Like "could have been done in shaders"(but isn't like it's done in the ps5), "offloaded onto dedicated hardware" and "run in parallel with full performance"(separate, in addition to, unlike ps5),...which he States is why he is saying 25TFs if ran with Ray tracing on.
I quote, and note the differences vs ps5:
"Without hardware acceleration, this work could have been done in the shaders, but would have consumed over 13 TFLOPs alone," says Andrew Goossen. "For the Series X, this work is offloaded onto dedicated hardware and the shader can continue to run in parallel with full performance. In other words, Series X can effectively tap the equivalent of well over 25 TFLOPs of performance while ray tracing."
"Could have been done in shaders"
"Offloaded onto dedicated hardware"
"Shader continue(s) to run in parallel with full performance"
I'm not disagreeing or countering anything he's saying; and I'm certainly not ignoring his statements. I'm just saying I don't think what he's saying is what you think he's saying.
Both systems have hardware accelerated ray tracing and both companies have indicated they're using RDNA2's RT. By definition, hardware accelerated means you're offloading at least some of that function to dedicated hardware.
None of this is inherently unique to XSX, he's just explaining what real time ray tracing is and effectively what both RDNA2 and RTX do. When he says "For the Series X.." he's not lying, he's just framing it to sound unique to the console which is what any good employee would do.
"Could have been done in shaders" - RT can be done in shaders on any gpu, it's just very ineffective, there's nothing special about that. That's why the dedicated hardware is created in the first place.
"Offloaded on to dedicated hardware" - The dedicated hardware is the intersection engine which is the AMD equivalent of an RT Core. The PS5 will have this dedicated hardware too, PS5 will not be doing the BHV component of RT on shaders.
"Shader continue(s) to run in parallel with full performance" - This is technically true and kind of the point of real-time RT, but the caveat is that RT isn't just one thing and the RT cores are simply accelerating the most computationally expensive component of the RT calculation (BVH Creation/Traversal). Some of that "full performance" will still be required to do the additional stages (casting the ray probe, shading the result etc.)
You seem to be under the impression that there are additional RT cores outside of the CUs. The RT cores are additional pieces of hardware placed inside the CU; these are the dedicated pieces of hardware he's talking about that will take that offloaded work and accelerate it.
Trust me, if MS had additional ASICs or solutions for accelerating the rest of the RT function (or providing additional functionality like BVH reordering) it would be a big deal and they'd be shouting it from the rooftops and stating it clearly in a way that wouldn't be misconstrued, DF would be all over it. All MS are doing here is conveying the RDNA 2 solution in layman's terms.
It's really this simple:
In RDNA2 and subsequently the Navi 2X GPU, the PS5 GPU and the XSX GPU: the CU contains both the shader logic and the intersection engine (which based on patents is tightly integrated into the texture unit).
In Turing/RTX GPUs the SM contains the shader logic and an RT Core.
There is no apparent special sauce tacked on in this respect..... the Intersection Engines/RT Cores are the special sauce!
They are the dedicated hardware that run in parallel.
Would it be nice to have additional acceleration and is it a possibility? Sure... But nothing anyone official has said on the matter indicates it at this point.