Github is and was never a problem, the information contained within it was 100% legitimate. The problem has always been, certain idiots with an agenda, using that information, mixing with their own BS and passing them as facts.
To this day, not one of them can answer about the actual CU count on Oberon. Not one, why? Because data is completely silent on shaders count. Meanwhile it provided shaders count for Arden at 3584, which meant 56CUs. All, you have is native 18WGP count for a BC compatibility/regression test. We have few, who on this very site argued it to be from Ariel iGPU profile for Oberon to be RDNA2. So which one is it? Either you cede Oberon to be RDNA1, meaning Github data can be thrown away. Or you can argue that Oberon is RDNA2, and all the testing is of RDNA1 iGPU i.e. Ariel, but then your interpretation of what is actually on Oberon made up completely from your backside and not on the basis of Github.
I've argued this for a very long time, when everyone was convinced Oberon is RDNA1. How can you be so sure about total CU count on Oberon when all you have is a regression test? How can you be so certain and box yourself with maximum limit of 36 or 40CUs, when a bigger NAVI was in existence in form of Arden in that very data itself? (We didn't know it was RDNA2 either) What exactly was stopping Sony from stepping beyond that mark?
Oh, I know, it was because PS5 iGPU was based on RDNA1. And given how power hungry they were and AMD highest performing NAVI discreet GPU tapping out at 9.7TF with 40CUs. Sony were always going to be limited by what they could design, as it would form part of the APU. (But not MS with Arden, because.... just reasons) If only I could be arsed to quote, pages and pages of this non-sense, which I had to sift through on a daily basis on these kind of threads on various places for months and months. All I had thrown in my face were Github data, and ro_game tweets.
But, now when AMD went out of their way to confirm PS5 is RDNA2, then Oberon was suddenly RDNA2 all along and their interpretation of Github data along with ro_game was wrong. I mean, you couldn't make this shit up! Anyways, if the Oberon was RDNA2 all along, it actually strengthens the argument of CU count even more. Want to one of the features of RDNA2? It is extremely power efficient, meaning it would allow AMD to put more CUs and build a bigger GPU. So, take this, we have Gonzalo an APU, housing an iGPU in form of Ariel which is based on RDNA1. So Sony, did spent $$$ to manufacture an APU on a tech/architecture which was never going into final PS5 (Killing the argument of Sony 'wasting money' on an APU they were not going to use). Hence, whenever the decision was taken of a redesign i.e. to switch from RDNA1 to RDNA2 (Ariel to Oberon, in this instance). What advantages did they take of this new RDNA2 architecture? We know they are tapping into VRS and RT, but did they also take advantage of efficiency gains to add CUs on the iGPU, or the decision was to simply clock it higher?
We DON'T know answer to any of this, Github data provides fuckall insight to this question which is what I've been preaching all along. We don't know the full makeup of Oberon, so stop boxing yourself within the confines of 36/40CUs. That restriction made sense if PS5 was using a power hungry RDNA1 architecture, but we know it doesn't. They could've clocked the GPU higher, or added more CUs and gone little wide and slower. We just don't know, anybody who is acting like they do on basis of Github data, are just chatting shit. /endrant
