Not Apple though. Apple just hand waves the numbers.

Apple does the thing with screens where they cycle between more shatter resistant and more scratch resistant and tout "3x more shatter/scratch resistant than the previous model"

My apologies for bringing insane US politics into a non-political post but I saw an example of this just today.

Don't forget "up to..."

False. They also stick fake numbers in there with zero regard for consistency.

just keep adding decimal places until you see a difference

I also love meaningless things like

"3x more effective at remowing plaques" or "10x kore tougher on stains". Yeah sure buddy it is...

very odd its not going 3.18, 3.211115, 3.222224, 3.233333 must be before the ipo announcement

It depends on what are you are trying to compare

YearFlagship ModelStandard FP32 Compute Power 2008NVIDIA GTX 280~0.9 TFLOPS

2010NVIDIA GTX 580~1.5 TFLOPS

2012NVIDIA GTX 680~3.0 TFLOPS

2014NVIDIA GTX 980~4.6 TFLOPS

2016NVIDIA GTX 1080 Ti~11.3 TFLOPS

2018NVIDIA RTX 2080 Ti~13.4 TFLOPS

2020NVIDIA RTX 3090~35.5 TFLOPS

2022NVIDIA RTX 4090~82.6 TFLOPS

2025NVIDIA RTX 5090~104.8 TFLOPS

I guess a good way to view it is, a stand mixer, a hand mixer and a whisk. All have purposes.. but you have to be using the right tool for the right workload to get the best performance.

It isn't better to scramble 6 eggs with a stand mixer before you put them in the pan, just use a fork. Doesn't mean the stand mixer won't blow the fork away when making ground beef.

Outdated. They stopped labeling the axes or for GPUs they label it some bullshit like "relative performance" so they can claim 16x performance with upscaling and slop-generated frames.

I remember when people got upset by things like this. I wonder why it's "ok" now.

Moore's Law is struggling against the limits of what's physically possible and the days of massive jumps in performance each generation are long over. It's hard to advertise a new CPU when it's running at nearly the same clock speed models ran at ten years ago and all you've done is thrown more L3 cache at an otherwise 99% unchanged previous model.

Even the number of cores has stagnated, and that was their Hail Mary to keep the performance gains coming. And I bet the Spectre/Meltdown mitigations eroded a large part of those gains too - IIRC existing chips lost ~12% performance after the firmware patches, and new ones had to be redesigned to handle speculative execution in a less efficient (but more secure) way.

I made a chart like this once, it was to encourage my wife durong the start of her weight loss journey