Discussion RDNA 5 / UDNA (CDNA Next) speculation

[marees]

Still really early in the cycle yet.

I wouldn't go one way or another until at least April-June next year.

Interestingly, Igor's Lab website has reposted wccftech rumour/analysis for its German readers

AMD RDNA5: Architektonischer Neustart oder strategischer Kraftakt? | igor´sLAB

Wenn sich ein Hersteller anschickt, seine GPU-Architektur grundlegend umzubauen, lohnt es sich, genauer hinzuschauen. Bei AMDs kommender RDNA-5-Generation (man munkelt auch „UDNA“) ist genau das der…

www-igorslab-de.translate.goog

 

[marees]

PS6 launch speculation

Is the PS6 Launch Date Delayed? New Sony Comments Hint It May Be.

Sony’s CFO has hinted that the PS5 is only halfway through its life cycle, sparking speculation that the PS6 release date could be delayed.

www.vice.com

Is the PS6 Launch Date Delayed? New Sony Comments Hint It May Be.​

Sony’s CFO has hinted that the PS5 is only halfway through its life cycle, sparking speculation that the PS6 release date could be delayed.

In her latest comments on PlayStation, the executive claims we are only halfway into the life cycle of the PS5.

“There are many active users enjoying the console. So from that perspective, we believe that the PS5 is only in the middle of the journey, and we are really planning to expand it even further.”​

Lin Tao’s latest interview may be at odds with the numerous leaks claiming that the PS6’s 2027 launch is already “locked in.” However, many PS5 players are not only happy with Tao’s comments but are also hoping PlayStation 6 gets delayed even further.

PlayStation Owners Want PS6 Launch Delayed​

many players reacted positively to the idea of the PlayStation 6’s launch possibly being delayed.

PS5 Support Could Overlap PS6 Launch​

when the PS5 launched in 2020, Sony continued making games for the PS4 for years. So, the PlayStation 5 could technically only be halfway into its life cycle while the PS6 still launches in 2027. Regardless, it seems like many PlayStation players are in no rush to jump into the next-generation, and Sony may be on the exact same page.

Is the PS6 Launch Date Delayed? New Sony Comments Hint It May Be.

Sony’s CFO has hinted that the PS5 is only halfway through its life cycle, sparking speculation that the PS6 release date could be delayed.

www.vice.com

 

[CakeMonster]

I frankly don't mind if its delayed. As a PC gamer I've typically been annoyed by games being held back in regards to graphics by console hardware and always looked forward to the next gen of consoles. But with the increased prices of everything lately, I fear the upcoming generation could be really underwhelming compared to current gen, to keep costs in check, and that would doom innovations in games for another 7+ years. I'd rather wait a couple more years and get consoles with hardware that can run bigger AI/ML models which I presume will be used for more than just upscaling and RT, along with plenty of VRAM for those uses.

 

[Vikv1918]

Maybe what he means is that the crossgen period for PS5 would be 3 years long, all the way to 2030, so we're halfway through the period where PS5 and Pro receives all games and is supported. Not necesarily that PS6 releases in 2030.

Edit : my bad, that article literally says the same thing

 

[inquiss]

With DDR5 pricing increasing, and GDDR following suit (but to a lesser degree), does anyone have any thoughts on how this will impact the relative strength of Nvidia and AMD's offerings in the next gen where AMD is expected to use LPDDR Vs Nvidia continuing with GDDR?

Cost differences between LPDDR and GDDR will be more extreme giving AMD a further advantage? What's the thoughts?

 

[adroc_thurston]

With DDR5 pricing increasing, and GDDR following suit (but to a lesser degree), does anyone have any thoughts on how this will impact the relative strength of Nvidia and AMD's offerings in the next gen where AMD is expected to use LPDDR Vs Nvidia continuing with GDDR?

Cost differences between LPDDR and GDDR will be more extreme giving AMD a further advantage? What's the thoughts?

AMD is still doing GDDR for 2 outta 4 parts.

 

[inquiss]

AMD is still doing GDDR for 2 outta 4 parts.

So the top end where the increase matters less will have the same issues but they will be more competitive (assuming price increases impact LPDDR less) in the mid to lower tiers? Will LPDDR be less affected, is that a reasonable assumption?

