Soundwave said:
The Switch is really not that underpowered of a system, it has 1/3 the teraflop performance of an XBox One, that's absurdly powerful, a PSP was no where close to 1/3 of a PS3 (and well below even a PS2) and the PSP was looked at a as technological wonder for its time (ditto for Vita, no where close to 1/3 of a PS4, not even PS3 performance). |
Teraflops isn't everything or a way to determine performance.
I.E. You can have a GPU with MORE teraflops be slower than a GPU with LESS teraflops.
I.E. The Radeon 7850 having almost 1 Teraflop less than a Radeon 5870 is faster.
I have already embarrassingly proven you wrong on this before, but you seem to keep regurgitating the same false information... But here we go again!
https://www.anandtech.com/bench/product/1062?vs=1076
Radeon 7850 is 1,761Gflop, the Radeon 5870 is 2,720Gflop, 7850 wins... And unlike your Comparison both chips are AMD.
But in the case of the Switch... In it's slowest 307Mhz mode it's only a paltry 157Gflop verses the Xbox One 1,310 GFLOP.
That is a 8.3x difference in the Xbox One's favour.
At it's Max clock (I.E. Docked, the mode that NO developer targets games for, they target Handheld mode, but you cherry picked docked mode.) the Switch is 393Gflop.
Meaning when we compare just single precision floating point, it is 1/3rd the capability of an Xbox One.
But in the real world... We are NOT seeing a 2.5x performance uplift going from Handheld to Docked, games don't perform 2.5x better, games don't look 2.5x better.
So we can surmise that:
1) Flops are bullshit in trying to tell us performance and capability.
2) The Switch still has bottlenecks and is held back by it's CPU, Memory, Integer and Storage performance.
The Switch is also based on nVidia Maxwell, Xbox One is based on Graphics Core Next... Everyone and their dog knows that Maxwell isn't very good at asynchronous compute like Graphics Core Next, so in number-crunching "Gflop" capacity, it's never going to get near an equivalent Graphics Core Next part in the real world, let alone one that is orders of magnitude faster.
But it also doesn't have to.
See... The other issue with your usage of Gflops is that... It doesn't tell the entire story, it's a mobile chip.
One GREAT way to save power and increase performance in mobile environments is to actually reduce the rendering precision and leverage FP16 math, which is a very common tactic developers use.
Meaning your "Gflops metric" is useless as it's literally not even used. Not single precision anyway.
Basically what it means is that, when two compatible (Must note the: Compatible!) instructions are used, the Maxwell GPU will combine them both and execute them on the FP32 compute engines, theoretically doubling floating point performance.
Obviously it doesn't and cannot scale like that in the real world.
But it also brings the Switch's base Half Precision floating point up to 314Gflop and a max of 786Gflop.
But these are theoretical numbers, not real world and not representative of actual performance... It's literally a by-product of multiplying Clock Speed, Instructions per clock and functional units and not an actual real-world benchmarked number.
Soundwave said:
A Switch can handle many modern PS4/XB1 era games, if it was a bit more powerful it would just flat out run I think almost all PS4 games. |
Lol. No.
Soundwave said:
For a handheld device that can literally run PS4 era games like DOOM Eternal, Fortnite, Overwatch 2, Wolfenstein, NBA 2K, FC Soccer, Mortal Kombat 11, Batman: Arkham, Witcher III, etc. |
All those games come with compromises. Often rather significant compromises.
Soundwave said:
The thing is the Mariko based Switch systems really actually could be clocked higher to get more like 600 GFLOPS docked performance (with no other upgrade required, this is just the result of the die shrink), that would be able to run pretty much any PS4 era game at a pretty reasonable level if Nintendo had allowed that. |
If the Switch leveraged Pascal rather than Maxwell, it could have clocked 50% higher on launch.
Again, your usage of Gflops doesn't tell the entire story as it's a theoretical denominator and not real world performance.
It would never touch an Xbox One or Playstation 4 because it's missing several key ingredients:
1) Ram capacity. 4GB with 1GB of that stolen for the OS, leaving developers with only 3GB is just not enough.
2) Ram Bandwidth. - 25GB/s even with Delta Colour compression is not going to provide the necessary fillrate for lots of alpha effects, resolution or quality Anti-Aliasing.
3) CPU performance. - ARM A57, 4-Cores with 1 core stolen for the OS, clocked at 1Ghz is simply not enough performance for demanding decompression of meshes and textures and support streaming of high-quality assets into the GPU in real time.
Just increasing "Gflops" is useless when literally every aspect of the console is holding it back, creating bottlenecks.
Soundwave said:
The Switch is a lot closer to a XB1 or PS4 than people think, if Nintendo let devs actually utilize the 600 GFLOPS a Mariko/Lite/OLED models can hit, they'd basically be able to run any PS4 game fairly well at that but even at the OG spec the fact that it can run things like DOOM Eternal and Dragon Quest XI is bonkers. |
The original Switch chip is capable of 1 Terfaflop when using half precision/rapid packed math.
Nintendo made hardware and software design choices to limit it, those who have cracked and modded their consoles and unlocked their clocks can game at 924Mhz or 473Gflop/946Gflop HP.
Still doesn't get near a Playstation 4. It never will. Even if it was 4 Teraflops.
Games perform better, sure. But often they still perform and look worse than the Playstation 4 and Xbox One version.
https://www.eurogamer.net/digitalfoundry-2019-switch-overclocking-analysis
The Switch's Bottleneck is simply more than just Gigaflops that you desperately cling to.
Soundwave said:
A Vita wasn't really anywhere close to a 360/PS3 though. It's 28 gigaflops, and XBox 360 is 250 gigaflops ... almost a 10x gap.
The Switch is 393 gigaflops docked, the XBox One is 1.2 teraflops, that's only about a 1/3 gap.
So a Switch is much closer to the actual home consoles it launched against (PS4/XB1) than a Vita was (versus PS3/360) even though Vita was somewhat impressive hardware.
This is why the Switch can run actual modern-gen titles like DOOM Eternal, Hellblade, Witcher 3, Persona 5, Fortnite, Overwatch 2, Dragon Quest XI, FC Soccer, while a Vita was not really getting many demanding games from the PS3/360. |
The Vita, Playstation 3+Xbox 360 were completely different from an architectural and technological perspective, architecture matters more than Gflops.
Just like Architecture matters more than "Bits". - Hence why consoles haven't moved from 64bit in decades... And the main reason for 64bit was to address more than 4GB of memory.
The Switch is much slower than 1/3rd of an Xbox One in the real world.
Yes it's running Doom, Hellblade, Witcher and more... But with significant compromises. And they ARE significant.
Still a great experience, but no way are they even remotely comparable releases.
Games like Overwatch and Fortnite are designed to scale to every single possible hardware combination in existence for mass-appeal and market... And thus have opted for clean and simplistic textures and geometry as well as baked lighting and shadowing to enable that scaling.
Games like Harry Potter showcase how limited the Switch really is, it simply doesn't have the memory.
Mortal Kombat 1 for instance was a bit of a joke on Switch.