AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

The thing is, we have no idea how stable their OC is across all games on the platform (PC). And more importantly how stable that OC would be across all retail cards featuring that chip. There are likely retail cards that can't hit that OC. And we don't know what yield cut-off MS was targetting. Without the possibility of salvage chips, they would have to go more conservative with clocks.

Regards,
SB

Yeah, MS have stated that every working 52 CU XSX chip is within their clock and power thresholds. They don't scrap anything that works properly. With the high cost of 7nm and limited capacity this is understandable.

For PS5, Mark Cerny stated explicitly during his Road to PS5 presentation that 2.23hz was as high as they could boost and be sure of operation. He outright stated that - he was not being conservative at the time, he was at the limit. And these are systems that must have a uniform performance profile across every single unit. And unlike PC cards, these decisions must be made many, many months before release. They can't set clocks for relatively tiny numbers of units (by console standards) 6 weeks before launch, and have them set to boost opportunistically based on that particular part and its local environment. MS had set their clocks by the end of 2019.

PS5 and XSX have slightly older designs than Navi 21 (which can actually drop to sub 2 ghz under stress!) and especially Navi 22, and you can't engineer based on hope.

It's a bit unrealistic when people look at clock speeds of binned parts, that enter mass production months later, with matured and likely tweaked manufacturing processes, with later revisions of the tech, and say "that's where consoles should be". Just look at how much has changed between November and March, never mind December 2019 and March 2021.

If MS and Sony were in the position of deciding on clocks for new parts today, they would be higher. But then you wouldn't be seeing those systems for 9 ~ 12 months.
 
Ev-ltPhWEAMo2iX
 
Yeah, MS have stated that every working 52 CU XSX chip is within their clock and power thresholds. They don't scrap anything that works properly. With the high cost of 7nm and limited capacity this is understandable.

For PS5, Mark Cerny stated explicitly during his Road to PS5 presentation that 2.23hz was as high as they could boost and be sure of operation. He outright stated that - he was not being conservative at the time, he was at the limit. And these are systems that must have a uniform performance profile across every single unit. And unlike PC cards, these decisions must be made many, many months before release. They can't set clocks for relatively tiny numbers of units (by console standards) 6 weeks before launch, and have them set to boost opportunistically based on that particular part and its local environment. MS had set their clocks by the end of 2019.

PS5 and XSX have slightly older designs than Navi 21 (which can actually drop to sub 2 ghz under stress!) and especially Navi 22, and you can't engineer based on hope.

It's a bit unrealistic when people look at clock speeds of binned parts, that enter mass production months later, with matured and likely tweaked manufacturing processes, with later revisions of the tech, and say "that's where consoles should be". Just look at how much has changed between November and March, never mind December 2019 and March 2021.

If MS and Sony were in the position of deciding on clocks for new parts today, they would be higher. But then you wouldn't be seeing those systems for 9 ~ 12 months.
valid and fair points. Months of improving process over what the consoles had to ship with.
 
What.
N7 D0 has been super-mega low for like a year now.

Not much left to improve; N7 is tapped out and superceded by N6/N5 duo for cost optimized/bleeding edge.
Are N6/5 what the 6000 series are based on?

If not, I do believe there are still changes that can be made to every design to improve parametric yield.
 
I responded to this on twitter since that user tagged me there, but this is in reference to a rather random sample scene vulkan RT implementation not Vulkan ray tracing in general (or ray tracing in general), While curious a change for how that sample was initially coded, I honestly do not think seasoned graphics programmers are going to be launching width 64 wavefronts on RDNA 2 in games with native RT implementations. Ampere, Turing, RDNA 2 are 32 width wave/warp - it is just a design quirk for GCN back compat essentially that RDNA 1 + 2 can also do width 64.

It is also constantly mentioned to do as such as a part of RDNA 2 best practices for ray tracing and other rendering from AMD themselves.
 
I responded to this on twitter since that user tagged me there, but this is in reference to a rather random sample scene vulkan RT implementation not Vulkan ray tracing in general (or ray tracing in general), While curious a change for how that sample was initially coded, I honestly do not think seasoned graphics programmers are going to be launching width 64 wavefronts on RDNA 2 in games with native RT implementations. Ampere, Turing, RDNA 2 are 32 width wave/warp - it is just a design quirk for GCN back compat essentially that RDNA 1 + 2 can also do width 64.

It is also constantly mentioned to do as such as a part of RDNA 2 best practices for ray tracing and other rendering from AMD themselves.
Further, just sort of following the thread, I also don't think those particular demos are reflective of real games. You can benchmark in a variety of ways to improve performance of the hardware, and that's fine, but the requirements behind games are different. At the very least, a difference in settings between low, med, high, ultra.
 
Last edited:
Magic Carpet was a very good 3d games for its time, I miss Bullfrog Productions *try to go back on topic by any means necessary*.

Well I'm convinced that last night @Bondrewd dropped the bomb about Van Gogh (9W APU with Zen2+RDNA2+LPDDR5) being exclusive to Surface devices so I'm waiting for him to wake up and spill more beans.

What is going to come out of here? Xbox Mini? X-Surface?
It also seems AMD is creating a whole new line of APUs with Van Gogh, since according to roadmaps that APU is already supposed to be followed by Dragon Crest in 2022.
 
Well I'm convinced that last night @Bondrewd dropped the bomb about Van Gogh (9W APU with Zen2+RDNA2+LPDDR5) being exclusive to Surface devices so I'm waiting for him to wake up and spill more beans.