 

[marees]

So the top end where the increase matters less will have the same issues but they will be more competitive (assuming price increases impact LPDDR less) in the mid to lower tiers? Will LPDDR be less affected, is that a reasonable assumption?

Nvidia will cheap out with lesser vram but force reviewers to flog their new 10x compression technology

AMD will try to market high vram cards to (naive) budget users, the same way camera companies market megapixels

 

[soresu]

Nvidia will cheap out with lesser vram but force reviewers to flog their new 10x compression technology

It doesn't matter who invented it as long as it can be used by anybody.

 

[adroc_thurston]

So the top end where the increase matters less will have the same issues but they will be more competitive (assuming price increases impact LPDDR less) in the mid to lower tiers? Will LPDDR be less affected, is that a reasonable assumption?

No, LPDDR is also mooning.

 

[inquiss]

No, LPDDR is also mooning.

Yeah expected as much, but from a lower cost base so still cheaper than GDDR per GB? If so the LPDDR move should be a strategic win..?

 

[adroc_thurston]

Yeah expected as much, but from a lower cost base so still cheaper than GDDR per GB? If so the LPDDR move should be a strategic win..?

Yes and yes but it's not why it's there.
LPDDR shoreline is for commonality with APU configs.

 

[inquiss]

Yes and yes but it's not why it's there.
LPDDR shoreline is for commonality with APU configs.

Yep, understand, just seems like a decision that has an extra unexpected benefit for AMD in the current environment of rising memory prices. Sounds like it will positively impact competitive positioning

 

[adroc_thurston]

Yep, understand, just seems like a decision that has an extra unexpected benefit for AMD in the current environment of rising memory prices. Sounds like it will positively impact competitive positioning

No, DRAM goes up still means DRAM goes up.

 

[inquiss]

No, DRAM goes up still means DRAM goes up.

Sure but if it goes up 50% and one is half the price of the other, then the cheaper base price one will be in a better position, no?

 

[MrMPFR]

“There are many active users enjoying the console. So from that perspective, we believe that the PS5 is only in the middle of the journey, and we are really planning to expand it even further.”

Late 2020 -> Late 2025 = 5 yrs, add 5yrs = Late 2030
Assuming PS6 is launch late 2027 = 3 yrs crossgen minimum.
With the rise of handhelds and many other factors strong incentives to prolong crossgen significantly. Unfortunately 2032-2033 sounds more likely.

I frankly don't mind if its delayed. As a PC gamer I've typically been annoyed by games being held back in regards to graphics by console hardware and always looked forward to the next gen of consoles. But with the increased prices of everything lately, I fear the upcoming generation could be really underwhelming compared to current gen, to keep costs in check, and that would doom innovations in games for another 7+ years. I'd rather wait a couple more years and get consoles with hardware that can run bigger AI/ML models which I presume will be used for more than just upscaling and RT, along with plenty of VRAM for those uses.

Ray tracing is very scalable. From Probe based DDGI to fully fledged ReSTIR based MLP pipeline (NG rendering). Game AI, physics and other systems can be scalable as well if devs bother.

It won't be underwhelming, well at least not in the post crossgen era. Their NG GPU is not RDNA 2 v2, it's not even RDNA 4 enhanced, it's a different beast altogether. For once there are actually strong indications that GFX13 is a forward looking GPU design.
Even if PS6 is only a 5070S - 9070 in raster it'll go beyond that spec in ML and PT and well beyond in work graphs accelerated workloads. So post crossgen the PS6 will be anything but boring and underwhelming when Programmable Shaders 2.0/Work graphs replaces EI whenever possible, and this is just one change.

Prolonged crossgen will take care of that for the next +7 years on its own.

Delaying it a couple years won't help. N3 -> A14 PPA scaling is terrible (relatively) and anything beyond N3 is going to be ludicrously expensive. An additional ~50-100% higher SoC cost is not feasible, especially not when N3 wafer prices are already terrible compared to N6/N7 in early PS5 cycle.

VRAM issue can be solved with better asset compression tech, data streaming (SF), and Work graphs (scratchpad), procedurally generated content... (ala HPG 2025 tree demo). Even a 20GB PS6 is doable.

