Hacker News Reader

Comment by throw10920 | Hacker News Reader

Not to undermine your point at all, but as a tangent, you might be surprised by how powerful phone compute is getting these days.

The iPhone 15 Pro approaches the multi-threaded GeekBench score of my Ryzen 3700x (which, while not new or high-end, is still better than what 90% of desktop users have) and handily beats its single-threaded score.

I agree that the difference in smoothness is due to software design decisions and not computational power, but thought it might be novel to see how the gap between nubile and desktop has shrunk over the years!

View on HN

Well, the devices I compare are a seven year old iPhone X, which has 20% of the performance of latest iPhone (in Metal API, at least), a desktop system which's equipped with a Radeon RX550 with 4GB of RAM, and an Apple MacBook Air M1.

Both of these systems are infinitely more powerful than my phone, and the latest iPhone possibly smoke that old AMD card, but not the M1, not by a large margin, at least.

Isn't the most important difference that the phone has a smaller screen?

That smaller screen on iPhone X has 2436x1125 resolution, which is only 33% less pixels than my desktop monitor (which is 2560x1440). Considering that the video plays full screen on the phone, and in a much smaller window on my desktop system, the inferior GPU can render that much more smoother than the desktop, which shows that GPU is not the problem here.

Also, grabbing the scrollbar on Firefox Linux restores the smoothness, supporting "tick" theory.

> Not to undermine your point at all, but as a tangent, you might be surprised by how powerful phone compute is getting these days.

... some phones, at least. Not everyone's on a state of the art flagship device.

Much more importantly for ARM phones video isn't decoded in the CPU at all, the SOC has video codecs built in. This allows things like all day netflix/hbo without draining the battery. AMD cpus don't have video codec coprocessors as far as I know. Not sure about GPUs though...

All of the big three GPU vendors (Intel, AMD, NVIDIA) support hardware video decoding and encoding onboard their GPUs for a very long time.

Intel and NVIDIA generally adds dedicated silicon for these tasks, and AMD prefers to use shaders for video tasks. They started doing shader based video processing in Radeon 8500 days, and they're very experienced in video processing, since they had AiO wonder cards which did also capture and even TV in some models. Trying to render analog signals nicely on a PC screen provided them considerable R&D and know how during that journey.

Correction: AMD has dedicated video encoder DSP called VCE[0].

[0]: https://en.wikipedia.org/wiki/Video_Coding_Engine

Be wary comparing benchmark scores between platforms. Different compilers for different platforms will make a massive difference to scores achieved, mostly due to the fact benchmarks typically do 'useless' work that compilers aim to detect and eliminate.

The compilers for iOS and OSX are largely the same and if you’re consistently using the same compiler family you’re probably ok even if the version isn’t completely identical (major LLVM versions aren’t frequently bringing massive changes to how the optimizer is working). Additionally the UB optimization stuff you’re referencing is all within a tight margin even across families (ie GCC and clang tend to apply very similar optimizations).

Reply on news.ycombinator.com