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

[Frenetic Pony]

One of the questions this all raises for me is the cost of building the BVH and the cost of updating it. If "next gen" games have geometry that's almost entirely dynamic (destruction everywhere as well as animation), what can you ray trace?

There's content to use restrictions for raytracing that simply might not be overcome, especially on a fixed bvh like Nvidia has. You can hack reflections somewhat, as they're not as noticeable if they don't match up 1 to 1. EG see the t-pose npcs in Miles Morale's reflections. Amusing but not terrible.

Shadows on the other hand need to be at least close to one to one, otherwise leaking and false shadowing occur, and that does look pretty terrible. There's been concern about how on earthy you can even manage to refit an entire moving forest of leaves in realtime even with a programmable bvh, let alone fixed function. You can see the effects already in CoD Black Ops:

You can see how the "wind" doesn't seem to effect the foliage almost at all, from the mostly static grass right at the beginning to the basically entirely static jungle a bit later. Meanwhile for rasterization vertex shader tricks have gotten so ridiculously cheap that you can move individual blades of grass separately all the way back on the Wii-U (see Breath of the Wild) without a problem.

 

[Dictator]

There's content to use restrictions for raytracing that simply might not be overcome, especially on a fixed bvh like Nvidia has. You can hack reflections somewhat, as they're not as noticeable if they don't match up 1 to 1. EG see the t-pose npcs in Miles Morale's reflections. Amusing but not terrible.

Level of Detail for BVH has been demonstrated on NV hardware - so it is not a limitation. What is happening in that miles morales screen shot is not level of detail causing t-posing Rather the game code for animation and characters not caring about how they look when the camera is not looking at them. A different "issue" altogether.

Basically, your gameplay systems are more lenient in fidelity and simulation but ray tracing now exposes the cheating.

Shadows on the other hand need to be at least close to one to one, otherwise leaking and false shadowing occur, and that does look pretty terrible. There's been concern about how on earthy you can even manage to refit an entire moving forest of leaves in realtime even with a programmable bvh, let alone fixed function. You can see the effects already in CoD Black Ops:

You can see how the "wind" doesn't seem to effect the foliage almost at all, from the mostly static grass right at the beginning to the basically entirely static jungle a bit later. Meanwhile for rasterization vertex shader tricks have gotten so ridiculously cheap that you can move individual blades of grass separately all the way back on the Wii-U (see Breath of the Wild) without a problem.

If you read the modern warfare 2019 ray tracing presentation - they mention how vertex animation grass and foliage is a limitation that they will go around in future products by actually using real mesh representation for grass and leaves and animating them that way.

https://www.activision.com/cdn/research/Raytraced_Shadows_in_Call_of_Duty_Modern_Warfare.pdf

Slide 61: "Alpha testing will most likely have to be replaced with high poly opaque BLASes, but actual performance/mesh complexity balance is yet to be measured."

That would be a positive thing for ray tracing against them as well as the fidelity of such assets. It will be nice to see how these games look once the last gen is dropped and the asset production can catch up to the new techniques.

 

[HLJ]

https://www.proshop.de/AMD-Radeon-RX-6000-Series-overview

It seems everything is supply constrained atm.
I added up the Proshop numbers for both AMD and NVIDIA (orders shipped to customers to date 25/11 2020):

RTX 3070 - 1.361
RTX 3080 - 1.135
RTX 3090 - 482
RX 6800 - 100
RX 6800 XT - 25

Will be interesting to observe the next few months and I hope Proshop keeps posting numbers like this for the future releases.

 

[Jawed]

https://www.proshop.de/AMD-Radeon-RX-6000-Series-overview

Within a week they've received 125 RX 6000 series cards. After 1 month they'd received 344 RTX 3080s:

https://videocardz.com/newz/danish-retailer-fulfiled-only-10-of-geforce-rtx-3080-preorders

and 78 RTX 3090s.

Nothing surprising here except that Azor guy on twitter giving people false hope that Radeon would succeed where GeForce, Zen, PlayStation and Xbox didn’t.

Do we have counts for shipped consoles?

It's at least in part trying to assimilate a fair number of other specialized software methods as well in the form of vertex, geometry, and tessellation shaders. Getting around the serialization of reading from the index buffer would be one fixed-function path replaced, although I'm not sure that on its own would be considered inherent to the hardware concept rather than a result of history. If Turing is used as an example, there were specific cases where the traditional pipeline could still win, like some tessellation scenarios. The continued evolution of the fixed-function pipeline also meant that the mesh shaders could frequently neglect certain things like the full suite of per-primitive culling options if the hardware could cull them easily. Part of the challenge would be weighing a fully-featured replacement shader versus the additional cost each programmable option incurs on the generic execution loop.

