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

That's what i have expected and hoped for. If there is no traversal HW, BVH might be arbitary, DXR implementation is backed with compute traversal and AMDs software BVH data structures.
If this is fast enough to b useful, it means full flexibility - custom BVH can be shared for multiple purposes, LOD is possible. All my initial complainds about RT might be resolved. (But AMD needs to expose intersection extensions.)
Full flexibility possibly comes with an hefty price tag since BVH traversal can be highly irregular and doesn't map well to SIMD cores.

umm... this sentences contradict themselves. Not sure what he tried to say (although german is my natural language), but it seems he means:
'AMD has confirmed traversal and shading can run in parallel' - which would imply traversal runs on FF unit like NV does.

I guess the truth is somewhat in between, but maybe somebody can clarify?
There's no contradiction. It means that while FF HW is performing some ray/box or ray/triangle intersections (which are part of the traversal process) the shader cores can be doing something else, like shading (same as texture mapping). Also they might be devoting some threads to traversal only while others are responsible for shading. Since traversal is partially done in SW there are many ways to skin that cat..
 
4 triangles per clock at up to 128 pixels/clock. A triangle can touch a single pixel or up to 32 and theoretically maintain peak geometry rate. For comparison I think the RTX 3080 is 6 prims/clock and up to 96 pixels/clock.
Thanks for the info. The driver leak and block diagram can look conflicting. This upgraded Raster Unit sounds really nice, and more efficient at rendering small triangles approaching 1 pixel in size.

Navi10 has Primitive Units, where each can send 1 triangle to the Raster Unit, and cull up to 2 triangles. I might be misinterpreting the driver leak, but is the Primitive Unit in Navi21 upgraded to send 1 triangle and cull up to 2 or 4 triangles per cycle?

BTW, would you know what the Packer Per Scan Converter is referring to in the driver leak? The below is what was confusing for Navi21:

num_sc_per_sh = 1
num_packer_per_sc = 4
 
he has a point but at the same time we only have the info we got

The 6800 beats the 3070 in almost all scenarios aside with ray tracing its a mix bag and add dlss on top let the 3070 almost double frame rates vs the 6080,

I think its a shame AMD couldn't show at least a beta of their DlSS competitor

Ultimately people will make the decision to purchase the GPU that best meets their needs. I think AMD will regain marketshare this round, just for simply competing at the high-end again, even if not in ray-tracing.
 
I don't use Twitter or Reddit so thanks for providing this info. Not that it really matters much in the grand scheme of things, but I like my hardware reviewers to be as objective as possible so I will continue to avoid HUB's reviews.

Personally, I don't totally agree with HUB, because I think RT is going to be important if you want to use the card for the next six years (console gen). I don't think they're really biased though. They're just more concerned with what they can bench know, and you can't really know if the current state will be reflective of the future. Just a different mindset, but I wouldn't call it bias. Also I'm pretty sure they've ripped AMD at times in the past.
 
Full flexibility possibly comes with an hefty price tag since BVH traversal can be highly irregular and doesn't map well to SIMD cores.
What about RTX cores - they're MIMD? Makes me wonder why AMD didn't go with something similar - patent issues?
 
4 triangles per clock at up to 128 pixels/clock. A triangle can touch a single pixel or up to 32 and theoretically maintain peak geometry rate. For comparison I think the RTX 3080 is 6 prims/clock and up to 96 pixels/clock.
I think the 128 is a number for retiring pixels, i.e. RB+ throughput.
 
If cyberpunk2077 doesn' move ray tracing cards then nothing in the near future will. I'm expecting people really want to play it with bells and whistles on.

Agreed. On a personal note I have to say, this is the only PC title on my radar right now and I can't wait to see how it looks and plays.
 
If cyberpunk2077 doesn' move ray tracing cards then nothing in the near future will. I'm expecting people really want to play it with bells and whistles on.
I think it will be dependent on what kind of net visual improvements RT brings and the performance achieved.
 
Thanks for the info. The driver leak and block diagram can look conflicting. This upgraded Raster Unit sounds really nice, and more efficient at rendering small triangles approaching 1 pixel in size.

Navi10 has Primitive Units, where each can send 1 triangle to the Raster Unit, and cull up to 2 triangles. I might be misinterpreting the driver leak, but is the Primitive Unit in Navi21 upgraded to send 1 triangle and cull up to 2 or 4 triangles per cycle?

BTW, would you know what the Packer Per Scan Converter is referring to in the driver leak? The below is what was confusing for Navi21:

num_sc_per_sh = 1
num_packer_per_sc = 4

From MacOs Leak:

You have 4 sahder engins:
num_se : 4
you have 2 Shaderarrys for each shaderengine:
num_sh_per_se: 2
and each shaderarry have 1 scan converter:
num_sc_per_sh: 1

so for scan converter you get: 4x2x1 = 8 scan converter

at macos10 the leaker did a failure by transfaring the data, he confirmed macos 11 is right. Its the same like in the linux driver.

rasterizer.jpg


https://www.pcgamer.com/a-linux-update-may-have-let-slip-amd-big-navis-mammoth-core/
linux.png
 
Last edited:
That vrs result is interesting. Not many games support it yet. Could be important this gen.
One thing to note though, the VRS chart measures performance gains in percentiles, it's not measuring actual scene performance, so a GPU can be CPU limited and not post huge gains with VRS enabled. Which is evident by the fact that the 2080Ti/3070 are posting bigger gains than even the 3080 or 3090.

comptoir-variableratebukyi.png
 
I like the 6800 I have not decided between a custom 3070 or a custom 6800. the 3070 is better for RT but worse at raster and only have 8GB it will age like shit. I made the mistake of get a 4GB card instead of an 8GB and now I have to pay the price of lower the resolution to play, I don't want to commit the same mistake again... I like RT but only for CP so far but realistically if it is gets to 60FPS I'm sold. Lets not be naive here and think today's implementation of RT will last 2 years with useful performance(max settings) yet along 6 years. Today RT is at the barely usable state and Nvidia can use it with DLSS but but the HW isn't there yet. I expect AMD RT to age better because of the consoles but at the same time looking at the 4K benchs they probly won't age as good as 16GBs cards with bigger bus.looks like 128MB is not enough to be as useful for 4K as it is for 1440p.

Idk why ppl say this is Zen2 when the flagship is not out yet. We will have to wait for the full unlocked 6900XT to see how much performance Navi2 really have.
 
Ultimately people will make the decision to purchase the GPU that best meets their needs. I think AMD will regain marketshare this round, just for simply competing at the high-end again, even if not in ray-tracing.

With no volume right now, it's complicated to gain anything. They don't take advantage of nVidia paper launch (worse, I've seen people shop for a 6800xt, but leaving with a 3080 because they start to be available here and there).
 
Back
Top