 

[SolidQ]

RDNA5 SD2?

Valve Says It Has a 'Pretty Good Idea' of What Steam Deck 2 Is Going to Be, Explains Why It's Holding Off for Now - IGN

Valve announced three new devices today: the Steam Frame, the new Steam Machine, and the new Steam Controller. But where’s the Steam Deck 2? In a new interview with IGN, Valve explains why it's still holding off.

www.ign.com

 

[branch_suggestion]

RDNA5 SD2?

Valve Says It Has a 'Pretty Good Idea' of What Steam Deck 2 Is Going to Be, Explains Why It's Holding Off for Now - IGN

Valve announced three new devices today: the Steam Frame, the new Steam Machine, and the new Steam Controller. But where’s the Steam Deck 2? In a new interview with IGN, Valve explains why it's still holding off.

www.ign.com

No, it is whatever tablet TDP part QC will have at the time.
AMD doesn't compete in that market unless there is enough volume for a semi-custom part. Valve got lucky with Van Gogh.
Qualcomm needed a design win like this desperately to try to legitimise themselves in where they currently suck badly, PC gaming.
Valve engineers now gotta work hard to make AArch64 a first class citizen too.

 

[Kepler_L2]

No, it is whatever tablet TDP part QC will have at the time.
AMD doesn't compete in that market unless there is enough volume for a semi-custom part. Valve got lucky with Van Gogh.
Qualcomm needed a design win like this desperately to try to legitimise themselves in where they currently suck badly, PC gaming.
Valve engineers now gotta work hard to make AArch64 a first class citizen too.

I don't know what's gonna be a bigger disaster, x86 emulation or QCOM GPU uarch trying to run any real-world™ graphics workload.

 

[branch_suggestion]

I don't know what's gonna be a bigger disaster, x86 emulation or QCOM GPU uarch trying to run any real-world™ graphics workload.

They get to learn the hard way.
Could've gone binned Medusa something or even something Intel wouldn't be awful, but alas.

 

[marees]

I don't know what's gonna be a bigger disaster, x86 emulation or QCOM GPU uarch trying to run any real-world™ graphics workload.

They get to learn the hard way.
Could've gone binned Medusa something or even something Intel wouldn't be awful, but alas.

I expect Valve to, at some point, drink the coffee & come around to RDNA 5

2029 Grimlake point 2 or grimlake point3? with zen 7 & RDNA 5+ should do the trick

Until then we can keep dreaming / speculating on Qualcomm handhelds

Outside option is Nv opensourcing their drivers & getting a cheap deal from samsung 2nm or Intel 18a (like they got for switch) — and also releasing an arm-mediatek product that is not borked out of the gate

 

[adroc_thurston]

2029 Grimlake point 2 or grimlake point3? with zen 7 & RDNA 5+ should do the trick

Until then we can keep dreaming / speculating on Qualcomm handhelds

Outside option is Nv opensourcing their drivers & getting a cheap deal from samsung 2nm or Intel 18a (like they got for switch) — and also releasing an arm-mediatek product that is not borked out of the gate

It's not like an IP issue.
Deck needs a 10W part and no one really makes those anymore.

 

[marees]

It's not like an IP issue.
Deck needs a 10W part and no one really makes those anymore.

Mendocino is 15 watts

Grimlake point 4 replaces that I believe. 2 open questions

1. Will Microsoft remove NPU tax & allow CPU/GPU stuff instead?
2. GPU capabilities of a Mendocino replacement

 

[Tigerick]

