What hardware you need

PxlMonk is built to work on ordinary machines, and the honest answer to “what do I need” depends on one thing above all: how many megapixels your camera has. A 24 megapixel file and a 100 megapixel file are the same work per pixel and four times the pixels.

Everything on this page is measured on real files rather than estimated.

The short answer

If you only read one line

For a 24 megapixel camera, 8 GB of memory and four processor cores are enough. A graphics card changes nothing about developing photos — it only speeds up the AI features.

Your camera	Memory: works	Memory: comfortable	Recommended
Up to 24 MP	2.5 GB	4 GB	8 GB
45–61 MP (most high-resolution full-frame bodies)	3–4 GB	7–10 GB	16 GB
100 MP (medium format)	5.5 GB	16 GB	32 GB
150 MP (the largest backs)	8 GB	16 GB	32 GB

The two memory columns are different questions, and both are worth knowing.

“Works” is what one photograph needs to open and develop at all — measured at the moment of opening, which costs more than having the photograph open does. A 150 megapixel file settles at 4.6 GB but passes through 7.9 on the way in, and it is that larger figure the table gives, because it is the one that decides whether the file opens. “Comfortable” is what PxlMonk needs before it can keep its working buffers in reserve instead of rebuilding them for every change. Below that second figure everything still works, but a full-resolution pass — an export, or the moment a noise-reduction slider settles — takes roughly twice as long.

That is why the 100 MP row recommends 32 GB even though 16 GB is enough to open the file. On a 16 GB machine a 100 megapixel photograph clears the comfortable threshold by about one percent — it works, and it has nothing to spare.

If a photograph is too large for the machine it is on, PxlMonk says so before it starts work rather than after: it knows the pixel count from the catalogue, so it can tell you the file needs more memory than there is instead of spending twenty seconds finding out. The message names the size and what is available. Nothing is wrong with such a file — it opens on a bigger machine.

Processor

Four cores is the useful minimum, and eight is where extra cores stop paying. Developing is spread across every core you have, but the gain flattens out: going from four cores to eight is worth having, and going from eight to sixteen is barely noticeable in the preview.

The exception is the AI features on a machine with no graphics acceleration — see below. There, cores keep helping well past eight, because the work is one long calculation rather than a picture divided into strips.

Graphics card

A graphics card does not speed up developing. Exposure, colour, tone, masks, sharpening, lens corrections and export all run on the processor, deliberately — the preview is designed to stay quick on a laptop without one.

What it does speed up is the AI features: AI noise reduction, generative retouching and face detection. There the difference is not subtle. On the processor alone, denoising a 24 megapixel photograph takes several minutes. With graphics acceleration it takes well under a minute.

If you never use those features, you do not need a graphics card at all. If you use AI noise reduction regularly, it is the single component that changes your day.

What your operating system changes

Memory and processor recommendations are the same everywhere. Graphics acceleration is not.

Windows

Graphics acceleration: yes, automatic. PxlMonk uses DirectML, which works with any reasonably modern graphics card — AMD, Intel or NVIDIA — as long as it supports DirectX 12. Nothing to install and nothing to switch on.

If your machine has no suitable card, the AI features quietly fall back to the processor and still produce exactly the same picture, just slowly.

The amount of graphics memory is not what limits you: the AI models work on small tiles of the picture at a time, so their memory use does not grow with the size of your photographs.

macOS

Graphics acceleration: yes, automatic. PxlMonk uses CoreML, so both Apple Silicon and Intel Macs with a supported graphics chip accelerate the AI features without any setup.

One thing worth knowing on Apple Silicon: the graphics chip shares the same memory as everything else. The memory table above already accounts for developing, but if you run AI noise reduction on a very large photograph on an 8 GB machine, that is the case where the two compete.

Linux

Graphics acceleration: no. The AI features run on the processor, so AI noise reduction on a large photograph takes several minutes rather than under one.

This is a deliberate decision rather than something missing: the options that exist would either restrict PxlMonk to one manufacturer’s cards, more than double the download for everybody, or require you to set up a separate runtime by hand. None of those seemed a fair trade for a feature many people never touch.

Everything else — developing, export, tethering, the film workflow — runs exactly as it does elsewhere. If you use the AI features often, more processor cores are the thing to spend on, since that is what carries the work here.

One further Linux-specific point that is not about speed: your desktop has to tell the system which colour profile your monitor uses, or PxlMonk cannot show colours accurately. See Colour management on Linux.

Storage

Nothing exotic is required, but two things are worth knowing.

RAW files are read in full every time a photograph is opened for the first time in a session, so a solid-state drive makes opening photographs noticeably quicker than a spinning disk — this is the one place where storage speed is felt directly.

PxlMonk never modifies your original files. Your library needs room for the originals, plus a small amount for thumbnails and the catalogue itself.