The point of my comparison: "work smarter nor harder".

AMD appears to be using primitive shaders, "to work smarter" in RDNA2. So we already have a proof. And that's before developers write mesh shaders.

The BVH structure would likely be a black box until there's more standardization. There's not a firm consensus on what the BVH should look like at a lower level, and it's often tuned to the packing and access requirements of specific architectures. I wouldn't relish the prospect of having to write a new BVH node format or heuristic for every cache subsystem that exists or could exist.
Part of winnowing down the possibilities would come from the hard data from deployed RT hardware, but without the acceleration that we have the data would not be forthcoming.

I agree. It would be like being able to work directly with the data in a render target, without the API getting in the way. Sort of how devs were flipping bits in XB360's render target... Alright if you have a single platform to target and you're a bit mad...

Perhaps "BVH Feedback" could be the concept we're looking for? Similar to Sampler Feedback?

However, is it certain that the most serious bottleneck for ray tracing is the acceleration hardware for the BVH? The most performant solution so far is the one with more dedicated hardware.

Yes, it appears NVidia is spending way more die area on ray acceleration... and all the comparisons so far are based on code optimised for that hardware.

Is it a clear win to have a generic software solution running on shader hardware. The AMD execution loop is partially programmable, and as such has a round-trip between the texturing domain, 32-wide granularity, LDS capacity and bandwidth contention, LDS latency, and a pointer-chasing workload on a long-latency memory critical path.
Making more of the process programmable means adding more dependence on things like the SIMD hardware, LDS contention, and more trips through the memory pipeline.
It may not be guaranteed that a more clever solution can get out from under greater fixed costs.

There's no doubt that you don't do a dumb port of a MIMD algorithm to SIMD hardware, we've got over 10 years' proof of that

At the same time it seems AMD is warning developers not to do fancy workgroup style coding in their custom traversal shader (LDS space is too tightly constrained, sigh), so maybe it's a dead end.

Maybe there's mileage in a hybrid of texture-space shading and ray tracing? Texture-space shading can be accelerated with sampler feedback, e.g. to identify parts of the frame where no rays should be cast or to identify dead rays that when cast produced no useful result (e.g. ray exceeded path length).

A custom BVH traversal algorithm that can sample from a texture might be useful

 

[HLJ]

Within a week they've received 125 RX 6000 series cards. After 1 month they'd received 344 RTX 3080s:

https://videocardz.com/newz/danish-retailer-fulfiled-only-10-of-geforce-rtx-3080-preorders

and 78 RTX 3090s.

This might be a fun thing to compare:
1 month:

Fulfilled:
RTX 3080 - 344
RTX 3090 - 99

Incomming:
RTX 3080 - 123
RTX 3090 - 55


1 week:

Fulfilled:
RX 6800 - 100
RX 6800XT - 25

Incomming:
RX 6800 - 13
RX 6800XT - 9

In Denmark there are also a lot of whine over the availability of new Xbox/PS consoles...and AMD CPU's
Seems everything is in tight supply atm.
Crossing my finger for that it doesn't affect servers.
Just ordered 34 x Dell P570 VXRails...less that a week of delivery-time...so for me it seem consumer space is hit way harder than Enterprise.

 

[DavidGraham]

How does the AMD Radeon 6800 handle Strife RTX / UE4 NvRTX_Caustics ?

Gives credence to the idea of AMD DXR drivers not being really ready.

 

[DavidGraham]

How does the AMD Radeon 6800 handle Strife RTX / UE4 NvRTX_Caustics ?

Gives credence to the idea of AMD DXR drivers not being really ready.

 

[CarstenS]

Within a week they've received 125 RX 6000 series cards. After 1 month they'd received 344 RTX 3080s:

https://videocardz.com/newz/danish-retailer-fulfiled-only-10-of-geforce-rtx-3080-preorders

and 78 RTX 3090s.

Maybe it's a bit too early to draw conclusions. The question is, can AMD re-stock the distributors/shops more quickly than Nvidia could.

 

[xEx]

Gives credence to the idea of AMD DXR drivers not being really ready.

And/or that RT on AMD is not as simple as turn it on.

