Agreed.

And bottlenecks can and will change depending on the game engine being used, game, the scene being rendered in the game and so on.

For example... If you were to throw Battlefield 1 on a Core i7 2600K with a Geforce 1060, you will likely be GPU bottlenecked.
But if you were to throw Ashes of the Singularity at that same system, the Core i7 will definitely be the bottleneck.

Not always though. See my prior posts in this thread.

MIPS holds a surprising amount of marketshare in modems/routers/switches/set-top boxes and so on.

It doesn't have the R&D of ARM or x86 though... And even in MIPS dominated markets, ARM is starting to take over.
My Asus DSL-AC68U has a dual-core ARM processor and 256MB of Ram whilst my previous modem a Billion 7800n had a Single Core MIPS CPU and 256MB of Ram for instance.
And it seems most modems/routers/switches/set-top boxes are slowly making the switch to ARM.

Did you read anything I wrote? Jaguar is a netbook CPU at 1.6 Ghz. If Sony used an i5 level CPU it would run circles around Cell. The other problem is that the Cell numbers are using the SPEs. The PS4 can use the GPU to perform calculations. Try to compare the puny 6 SPEs of the Cell with a few hundred GPU cores doing number crunching. That was the point that made Cell obsolete (even for HPC): GPUs simply destroy the SPEs in the massively parallel tasks both are supposed to do.

It's also insanely naive to simply compare brute floating point numbers, since this is not the only operation these CPUs are supposed to do for games. What about cache access? And specific instructions? Branch prediction (Cell doesn't have it)?

Cell is a single core CPU with 8 SPEs. Jaguar has 8 real cores. The SPEs can do numeric operations, but that's pretty much it. If i want my CPU to run an AI routine, the Cell can only deal with one thread since the SPEs are SIMD. Jaguar could run 8 threads. Basically floating point operations are the only thing Cell beats Jaguar. Any stuff that creates ramifications or recursions can't (shouldn't) be done on the SPEs. They also have limited memory access. They are simply 1% of a GPU crammed on the CPU die.

The SPEs work like GPU cores (again, SIMD). So, it doesn't help when you can simply use the GPU itself with hundreds or thousands of cores to do the operation.

Affirming that the Cell is better than Jaguar due to floating point performance is the same of assuming a Bugatti Veyron would win a F1 race because its top speed is higher than F1 cars.

Cell is only better than Jaguar at some floating point, there are different types. - And even then it will only beat the base Xbox One and Playstation 4, the Playstation 4 Pro and Xbox One X will still win in even single precision.

I'll point you to my post which has the explanation on it all.
http://gamrconnect.vgchartz.com/post.php?id=8674270

I never said anything like that...you’re putting words into my mouth. I just hinted at the fact that not everything about the Cell seems ridiculously outdated by todays‘s standards. Of course, as a whole, it is outdated by now. It’s over a decade old now.

You were simply pointing that that Cell is better in a single, extremely specific scenario, when compared to a modern low-end CPU. While ignoring that in the other 99% of the stuff a CPU has to do, even Jaguar is faster. And also ignoring that even on the PS4 devs could offload the tasks the SPEs did to part of the GPU cores and achieve much better performance.

Cell was a bad idea 10 years ago. It's not because it's outdated. It's basically a CPU with a tiny GPU-like array of cores attached to it. Bad development tools and plenty of quirks killed its HPC potential. That and the availability of Nvidia GPUs with way more cores and better dev tools.

Good post there. I would only add (for the people that think Cell is a 9-core CPU) that Cell has a single core plus 8 SPEs that are SIMD. They are not comparable to regular MIMD CPU cores and are way more limited. Like GPUs, they excel at specific tasks that are parallel.So, the correct comparison is the SPEs against GPUs, a scenario where they are massively outperformed.

The closest thing we have to Cell nowadays is the Xeon Phi line by Intel (and that's quite different anyway).

I also think people are missing the point. IBM, Sony and Toshiba reportedly invested US$ 400M on Cell development. It's insane for Sony to expend that much every gen just to R&D the CPU of their consoles when they can get a off-the-shelf part with zero research cost. Unlike Intel or AMD, Sony isn't on the CPU business, they better expend their money researching stuff that is directly connected to the console itself. Last gen they poured billions in R&D and lost tons of money. This gen, they had lower R&D costs and actually got way more money in the end.

It's ridiculous to think they will return to the strategy that was almost making them go bankrupt instead of sticking with the current approach that basically saved the company. Even MS, that was less successful, still profited with the X1. These companies are in to make money, not to try to develop magical processors that will unlock secret superpowers.

The Cell was intended to give IBM/Sony/Toshiba a dent at the HPC market, with the PS3 acting as poster-boy. But it got ridiculously clear that Nvidia got the advantage with their GPUs. Even Intel is suffering to push the Xeon Phi (that beats Tesla at task parallelism), simply because devs are already to invested on the CUDA mindset and most HPC applications have awesome libraries that support CUDA (TensorFlow, etc).

It's possible but it doesn't make sense.

THe more likely solution is to run PS3 emulation on a much superior Processor which is the Ryzen CPU. Best case scenario for the PS5 IMO we're looking at the following specs, but customized:

PS5 (2020 - 2021)
Ryzen 5 1400+ (Quad-Core CPU, with 8 threads)
10 - 15 TFLOP AMD GPU (closer to 10 $399, closer to 15 $499)
16GB of GDDR5+ / HBM2
1 TB HDD

$399 - $499

Basically a top end gaming PC from today is the range the PS5 should be in Raw performance wise if it launches in 2 - 3 years, unless there's a massive jump in the fabrication node, which is unlikely.

The performance of the CPU should be enough to run PS3 emulation, considering it's being done currently (although slowly). On top of that there isn't a massive need for PS3 emulation, considering several of the best games have already been remastered for the PS4, outside of a few key franchises like Resistance, Warhawk, Killzone, etc... As much as I enjoyed the PS3, there's hasn't been a need for me to go back to that catalog for years now. Maybe at launch, but that was 4 years ago, and it will be the same for the PS5.

It's great to have all generations on 1 platform, for the sake of being able to please peace of mind, but it's not a necessity. Having PS4 BC at launch is the most important, and having those PS4 games run based on the Pro's specs or higher is the route that makes the most sense.

I think each of the three had their own interests, as to my knowledge, Sony was planning on putting Cell on every home electronics device it was to manufacture. Linux kernel was ported on Cell, so someone was clearly planning it on HPC and servers and such.

IBM was clearly intending it for HPC. But it failed badly due to lack of performance against GPUs and terribly bad dev tools. Sony had this weird plan about home electronics, that I really never understood. If you want to pack a punch on electronics, just use an ARM SoC. In their defense, Cell was developed before the smartphone arms race made ARM SoCs improve dramatically fast.