The Apple M4 has a faster GPU than the Xbox Series S

Posted by TechExpert2910@reddit | hardware | View on Reddit | 17 comments

**➡️ Here are some modern GPUs' FP32 TFLOPs (a measure of performance):** Apple A10X (iPad Pro 2017) - 0.76 Apple A12X (iPad Pro 2018) - 1 **Xbox One S - 1.4** **PS4 - 1.8** Apple A17 Pro (iPhone 15 Pro) - 2.1 Apple M1 - 2.6 Apple M2 - 3.6 **Xbox Series S - 4** **M4 (iPad Pro 2024) - 4.32** Interestingly, Apple's M series SoC's TFLOPs are comparable to AMD's latest GPU architecture (the consoles are based on older AMD GPU architecture, so each Apple M teraflop can lead to more real-world performance than those consoles)! The M4 also has neural-engine-enhanced Metal FX upscaling and hardware support for mesh shading, which modern consoles lack. Ray tracing acceleration too! And it actually shows - right now, games like COD Wardone Mobile and Genshin do look just like their modern console counterparts at max settings, high iPad resolutions, and 60-120 FPS. This is how I calculated the M4 TFLOPs: The M4 has 22% higher Antutu GPU scores and 18% higher Geekbench 6 Metal scores compared to the M2. Adding 20% to the M2's known 3.6 TFLOP number, we get 4.32 TFLOPs.

17 Comments

[-]

CatalyticDragon@reddit

>Apple M teraflop leads to much more real-world performance than the 8th generation consoles How exactly are we calculating the occupancy for these devices to make this claim? >The M4 also has neural-engine-enhanced Metal FX upscaling 1. Everything has upscaling. 2. Consoles have access to "neural engine" based upscaling too. XeSS being one example (not that anybody wants to implement it). 3. Metal FX upscaling is [AMD's FSR](https://www.notebookcheck.net/Apple-s-Metal-FX-upscaling-is-powered-by-AMD-FidelityFX-Super-Resolution-technology.784429.0.html) which is already in many console games. >and hardware support for mesh shading, which modern consoles lack Nope. The [Xbox X&S ](https://learn.microsoft.com/en-us/gaming/gdk/_content/gc/graphics/newfeatures/ms/ms-scarlett)has them. PS5 also where is is called "primitive shaders". >Ray tracing acceleration too! Just like a console, or even the Steamdeck. >games like COD Wardone Mobile and Genshin do look just like their modern console counterparts They should. These are not complex games. They should run well on a $1000 state of the art modern SoC. >Adding 20% to the M2's known 3.6 TFLOP number, we get 4.32 TFLOPs Apple says the M4's GPU is "up to 4x faster than M2 (3.6 TFLOPS)" which means 14.4 peak TF. That's roughly around RTX 4060 (15 TF) level. But real world performance involves more than just theoretical TFLOPs. Memory is crucial and the M4 has 120GB/s vs 448 on a PS5 and 560GB/s on XboxX. Even the 4060 more than doubles it with 272.0 GB/s.

Anotheeeeeeant@reddit

They don’t though, some of his arguments is a lot wrong. COD Warzone mobile at peak graphics looks worse than pc on its lowest settings. I actually compared the 2. [https://www.reddit.com/r/CODWarzone/comments/1bk9he7/warzone\_mobile\_vs\_warzone\_pc/#lightbox](https://www.reddit.com/r/CODWarzone/comments/1bk9he7/warzone_mobile_vs_warzone_pc/#lightbox)

IntrinsicStarvation@reddit

>How exactly are we calculating the occupancy for these devices to make this claim? I would like to see this too.

From-UoM@reddit

MetalFX uses Fsr 1.0 for older intel based systems. On M series chips its different. The Neutral core shows usage when you enable it on M series chips

jasswolf@reddit

MetalFX upscaling appears to use FSR 1.0 as part of its initial base before an ML pass, but it's not *just* FSR. That being said, FSR seems to be set to get some sort of ML-accelerated component with the release of the RX8000 series.

twhite1195@reddit

And the iPad pro is like $1000 and released this year and the series S is $300 and released 4 years ago...like... Of course technology improves

CMDR_Supagoat@reddit

It’s not just that, though. The ipad used so much less power.

Bderken@reddit

Why are you being negative, it’s nice to see exactly how technology is improving by somewhat quantifying it.

juhotuho10@reddit

TFlops aren't comparable between GPU architectures, TFlops and performance very often don't even correlate all that well

They absolutely are. A flop is a flop, this is an internet myth. The ACTUAL information it is telephone gaming is: Nothing ever actually gets to its peak theoretical performance. Ever. Peak theoretical performance <> real world sustainable performance. Different architectures can get closer to their peak theoretical numbers than others. If two systems with the same peak theoretical performance are compared, and one is performing better, the flops are exactly the same flop for flop, one system just isn't actually performing as many flops as the other.

This is awesome. Just wish we could play console games on an iPad, natively.

SkruitDealer@reddit

That's the main problem with iPad and many Apple products. All aspiration, but the garden of Apple strictly gates software that would unlock their magnificent hardware. Feels like they built a fighter jet that can only travel on railroads - "for your own safety" and "quality". In the end, it's just a sinful waste of top-tier, short-supply manufacturing capabilities. It is downright abusive to their customers and directly contributes to wastefulness. With an M4 in an iPad, and Apple's full control over developers (everything is ARM compatible at this point), I shouldn't need an iPhone and a MacBook to access all the software available in the garden. It's such a fucking scam.

And ROG Ally is over 2x than the series s and almoat as powerful as the Ps5 with its 8.6 tflops Roght everyone? /s

OverlyOptimisticNerd@reddit

Asus is advertising FP16 flops to inflate the number. Also, it’s based on max sustained clock rate and full performance of the GPU, which it cannot do with a max sustained 30W (desktop equivalent APU is 65W and up, the 8700G). The Ally is more powerful than the Steam Deck (1.6tflops) but generally is only 20-50% faster at 30W than the Steam Deck at 15W. The Ally probably has a real world FP32 of ~2.5tflops.

No it is Fp32. YOU CANNOT COMPARE FP32 BETWEEN DIFFERENT ARCHITECTURES. For example rdna2 RX 6800xt of 20.74 fp32 tflops is significantly faster than thr rdna3 RX 7600 of 21.75 The rx 7800xt of fp32 37.32 tflops is slower than the RTX 4070S of fp32 35.48 tflops

SeniorFoodmaxxer@reddit

And it’s going to throttle twice as fast. What’s the point of these metrics, again?

It very well should have a faster GPU. It costs three times more and is three and a half years newer. You know what else has a faster GPU than the Xbox Series S? The Xbox Series X!

Reply to Post

17 Comments