What he got them right so far:
XSX is 12 TF ✔
XSX RDNA2 (7nm+) ✔
PS5 has faster SSD ✔
Many people made those guesses, not just Tommy. And a lot did so before he did.
The guy is playing a guessing game where the odds are 50% for or against the guess. If you're piggybacking off what a majority are already speculating that ups your odds to land a right guess. That's all he has demonstrated so far: darts at the board with some landing as lucky guesses, around very vague specifications.
But as already mentioned, there were other people guessing at those things before Tommy did, they just weren't calling themselves insiders. Even other, credible insiders were speculating a lot of these things and Tommy would show up to piggyback off of that. I mean when you actually think about the possibilities, your ranges narrow down. XSX @ 12TF wasn't a hard guess: you're talking about a new console for a new generation so you'd think it would at least double the count of the X. RDNA2 was a relatively safe guess given when the system is set to launch, plus there was speculation on both full-blown RDNA2 and RDNA2 roadmap features pulled to an RDNA1 spec well before Tommy showed up. PS5 SSD being faster has been a speculation for a long time (and FWIW, that actually still hasn't been confirmed. It's just the favored trend in speculation for the SSD atm).
There's nothing to Tommy's speculation that seems insider-quality to me, they're no more in the know than the rest of us and I would argue actually less in the know on some terms if they don't know about certain market realities, where certain tech is at progress-wide for mass production viability, specifics on how some architecture works (I remember them claiming specifications that were pretty nonsensical or even impossible, such as with certain memory configurations), etc. That could apply to some of the other insiders too but they have track records (and some actual verification) on their side.
BGs
even highlighted 13.8TF as far as I unterstood it correctly. Would be extremely sick, but I'm still not really convinced.
I think he got that figure after reading the ITMedia article, but that article had some weird wording when mentioning it, saying something to the effect of "equivalent to 13.8TF of Radeon VII".
The wording there is open to questioning, just like Phil's wording of "twice the power of X" was at the TGAs. And the ITMedia article also gave some very low-end speculative figures as well, so I think it was just running a gamut of possibilities for the type of hardware power PS5 would need to hit certain targets, it wasn't actually making a claim for PS5 being of those mentioned targets.
If you translate 13.8 Radeon VII (a Vega II card, aka GCN 5.0) to RDNA, it gives you between 10.35TF Navi (low end, if efficiency is closer to 25%) and 11.04TF Navi (high end, if efficiency is closer to 20%). However, if you translate it to RDNA2 (and I'm going to be conservative here and give RDNA2 efficiency as being 10% more than RDNA1 and up to 30% more; some articles are claiming it's 50% but that seems very large and kinda PR'ish tbh), you can hit around 13.8TF Radeon VII with either: 9.66TF RDNA2 (30% efficiency over GCN), 8.97TF RDNA2 (35% efficiency over GCN), down to 6.9TF RDNA2 (50% efficiency over GCN) or even 6.21TF RDNA2 (55% efficiency over GCN). However I don't think the last two are going to happen because I doubt RDNA2 is that much of an efficiency boost over RDNA1.
As usual there's caveats; it depends on if PS5 is full-on RDNA2 or has RDNA2 features from the roadmap pulled ahead onto a customized RDNA1 chip. In that case the chip would still be RDNA1 but would also have RDNA2 features like VRS and RT on it, obviously. And the reason I still think that's the more probable scenario right now is because it's also the one thing all the Oberon benchmarks fit a pattern for, until (or unless) there's another stepping or revision that has documented proof show up, showing extra hardware (particularly CUs).
That's the main reason my own PS5 speculation still ranges from
10.44TF - 11.05TF RDNA1 w/RDNA2 features integrated into the silicon, as a best-case scenario. But again it needs another benchmark result to show up with hardware able to fit that amount, to cement that.. Worst-case scenario is it's 9.2TF - 10.24TF with the current line of Oberon chips. A larger chip able to push ahead of 11.05 while staying within the sweetspot range is less likely since we know Oberon E0 is as recent as December 2019, but it's possible. They could also push a 48CU chip to 2GHz for 12.28TF, which for all we know could be what they're doing. But again, we need a benchmark to show up with that type of setup to lend hard evidence to it, and it hasn't happened yet. Given the pretty wild development history PS5 seems to have been going through, it's more likely they'd have squeeze in another 12CUs to get to 52 but with 4 disabled to improve yields, working with the spec they'd have been working on since 2019 rather than completely scrapping it (and losing the multi-millions in R&D and architectural planning) to roll with a whole new chip just because of a competitor.
(* They could also just blast all 52 CUs of such a chip on and push to the upper end of the sweetspot for 11.9TF, virtually identical to XSX's 12 (I'm going to say XSX is 12.18TF since 56 CUs at lowest end of sweetspot range gives 12.18.*)
So even while I like and appreciate what guys like
BGs
say on the matter, due to a lot of reasons (including NDAs I'm sure) they can't give too much context. And without that context I have to go to other channels to fill in the picture, since they have more of that context (data) and, moreso, data that's been showing a consistent pattern and been verified through actual products fitting that data precisely.