 

[chris1515]

Level of Detail for BVH has been demonstrated on NV hardware - so it is not a limitation. What is happening in that miles morales screen shot is not level of detail causing t-posing Rather the game code for animation and characters not caring about how they look when the camera is not looking at them. A different "issue" altogether.

Basically, your gameplay systems are more lenient in fidelity and simulation but ray tracing now exposes the cheating.


If you read the modern warfare 2019 ray tracing presentation - they mention how vertex animation grass and foliage is a limitation that they will go around in future products by actually using real mesh representation for grass and leaves and animating them that way.

https://www.activision.com/cdn/research/Raytraced_Shadows_in_Call_of_Duty_Modern_Warfare.pdf

Slide 61: "Alpha testing will most likely have to be replaced with high poly opaque BLASes, but actual performance/mesh complexity balance is yet to be measured."

That would be a positive thing for ray tracing against them as well as the fidelity of such assets. It will be nice to see how these games look once the last gen is dropped and the asset production can catch up to the new techniques.

This will be so good no more strange looking vegetation.

 

[pharma]

And/or that RT on AMD is not as simple as turn it on.

That is likely accurate and aligns with some recent info from @Dictator.

So you could make a different level of BVH depth if you wanted on console or other custom things - something not at all provided in DXR. On the PC side, AMD is having their own traversal kernel in the AMD driver and cannot be accessed in DXR. Nvidia is doing what ever the heck Nvidia is doing there in hardware on the RT core.

https://forum.beyond3d.com/posts/2178552/

 

[Scott_Arm]

So the AIB pricing is ...

The MSRP is $120-150 USD above AMD's announced 6800XT price. Pretty crazy.

Edit: Not distributors or retailers that are increasing the price. Powercolor and Sapphire pricing is "low margin" for those companies.

 

[Kaotik]

So the AIB pricing is ...

The MSRP is $120-150 USD above AMD's announced 6800XT price. Pretty crazy.

Edit: Not distributors or retailers that are increasing the price. Powercolor and Sapphire pricing is "low margin" for those companies.

Something doesn't add up, they mention "10% profit margin" at current prices. If that was correct, they sold references at about 0% profit margin (or even negative, didn't actually do the math, just estimation) and AMD is selling them at loss from their own shop.

 

[trinibwoy]

This will be so good no more strange looking vegetation.

It would be fantastic.

You can specify unique transformations per instance in the BLAS but presumably deformations would still be done at a relatively high level (shrub, branch etc) and not for individual blades of grass or leaves.

 

[Jawed]

So the AIB pricing is ...

The MSRP is $120-150 USD above AMD's announced 6800XT price. Pretty crazy.

Edit: Not distributors or retailers that are increasing the price. Powercolor and Sapphire pricing is "low margin" for those companies.

Since when have custom premium AIBs been MSRP? Not in my lifetime.

 

[Nebuchadnezzar]

Something doesn't add up, they mention "10% profit margin" at current prices. If that was correct, they sold references at about 0% profit margin (or even negative, didn't actually do the math, just estimation) and AMD is selling them at loss from their own shop.

AMD has a profit margin from the chips. Board partners have to buy the chips.

 

[Scott_Arm]

Since when have custom premium AIBs been MSRP? Not in my lifetime.

Is $800 for a card with a reference model at $650 normal? That seems insane for an air cooled card.

 

[Leoneazzurro5]

Is $800 for a card with a reference model at $650 normal? That seems insane for an air cooled card.

https://www.techpowerup.com/review/evga-geforce-rtx-3070-ftw3-ultra/

110$ above the MRSP

https://www.techpowerup.com/review/evga-geforce-rtx-3080-ftw3-ultra/

same

https://www.techpowerup.com/review/colorful-igame-geforce-rtx-3080-vulcan-oc/

180$ above MSRP

 

[xEx]

So we have to get a last gem card until prices become sane again. I heard 2080S are in good prices now

 

[Kaotik]

AMD has a profit margin from the chips. Board partners have to buy the chips.

In case of reference models, all the cards are supplied by AMD as whole rather than as chips. AMD is selling them at MSRP. AIBs reference models were marked up clearly everywhere from those (regardless of who's behind the markup, be it AMD pricing, AIB pricing, importer, retailer, whoever).
AIB customs seem to be even more expensive and if they quote 10 % profit margins for those, there's just no way they made any money from reference models and AMD can't have normal margin on their reference either.
AMD needs to take into account that profit margin per chip when they price their card, no matter if they're selling it themselves or via AIB, or lower their own profit margins which I can't see happening really.