q2 2022, Intel will also revise their mobile chips at least twice during that timeframe.

If it were released tomorrow I'd probably agree, but it's not. Vega will be as old as skylake is now, by the time it's retired

Also RDNA2 has tons of new features: VRS, mesh shaders, which would help IGPs in perf considerably

 

[uzzi38]

You can't make such decisions looking at things "at present". Tiger lake will be out before the year end and will overtake current Vega performance. The chip you are talking about will come out Q1-Q2 2021 and last till Q2 2022, Intel will also revise their mobile chips at least twice during that timeframe.

If it were released tomorrow I'd probably agree, but it's not. Vega will be as old as skylake is now, by the time it's retired

Also RDNA2 has tons of new features: VRS, mesh shaders, which would help IGPs in perf considerably

Gen13 isn't going to release before EoY 2023... but I agree, AMD should really push for RDNA2 in APUs way before that.

And yeah, RDNA2 will help a whole load. Quick reminder that even at 4K the 5600XT performs within 2% of the also bandwidth-bound Vega64 with only 282GB/s bandwidth vs 484GB/s.

 

[AtenRa]

Yeah still usingVega on Cezanne in 2021 would be a huge blunder by AMD. Even if Tiger Lake (and it's successor) wont murder it in performance, they would in features as Vega would be 4 years old tech.

Let's recap:
* Desktop Vega was released in August 2017 (being delayed from may)
* Raven Ridge mobile zen1 APU using it was released October 2017 mere 6 months after the initially planned release date

* RDNA1 was released June 2019
* Renoir was released Q1 2020 and couldn't really use it as it was on a different node (N7 vs N7P), unlike Raven Ridge
* RDNA2 Big Navi is coming "in a couple of months" according to Lisa's last interview, the design should be fixed for a while already (ramping up console chips)
* Cezanne is coming in H1 2021 and is also at least on N7P so there should be no problem using at least RDNA1 (worst case)

What features ??? If you mean Display and Video Codec then VEGA 7 has new features vs Vega 64

As for the different node process i dont believe it has anything to do with the ability to use VEGA architecture in APUs, VEGA 64 was made at 14nm and VEGA 7 at 7nm.
Same with APUs, RavenRidge and Picasso with VEGA were made at 14nm, Renoir with VEGA was made at 7nm.

I dont know if ZEN 3 APUs will use again VEGA mArch but its not impossible if they believe they will still have the performance lead.

 

[AtenRa]

Gen13 isn't going to release before EoY 2023... but I agree, AMD should really push for RDNA2 in APUs way before that.

And yeah, RDNA2 will help a whole load. Quick reminder that even at 4K the 5600XT performs within 2% of the also bandwidth-bound Vega64 with only 282GB/s bandwidth vs 484GB/s.

VEGA 10 used in VEGA 64 graphics Cards and APUs (Raven and Picasso) and VEGA 20 used in VEGA 7 graphics cards and APUs (Renoir) are not the same.
AMD could even have created an even more refined version for Cezanne , thats not impossible

 

[DrMrLordX]

AMD could even have created an even more refined version for Cezanne , thats not impossible

That would be my guess.

 

[LightningZ71]

The difference between Vega 10 and Vega 20 is effectively minimal. There wasn’t much added to the encoding engine. They clock higher. And there was a little tweaking to the internal data pathways. It was largely just focused on higher clock speeds, and that’s absolutely what Renoir has. Yes, there were added data types and machine learning instructions, but those are hardly relevant in an APU. It also looks like the double precision improvements that we saw in Radeon VII aren’t present.

so, what would further refinements to VEGA look like for Cezann? I can see AMD just making further clock speed tweaks and maybe also adding back the compute improvements as they may actually be important in the future.

 

[flash-gordon]

I dont believe we will have console performance levels on desktop APUs.
AMD cant push this without cannibalizing their partners consoles. They will always keep it segmented.

 

[Gideon]

As for the different node process i dont believe it has anything to do with the ability to use VEGA architecture in APUs, VEGA 64 was made at 14nm and VEGA 7 at 7nm.
Same with APUs, RavenRidge and Picasso with VEGA were made at 14nm, Renoir with VEGA was made at 7nm.

I pointed out the node processes to highlight something that has been mentioned on this board before.. That takes 8+ months to develop an architecture on a new node, even if it's highly synthesized, the only exceptions are some nodes that can reuse old masks e.g. Global foundries 14nm -> 12nm , TSMC 7nm -> 6nm. That in turn takes time and money and most importantly it also uses up people (from a limited pool @amd) for those 8 months. That's why AMD probably wants to keep as many designs on a similar process as possible (to speed up designs reusing components)

One of the reasons why AMD could easily use Vega in Renoir and not RDNA1 was precisely that they had already released 7nm VEGA already in february 2019 (being their "pipe-cleaner" for the process).
Vega only exists on 14nm and 7nm bulk not N7P or N7+ while RDNA1 is available on N7P and RDNA2 on either N7P and/or N7+ (depending who you believe regarding consoles).

I very much doubt that zen 3 APU is still on bulk 7nm as that would mean backporting Zen 3 core (from Vermeer chiplet, that we know to be "improved 7nm"), that in turn made RDNA seem like a more likely candidate.

And if they are concerned about die-area (which they surely are) they can just cut ~30% of the CUs and still get comparable performance:

 

[soresu]

One of the reasons why AMD could easily use Vega in Renoir and not RDNA1 was precisely that they had already released 7nm VEGA already in february 2019 (being their "pipe-cleaner" for the process).

Weren't the first MIxx Vega 7nm cards out earlier than that?

 

[Gideon]

Weren't the first MIxx Vega 7nm cards out earlier than that?

