Generally speaking, you offset processing power with storage, precalculating values. If the SSD can supplant processing power, it'll be at the expense of larger games and more data, bringing with it some other issue like download sizes and how many games can fit on the SSD.
First of all are you just purposely inflating the percentage difference now? It's 18% power difference and that's a ceiling, realistically it won't reach that much due to other factors mentioned in my previous post. And where's that 2+TF Navi2 come from? Math tells me 12.14 - 10.28 = 1.86TF. The split pool ram is obviously a cost saving decision which in turn makes it a hindrance if game graphics require over 10GB and hell I'll bet my house games a few years down the line would more than likely go past that amount. Also this minuscule difference in CPU power is pretty much non existence when you're rendering over 1800p. Again, clock speed vs more CU is load dependent and how you program your engine to take advantage of such, DF actually mentioned it in their video and you just omitted that part like you always do. Look, there's a lot we don't know but for what we do know let's not inflate those numbers to suit your narrative shall we?
You can't use redundant CUs for anything; the moment you use something, it's no longer redundant. Tempest was engineered and included deliberately. It remains to be seen where it fits into the SOC, whether part of the GPU or isolated as an independent unit. Though given it has its own controller and memory accessing, I expect the latter, a stand-alone audio processor that happens to be derived from an RDNA CU as opposed to derived from a Tensilica DSP or Cell SPU or bespoke ground-up design.
I can only go on what has been reported, and as far as I know, they dont make GPUs with 41 Cu's?
Nothing, because they're redundant - I don't think it works like you're suggesting but I'm low down on the tech food chain here (bottom rung
)Well that's my question - how to they know? Devs, magic or another solution?
No, I'm suggesting that RDNA2 didn't exist when Github was run - I believe (and could be completely wrong) that it was not RDNA2.
You're right but wrong (kind of) as Cerny said 'PS5 CU <>= PS4 CU, and to be more accurate I think it was 36CU = 58CU so over 50% better...so whilst the XSX is 12TF it's acting nearer 20TF - or rather over 3x the Xbox One X. So in 3 years if you double the power then that's significant power to RT and shoring up performance that I for one would buy.
lol, it did seem odd and fanboyish which is why I said what I said about trolling...so apologies back.
Sorry, I tend to dip in and out and often stuff goes over my head to I miss bits out. I don't know if I could find it either, if you could link? But essentially I assume you're saying that although the different speeds are for different types of work, it doesn't affect devs workloads?
My thought...I would have thought it's 'just' an extra bit on the chio as apposed to 1 of the assumed 4 disused CUs being repurposed.
I think it just means they took a CU and reworked it into a new design.
Never use a second hand source when the first hand source is available:
The hardware architecture would have had all or most of the capability from scratch. Technically, the current gen did as well, although Jaguar's architectural turbo feature was mostly unused until a new revision on a different process.
The Tempest CU is described as not having caches, so that would bypass the caches used by the rest of the architecture. It's not clear where the unit is in the overall system, so whether it is covered by the cache scrubbers mentioned for the disk I/O block isn't clear. The SPU-model does well with getting high vector utilization out of highly regular streaming workloads. The motivation is to combat the memory wall with explicit control of all memory access, which in algorithms where the access patterns are regular or predictable can be more efficient than relying on the standard CU's dependence on heavy concurrency over a long-latency memory pipeline.
MSFT also handed them PR win after win before the current gen launch on top of the 2/3 GPU performance. That's not happening this time, quite the opposite.
I agree with your point that you don't need the best hardware. That said PS4 did have the best hardware, and PS2's competition wasn't that serious. I also wonder how well the PS4 would have done (at least in US/UK) had MSFT not made so many mistakes.
