what's random? ps3 cpu is still above ps4 in raw numbers, that's why i said architecture aside.

It's like saying that a Pentium D @ 3.73ghz is "above" any Core i3 2xxx or above CPU.  Even at half the mhz, it wrecks the older chip in every way.

He is right, PS3 CPU was beast and in some tasks its stronger than PS4 Jaguar CPU if comparing core per core. People dont realise but PS4/XB1 Jaguar CPU is not strong and it's actually bottleneck for PS4/XB1 hardware.

Point that Wii U CPU and Switch CPU have 1GHz don't mean nothing because you comparing basically upgraded GameCube CPU tech/architecture with ARM tech/architecture from 2014, on same clock A57 is more capable and more stronger even PS4 Jaguar CPU, and its more capable than Xbox 360 even on 1GHz.

Here you have comparison of AMD 4-core Jaguar and ARM 4-core A15, have on mind that A57 is 4 years newer CPU and much more capable than ARM A15.

https://www.notebookcheck.net/SoC-Shootout-x86-vs-ARM.99496.0.html

There are some benchmark in which the PS3 CPU actually does a better job than the PS4 CPU.

http://www.cinemablend.com/games/PS4-Clobbers-Xbox-One-Ubisoft-GPU-Benchmark-67877.html

What that benchmark tells us is that the Cell's architecture was designed to have certain special purpose features (similar to that of a GPU.) For general-purpose tasks (which is what CPU's are designed for) the Jaguar is very much the superior processor. The Cell (and to a lesser extent Xenon) were designed to maximize floating point operation performance (flops) at the cost of pretty much everything else, it makes sense that it would somewhat outperform the Jaguar when it comes to GPU Benchmarks. 

Yes.

Finally a poster that speaks my language and is accurate.

This is probably the most accurate way of guaging performance using a singular metric as it is based on Drhystone MIPS.

And the good part about using DMIPS is that you are pegging it against a reference machine, not using theoretical numbers plucked out of thin air that the chipset may never even remotely reach.

That's not to say it's going to be a perfect comparison though. But it gives a good guestimate.

No.
Cell exceeds the Playstation 4 in THEORETICAL numbers. That's Flops my dear watson. And is not representative of any real world performance.

The Playstation 4's CPU is indeed superior to Cell.
The games speak for themselves anyway.

This. Megahertz and Core counts isn't everything.

Based on what metric?

Because the games say otherwise. In general, Playstation 4 and Xbox One games have significantly more A.I characters and Multiplayer matches with significantly more players.

Sadly they didn't elaborate on the testing conditions. It is likely single precision floating point heavy, leveraging iterative refinement on the Playstation 3. Games tend to be built using more than that.

Throw double precision, integer and the Playstation 4, Xbox One and Switch CPU's would obliterate the Cell.

What they are trying to sell you on in that article is that the GPU is well suited to things like Cloth Simulation, Physics etc'. Which is blatantly obvious.

The Cell has a ton of architectural hindrance's that limit it's potential.

It favours bandwidth over latency for one, SPE's can only access it's own scratchpad memory and not the memory of other SPE's.
The SPE's are also only dual-issue, in-order designs, there is no branch predictors and will thus heavily rely on compilers.

In essence, the Cell falters compared to proper CPU designs in many tasks... And for Floating Point Heavy scenario's, a GPU is better anyway.

Sony advertised the Cell as some kind of powerhouse and made people believe it was some kind of super computer in a chip, the Cell was a transistor conservative design on an older fabrication process to offer good performance at a low price.

Are you seriously comparing games that were made for a split 256/256MByte memory layout with games that are made for a 8GByte unified memory layout?

The SPU's can do everything a "normal" cpu can do. They are as fast on integer as they are on fp. They are just extremely simple or "dumb" in comparison to a modern cpu core (In-Order, no branch prediction, etc...). That's what's making them so fast and so hard to code for. You can (partly) optimize against these "flaws", but this takes time and is worsen by the fact that your normal pc codebase doesn't care about them.

You can't really say "this" is better then "that" without taking in to account difficulty to code for. It doesn't matter if one is 50% "stronger" than another if the bulk of developers can't write code well enough to show that advantage.

In Short Cell Is weaker to Kabini because of generation gap. But still you cannot compare them.

Well, it is kinda hard to come along real numbers, as console CPUs are usually only used in console and have no benchmark-runs. It is easiest for the Cortex, because it is a widely used CPU besides the Switch. For instance here you have results (in comparison to A15) for 3 major benchmark-suites (SPECint2000, Geekbench Integer and Geekbench Floating Point):

http://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/6

The Cell was used in supercomputers too, so I see some Linpack-data float around, but that confuses me actually. In Supercomputers you use hundreds or even thousands of CPUs at once, you would have to calculate back to the single CPU. However, here you have an article that uses Linpack and some other tests:

https://www.ibm.com/developerworks/power/library/pa-cellperf/

I found nothing on the other CPUs really. But XBox 360 and WiiU both use PowerPC-design (which is also the base of the Cell).

Sadly you can't compare the GFlops in Geekbench Floating Point and Linpack, as Linpack uses just pretty simple operations (matrix-multiplications and stuff like that) while Geekbench uses Real World algorithms (Sharpen, Blur, Mandelbrot). If you want, you can see that the Cell-BE performs in Linpack between 9.67GFlops (1000x1000) to 200 GFlops (Matrix multiplication), while the Cortex does in Geekbench between 2 GFlops (DGEMM) and 6 GFlops (SGEMM). I ignored the ST-results, because that means Single Thread - only one core is used.

This points toward a bigger floating point performance of the Cell (but as I said, simpler operations in Linpack, for real comparison you would use the same benchmark). But this is not surprising, as the Cell was also used for graphics calculations (which are using floating point), while the Cortex in the Switch does the main processing and the GPU calculates the graphics. And the Tegra X1 has a more powerful GPU than the RSX.

I wish all console-makers would do a SPECsuite-benchmark and release the results, these would bring comparable numbers. But they'll never do that.

Do you have a source for that?