Yeah, of course you're correct. I looked up VII @ techpowerup database, but forgot MI.

MIxx versions were released in Nov 18th, 2018.

 

[Shivansps]

If AMD uses Vega for Cezanne they only need to increase the CU number from 8 max back to 11 again... or maybe even 14...

If the 5400G ends up having Vega 10-11CU at 2ghz, it would be a nice improvement over the (likely) 7CU@1.9Ghz on the 4400G Renoir.

AMD needs to aim for "as close as possible" to RX570 perf on DDR5 APUs, wharever else happens on DDR4 apus is not important anymore.

 

[LightningZ71]

On an improved process for Cezanne, assuming that it can get at least 15-20% improvement in density, then it shouldn't be too far fetched to think that they could fit two more CUs (total 10) and clock them another 10% higher to around 2000-2300Mhz depending on the model. That would effectively be a 30+% improvement in throughput. I suspect that first gen DDR5 won't give drastically greater numbers than high end DDR4 sticks that are well tuned. If we assume that there will continue to be the same generational improvements to graphics performance, then it'll be firmly in between RX 550 and RX 560 levels of performance on average. There will be some cases where the memory bandwidth limitations will be too great and it won't get there, but in others that aren't quite as bandwidth sensitive, it'll easily get there.

I don't think that we'll see anything close to RX 570 performance from an APU until they go to RDNA/2 for the iGPU and we get into the faster DDR5 batches. RDNA has improvements in bandwidth throughput efficiency that could be profound at that level, especially when combined with what DDR5 should be able to deliver by then.

Remember, when you're asking for RX570 levels of performance, you're basically asking for what Intel and AMD were putting out with their combined package that had a "VEGA" (really modified polaris) gpu, an intel laptop CPU, and a single HBM stack combined. Yes, technology advances, but that single HBM stack was providing a considerable amount of VRAM throughput, to the tune of 204GB/sec. For comparisson, in an ideal scenario, Dual channel DDR4-4800 could potentially provide about 75GB/sec, and that's going to be shared betwen the gpu and the CPU cores. Dual channle DDR5 -5200 is expected to push 82.4GB/sec and DDR5-6400 is expected to top out somewhere just north of 100GB/sec in a dual channel config. So, the MAX that we're HOPING for in an APU would be about half the memory throughput of KadyLake-G with the HBM stack in a dedicated setup.

At best, we're looking at relegating the Radeon RX-5500m product line into redundancy. The 5600 should stay comfortably ahead of it. That's current products. Cezanne won't be around until net year. By then, RDNA 2 will be on the market with 6000 series products with improvements in efficiency over the 5000 series.

As long as APUs are limited to dual channel RAM (or, quad channel as in LPDDR4X, etc) you won't get the kind of performance that a dGPU can deliver. For every generation of APU, the GPUs will have at least four times the memory bandwidth (or a lot more in higher end products) and will be far more capable just due to the ability to separately dissipate heat and not have to fight for their memory bus. That's not to say that you won't be able to game on APUs, you just have to have reasonable expectations. 1080p with modest quality settings should be reasonably achievable in this and next generation APUs.

 

[Shivansps]

Yeah, RX570 level performance in a APU is more of a wish than something that will happen in the next two years. Nevertheless, i belive AMD can hit at the very least for RX560/GTX1050TI levels. The 3400G already is almost RX550 performance, to the point that a RX550 is not considered an worthwhile upgrade over Vega 11. Renoir is not going to be much of a upgrade over 3400G Vega 11, but still should push it closer to RX550.

The first DDR5 APU should around 80GB/s and at least RDNA 2, i think it should be able to beat 16CU Polaris GPU with just 112GB/s, and that would be very nice actually.

32CU and 224GB/s is another matter enterely.

 

[TESKATLIPOKA]

With limited bandwidth, size rules the day for the iGPU, and there, VEGA currently wins. Now, that's not to say that RDNA2 and potentially 3 don't improve on that. We just don't know. Currently, there isn't a major feature that RDNA supports that VEGA doesn't, it's just faster and more power efficient in certain cases, at the expense of die area, and when VEGA and RDNA are throttled, and on a similar node, the difference is trivial. So it doesn't make sense currently, on N7, to push for RDNA. It may at N5P.

Vega may be smaller than in RDNA1, but RDNA1 has a much better IPC(35-40%), so It doesn't necessarily mean Vega has better performance/size ratio.

Vega 64 is 4% faster at FullHD and 9% faster at 4K than 5600xt.
Vega 64 -> 64CU, ~1530mhz, 484GB/s
5600XT -> 36CU, ~1700mhz, 288GB/s
If you scaled them both down to 8CU to be comparable to IGP in APU while keeping their average clockspeed It would look like this:
Vega 64 -> 8CU, ~1530mhz, 61GB/s
5600XT -> 8CU, ~1700mhz, 64GB/s
Vega would end up with lower bandwidth requirement, but RDNA1 would be significantly faster thanks to much better IPC and higher clocks so It would have much better performance/bandwidth ratio than Vega.

Naturally RDNA1 is not the same and RDNA2 will most likely end up bigger on same process, but performance or bandwidth requirement should be better rathen than worse compared to RDNA1.
So my conclusion is that It doesn't make any sense using Vega architecture for the next gen of APUs.

 

[SPBHM]

I dont believe we will have console performance levels on desktop APUs.
AMD cant push this without cannibalizing their partners consoles. They will always keep it segmented.

maybe he means current consoles? like PS4 performance? that might happen,

because yes. PS5/SeriesX have RDNA2 with something around 400/500GB/s memory, no way DDR5 ram is going to get near enough