I appreciate the acknowledgement. Truth be told I was trying to straddle in the middle as a means of not casting either source type completely out of the picture. It was just so ridiculous to me to turn away actual data just because it didn't paint the picture some wanted to even consider accepting as a possibility.
At the end of the day the Github leak and testing data were the closest to predicting both systems' specs, but it's easy to see where the final systems varied from even that.
-XSX: The Arden data was with the full chip (56 CUs) enabled and clocked to a point to still deliver performance equivalent to retail unit. IIRC MS did the same thing with the X; devkit chips had the full chip enabled, final units disabled 4 CUs.
-PS5: Since Oberon C0/E0 was the one revision we didn't get full info on, it's probably the one where the clocks got an increase from 2.0 GHz to 2.23 GHz. It was also the same revision with changes to the memory controller to support up to 512 GB/s bandwidth (since the system confirms 448 GB/s, the Japanese side probably won out because they wanted the cheaper GGDR6 chips. Japanese side is also the one favoring selling the system either at-cost or for profit, by the way. So keep that in mind when PS5 price gets announced down the road, you'll know which side won that deliberation as reflected in the MSRP)
The way I saw things from there was to try taking the more logical insider claims and fitting them with what that data was already telling us. When a user on ResetEra made a graphic speculating a 48 CU chip for PS5, it seemed plausible. When I saw users on Beyond3D going in-depth about the Ariel iGPU profile testlists, those seemed plausible. When I saw Rogame's tweets listing Oberon as a Navi 10 chip, I figured that was referencing the Ariel iGPU profile testlists. Sony already had a history of disabling part of the PS4 Pro's GPU for PS4 BC, so it seemed possible PS5 could've maybe had done the same. The rumors of the SSD, RT etc. made it seem less likely it would be able to be a chip same size as XSX and yet deliver ALL of that for the same price or cheaper, so 48 CUs seemed like a sweetspot limit. Anything larger, I held out until maybe another Oberon revision surfaced showing a bigger chip, or a full testing data leak for Oberon C0/E0 showed up showing the actual full chip. Obviously, neither of those happened.
So when it came to insider claims the only ones I actually fully gave major thought to were one of
@HeisenbergFX4 's spec rumors for PS5 (10.5 TF, strikingly close to what the PS5 actually is btw), and
Gavin Stevens
's (11.6; technically a high-clocked 48 CU chip could hit those numbers while not being RIDICULOUSLY clocked). The others, I was more ho-hum on and wanted to wait until another revision leak came out to really give more credence to them. And then the 13 TF stuff that Tommy Fisher (and others) were speculating, just never registered with me. Seemed fantasy-tier, but still more believable than some of the Reddit/4chan "leak"s saying 14 TF or even 15 TF...all of that with a system people were expect to be $399 - $499 btw. Just completely unrealistic.
But it was that ITMedia article that helped put potential real PS5 performance into perspective, when they brought up Radeon VII and 13.8 TF. Just convert that to RDNA #s and you get 10.35 TF for low end to 11.04 for upper end, based on 20% - 25% IPC gain range of RDNA over GCN. And turns out PS5 is 10.3 TF (with big rounding which I don't prefer doing, when it's really more 10.275 TF; same way I don't like rounding XSX to 12.2 or even 12.15, when it's actually 12.147 TF)...not very far off from 10.35 TF at all.
One thing I will say about Tommy is he either got extremely lucky with the XSX guess or he was a MS insider the whole time. Some people were even saying it was TimDog. Guess it's possible
. But them getting that one right obviously made a lot of people think their PS5 numbers would end up right, too. Still seemed unlikely IMO, but up until Road to PS5 anything was technically possible.
End of the day it's disingenuous to call PS5 a 9.2 TF system. It's a 10.275 TF system that might drop to 10.07 TF if going by Mark Cerny's 10% power / 2% frequency drop figure in his speech. For now that should be the absolute lowest anyone can expect TF performance to drop on PS5, until (if ever) we get actual games coming out that push system in such a way where it has to go lower in max load situations. But for now 10.07 TF should be the floor, and seemingly uncommon/rare floor at that until there's a pattern of games that show persistent drops to that level. 10.275 TF should be the ceiling and assumed base for now.
Just wanted to clarify some of my thought process speculating on the systems and why I was going abut the way I did in this thread during that time. Takeaway
hopefully should be, in the future,
NEVER turn away a potential source of speculation when it has actual data attached to it and there is a pattern (and tangentially related corroboration of other data, circumstantial, coincidental etc.) to that data. And also,
DON'T go hard 100% into any given source type. Oh, and don't attack insiders just because they may've gotten some things wrong; can't control the quality of your sources after all. And it's not like insiders got everything wrong (things like the PS5 SSD for example, they've gotten right. Would be nice to learn more on SSD setup in both systems tho among other things).
Hope convos continue to be civil and fair & balanced throughout