Updated RDNA5 Lineup Speculation

• As explained in SWV thread, RDNA5 will get double SP per CU. Thus AT2 with max 70CU will get 140CU in old format. That's explain 20% faster performance than RTX4080. It also means AT2 GPU is severely bounded by memory bandwidth.
• Therefore, AMD do not need to clock as high as RDNA4, I am expecting 2GHz+ not ~3GHz. It also means AT2 has headroom to grow. That's why I suspect AMD is reserving XTX model with full die of 70CU for future 40Gbps and 4GB GDDR7 die to appear. That explains the cancellation of AT1 cause AT2 XTX is good enough to compete with upcoming Rubin-70Ti with 24GB 256-bit memory bus.
• There are leaks saying RDNA5 dGPU will be released in Q2 next year. I am actually expecting early announcement in Q1. Thus, the cancellation of RTX-50 Super make senses because it will be bloodbath for NV : no amount of overclocking will save RTX-50 Super series. NV needs to speed up the release of Rubin dGPU. If Rubin dGPUs are indeed fabbed by 3N (variant of 3X), then the earliest release date would be Q3 next year. That gives AMD early head up of next gen dGPU war.
• That's why I am predicting AMD will set higher price point for AT2-70XT and AT2-70. AMD will keep selling RX9070XT until NV able to launch Rubin-60.
• AMD most likely will keep selling N48 in the form of 9070GRE by then. And no, AT3 and AT4 are NOT for cheap dGPU lineup, period. Now that we know Medusa will have XDNA3, where do you think the NPU will reside in AT3, huh?

 

[reaperrr3]

• As explained in SWV thread, RDNA5 will get double SP per CU.

RDNA4 (and RDNA3) already have 128 ALUs per CU, it's just that the current WMMA/dual-issue implementation is too limited and doesn't trigger often enough, so AMD stopped advertising dual-issue FLOPs.
RDNA5 will have the same ALU count per WGP/CU, only with much better utilisation.

• Thus AT2 with max 70CU will get 140CU in old format.

No, see answer above. AT2 is equivalent to 35 WGPs/70CUs in RDNA4.

• That's explain 20% faster performance than RTX4080.

That's only ~30% faster than a 9070XT, strongly pointing to your 140 CU theory being completely wrong (and it is).

• It also means AT2 GPU is severely bounded by memory bandwidth.

No.
~10-20% higher CU IPC and ~10-15% higher clocks, that's where the ~30% more perf are coming from.
With 32-36Gbps GDDR7@192bit, combined with bigger/smarter CU caches and bigger L2, that's enough bandwidth for that GPU.

• Therefore, AMD do not need to clock as high as RDNA4, I am expecting 2GHz+ not ~3GHz.

Since you're wrong about the CUs, you're going to be wrong about this one, too.

AMD wants margins, that means as much perf as possible within as little area as possible.
That means they'll want higher clocks.

• That explains the cancellation of AT1 cause AT2 XTX is good enough to compete with upcoming Rubin-70Ti with 24GB 256-bit memory bus.

Nah. AMD just isn't bothering with competing against GR203, because too many people don't want (or can't afford) cards above 600$, and those that do will probably go Nvidia anyway.
AT1 was likely considered too costly to make for the projected volume and margin.

AT2 is designed to compete with the 6070 and 6060 Ti (and go into Xbox Next), that's where the volume is at, where fixed design cost can get amortized much faster.

• There are leaks saying RDNA5 dGPU will be released in Q2 next year.

The only part from AMD that might release in Q2 next year is CDNA5...
RDNA5 will be more like Q2/3 2027.

• Thus, the cancellation of RTX-50 Super make senses because it will be bloodbath for NV

The only reason for that is lack of competition and surging memory prices.

• AMD most likely will keep selling N48 in the form of 9070GRE by then.

While AT2 is on a more expensive process, it's also smaller, has a narrower memory bus and lower TDP (cheaper PCB), so there won't be any incentive to continue N48.

• And no, AT3 and AT4 are NOT for cheap dGPU lineup, period.

Of course they are. Just because they're also going to be used in some APU SoCs doesn't mean they won't get used as dGPUs as well.

And of course, "cheap" is relative, yeah.
True low-end is largely dead due to lack of volume below the $299 price point.
Either an APU is enough for you, or you need at least a 9060/6060 class card with 12+GB.

• Now that we know Medusa will have XDNA3, where do you think the NPU will reside in AT3, huh?

What are you on about?
RDNA5 parts just don't need an NPU, because the AI capabilities of the GPUs themselves will exceed most NPUs.

The Medusa 3nm-SoC-Die only still has an NPU because it'll be used as stand-alone APU as well, and its small RDNA3.5 IGP can't handle AI workloads.