If you, Tommy Fischer are saying PS5 is 13 TFLOPS, then why is
HeisenbergFX4
saying its between 10-11? All three of you should be stating its 13 TFLOPS.
Better question is why has no silicon surfaced in testing results indicating a PS5 chip fitting specifications to reach that number? I mean it definitely sounds like a great target to hit but we have GPU database entries on many, many other AMD GPUs and APUs that have been officiated in one way or another and, more importantly, we have actually datamined information for.
And it still doesn't answer why a certain other chip has been tested as recently up to December from last year. Unless the base/Pro plans are actually holding some validity, or there is another PS5-related APU with data that will show up very soon (if Sony's actually going to talk of specs or specs somehow get leaked out this soon, it shouldn't take more than a month for such a chip's data to be found in AMD's database bank), be it another Oberon stepping or a revision chip.
Because given what PS5 chips exist that we have actual evidence of existence on, there's but a few ways to reach the numbers Osiris mentioned in their post:
-Full active 40CU Oberon E0 clocked @ 2600MHz (13.3)
-Full active 40CU Oberon E0 clocked @ 2550MHz (13.05)
- Oberon E0* with 48CUs clocked @ 2125MHz (13.05)
- Oberon E0* with 54CUs clocked @ 1890MHz (13.06)
- Oberon E0* with 56CUs clocked @ 1815MHz (13.009)
- Unknown Chip
Highlighted that specific option because it mirrors the butterfly setup, 18 x 3 = 54. However, the E0* ones are scenarios where E0 is a larger Oberon chip than the others. We don't actually have data yet on the CU setup for Oberon E0, so it's possible it could have additional units. After all, it had a fix for the memory controller on it as well. However, one thing working against this, is that steppings with that large a change between previous ones usually get a full revision. IIRC that usually means a different codename to go with it.
So E0 could fit the criteria the whole time and we just don't know it yet, or there's another chip on the database (by now it would absolutely be recorded onto the database, it's already early March) fitting the criteria. But if it's another chip altogether, why was E0 tested as late as December 2019? Wouldn't it actually be more hopeful that E0 is this "other chip" and has the means of providing the higher numbers sanely?
We also have to ask what chip is in the latest devkit that was supposedly sent out. The V-shaped devkit had one of the pre-E0 Oberon steppings in it, and that devkit was sent roughly around the time of E3 2019 (or shortly after). Given the timing insiders were claiming the new devkit was sent and E0's logging, I'm going to assume the latest devkit has the E0 stepping in it. We already know E0 upped the bandwidth some, but we don't know if it has additional CUs on it (being a bigger chip than the previous steppings). If it does, would such an addition still fit in giving it a stepping designation, or fall more in line with a full revision?
I mean, I'd
like to believe it, but as you say there's still too many other things of valid concern to prevent that just yet. But that's just my opinion. Either way we'll be finding out soon where PS5 actually is in terms of at least some of the major specs, and probably get additional details on XSX like the storage and memory amount/bandwidth. All leading up to E3 of course.