What is going to come out of here? Xbox Mini? X-Surface?
It also seems AMD is creating a whole new line of APUs with Van Gogh, since according to roadmaps that APU is already supposed to be followed by Dragon Crest in 2022.
I'm having hard time believing Van Goghs being branded as 4000-series, or does he mean 2022 Surfaces? Since this years updates, while bringing more AMD options, are using R5 4680U and R7 4980U
 
I'm having hard time believing Van Goghs being branded as 4000-series, or does he mean 2022 Surfaces? Since this years updates, while bringing more AMD options, are using R5 4680U and R7 4980U
That's the Surface laptop though right? What about the next Surface Pro (8?) which could be arriving this year?
 
If Van Gogh it's a custom product it may not even be branded like the other APUs at all.
Btw, it is obvious that when going for a multi-chiplet architecture packaging complexity and costs will be higher. this was already the case for Zen CPus. In the case of GPUs, things are much more complicated than for CPUs due to the sheer amount of bandwidth and data needing to be exchanged. TSMC and other manufacturers presented a lot of solutions for advanced packaging, but prototypes <> commercially viable. We'll see which the real availability will be, but it's clear any delay in the packaging project pipeline will reflect on the actual availability.
 
What is going to come out of here? Xbox Mini? X-Surface?
It also seems AMD is creating a whole new line of APUs with Van Gogh, since according to roadmaps that APU is already supposed to be followed by Dragon Crest in 2022.
Seems more like a reboot. They had it but paused after Stoney Ridge for unknown reasons. Maybe they were projecting Intel to do better esp. in this <15W space, but it turned out not.
 
what even happened here
Why would you conflate IP with a particular a chip design or layout
Because they DTCO IP families for target nodes; that's how everyone does it (unless you're Intel. Well, unless you *were* Intel).
No magic voodoo for consoles here; they get what AMD actual is using (i.e. reference implementation with whatever higher-level tweaks particular customer wants).
A console manufacturer might have to make a decision months earlier.
No, why would they?
Console stuff is determined enterily by longer-term pathfinding roadmaps more than anything else (and, say, MS actual isn't pretending it's not).
The actual IP availability dates aren't anything too different.
So RDNA3 has bad availability for 2022?
Not bad, but you'll pay a pretty penny for it.
Van Gogh is a Surface exclusive?
Well it was.
It's now officially in limbo.
So that super awkward appearance of Panay at AMD's CES presentation was supposed to be about Van Gogh but instead they did those cringe executive mutual backrubs?
Yea, MS h/w refresh went off the rails during 2020 for unspecified reasons which is why we have like super-odd looking spaced-thru-the-year refreshes from MS.
comment was on how long AMD finalised target clocks for the RDNA2 GPUs
Better, AMD had a particular freq target for RDNA2 looooooooooong before Sony even got dibs on A0 Oberons.
They've even missed it a fair bit.
If Van Gogh it's a custom product it may not even be branded like the other APUs at all.
Not really custom, but made for MS first.
But then something happened.
 
Last edited:
I'm having hard time believing Van Goghs being branded as 4000-series, or does he mean 2022 Surfaces? Since this years updates, while bringing more AMD options, are using R5 4680U and R7 4980U

These were the silent updates on the Surface Book line, right? I think those are using just slightly binned 15W parts.

I believe Van Gogh, being a 9W chip is (was?) going into a new line of products.
Or maybe it's just what they're putting into the new Surface Pro line, though I'm inclined to believe Microsoft doesn't like to deviate too much from their comfort zone with Intel. Also, it's not like Intel is getting too much behind on the 15W line of products where the parallel task productivity of an 8-core SoC isn't critical.


Well it was.
It's now officially in limbo.
So if Microsoft doesn't take it, what are the chances that AMD makes Van Gogh available for e.g. handheld gaming PC makers like the AYA Neo or Dell UFO? Do they have any volume of that chip laying around?

Also, what happens to its refresh Dragon Crest?


Better, AMD had a particular freq target for RDNA2 looooooooooong before Sony even got dibs on A0 Oberons.
They've even missed it a fair bit.
3GHz+ double confirmed for 5nm RDNA3, got it.
j/k
 
I guess I've missed a significant portion of the discussion, but why it's just defects that should affect the clocks? Power consumption will go up if your GPU clocks higher (especially in fixed clock mode) even if the voltage remains the same, pretty sure thermal considerations (as well as potential RMA issues due to clogged fan / fin stacks) are also playing a role in how Sony / MS decides what clock their chips should run at. There's also a high frequency (on serdes level) and very wide (on a logic level) IF interconnect inside the same chip, which would probably be very sensitive to all kinds of EM interference.

In any case, why clocks even should matter if we are speaking about an ecosystem (as in console products), not about particular product that directly competes with another to be a part of a bigger system? Pretty sure Switch had success not because it used an nV mobile chip (judging from my experience with GoForce 5500 and Tegra3 GL, in spite of it >.<), but because Nintendo managed to create a successful ecosystem around that console. Of course, there's a weird competition of numbers between xbox and ps generations, but in this case it seems both vendors opted for different marketing tricks (NVMe vs compatibility / high fps in old games)
 
I believe Van Gogh, being a 9W chip is (was?) going into a new line of products.
It's more of a general premium FF thingy for AMD.
Or maybe it's just what they're putting into the new Surface Pro line, though I'm inclined to believe Microsoft doesn't like to deviate too much from their comfort zone with Intel.
SP is also in limbo and got a very silly refresh recently with SP7+.
So if Microsoft doesn't take it, what are the chances that AMD makes Van Gogh available for e.g. handheld gaming PC makers like the AYA Neo or Dell UFO?
Well it's technically alive but no one knows for sure.
The real issue is volumes at this point.
Also, what happens to its refresh Dragon Crest?
Still on roadmaps.
3GHz+ double confirmed for 5nm RDNA3, got it.
Well it's not off the tables, N5p being pretty speedy and all.
 
Back
Top