What does hardware acceleration accelerate ?

[cgidesign]

I found many threads about hardware acceleration issues but none explaining what is accelerated if the feature works.

E.g.

• brush strokes are processed faster
• live filters are processed faster
• adjustments layers are processed faster
• ...

Has Serif posted such information somewhere?

[GarryP]

Go to the Help for your Affinity application, type hardware into the search box, and follow the links.

It’s unlikely that Serif will give more-detailed information about how their software works internally as that could benefit competitors.

Basically: Turn it on; see if the stuff you are doing works faster/better than when it’s off; and if not, then turn it off again. (Or the reverse if it's already on.)

[James Ritson]

Hi @cgidesign, hardware acceleration is used for all raster operations within Photo—that includes basic layer transforms through to filter/live filter and adjustment layer compositing. In cases where a GPU with strong compute performance is present, the performance difference can be huge. Take M1 Max for example: the CPU is no slouch, and scores very highly on our vector benchmark, but raster performance is only around 1000 points. In comparison, raster performance on the GPU is 30,000. That's a linear scale, so to say it's 30x faster than CPU is no exaggeration.

The fact that all major raster operations are accelerated is a huge undertaking, and means it's not really appropriate to compare our hardware acceleration implementation with what other software has. I understand there's a lot of frustration around the choice to disable support for AMD's Navi cards, but Photo compiles new kernels every time you add an adjustment, filter, use a tool etc for the first time during the session. It was simply unusable on Navi because the kernel loading is unacceptably slow, to the point where it could take up to a minute for a RAW file to be initially processed and displayed in the Develop Persona (multiple kernels are loaded simultaneously in this scenario).

If you start to introduce vector operations to your workflow, such as text, vector shapes, poly curves and so on, this is where the GPU must copy memory back to the CPU, and this is where most bottlenecks occur—unified memory architecture (e.g. Apple Silicon) manages to reduce this penalty somewhat, but as of this current time it's still not as fast as compositing purely on the GPU.

The Benchmark (Help>Benchmark) will give you quite a clear idea of the performance improvement you can expect. In particular, focus on the Raster (Multi CPU), Raster (Single GPU) and Combined (Single GPU) scores. They will indicate CPU compositing performance, GPU compositing performance and shared performance (copying memory between them when using a mix of raster/vector operations) respectively.

Hope that helps!

[cgidesign]

Thank you both. I found the information in the help menue.

Unfortunately still not working very well (same as my first try a long time ago).

Tried with a big airbrush in a pixel layer on top of various adjustments and other pixel layers. After a few quick strokes APh freeses and i have to wait for several seconds until it recovers. So, no hadware acceleration for me 😪

Hoping for Version 2 now.

Notebook: Nvidia RTX 2070 (8GB vram / used 3.8 GB), core i7 tenth gen, 16 GB ram.

[lacerto]

Obsolete.

[Roland Schwarz]

That doesn't explain why brushes in a picture 25000px x1600px in open GL are working properly but with metal acceleration settings lagging at the beginning of every stroke. In the contrary. After this explanation it makes even less sense. But anyway - thanks a lot for the explanation! And thanks to the serif team for great apps!

[cgidesign]

39 minutes ago, Roland Schwarz said:

thanks a lot for the explanation! And thanks to the serif team for great apps!

Same from me. Asking for optimization in my case means, I am using the products 🙂. I am not posting in the adobe forum anymore because I replaced their products with affinity (and a whole bunch of others).

[PaoloT]

On a 12-core MacPro 2013, with two graphic cards, the benchmark is really impressive. And I've just the basic GPUs!

Paolo