Some folks might be wondering where performance in their system might be lacking, and as we all know in this game of making games - creativity is greatly hampered by the amount of effort involved to see your work progress. Having a quick editing environment allows you to move quickly in order to maintain a constant creative outlet and not get bogged down by inconvenient, limitations, or overall click and wait performance experiences.
I just started working on the kit and have two fairly Apple -> Orange workstations that can help shed light on the differences one might see as well as some recommendations to consider in order to identify performance gaps of your workstation and how best to work around these limitations if component upgrades are not possible.
To start, Memory, CPU, and Disk are the core bottlenecks for a smoothly running workflow. These are the components that generally result in the stop the world, “time to make a cup of coffee” times…
an i7 workstation I have has 1 proc, 4 cores, 8 threaded with 16Gb of ram on solid state. If you run task manager and expand out processes you will see that when Unreal is cooking shaders it directly correlates to processes that have affinity to your CPU’s cores. This means the more cores you have, the more shaders you can process in parallel but with so many processors to choose from there are caveats and my two example machines describe these caveat situations.
When you have a 1 proc, 4 core, 8 threads machine you cook at a frequency of say 2.2 - 3.4 Ghz whatever your processors core frequency is rated at. Whenever you adjust textures to the terrain a chunk of the terrain has to recompile the shaders and depending on your mesh density it could be hundreds of shaders, each shader takes a few seconds to process and with 8 threads you are looking at 20-30 seconds of reduced CPU performance while those chunks are being cooked (even if you moved on to another chunk, you will see reduced responsiveness even if that chunk does not require a shader update.
A Dual Xeon workstation I have has slower cores 2.2ghz but 2 processors with 4 cores each, 16 hyperthreaded. With 16 shaders compiling in parallel even at a slower core frequency give or take the enterprise cache on the processors, it surprisingly chews through shader recompiling at double the rate. This is especially important when you are making new height maps and constantly flipping the ground mesh to find what grayscale works best.
Also for each texture type that can get applied to the ground terrain, when you blanket import “black” images to flatten TheIsland back to blank for fresh painting - you are looking at around 14,000 shaders per image and I usually import a second one once the UI starts responding again which brings me up to 28,000 cooking. This increases as you add more textures that need to blend on the same segment of the mesh. With an 8 core machine you could see 875 seconds worth of compiling and potential maxing of the 16GB memory range. With the Dual Xeon it comes out to about 400 seconds although ive not timed it but the “feeling” sounds about right with what we expect give or take the enterprise vs consumer silicon difference.
If you are CPU bound (100% cpu) then cores is what you need.
16Gb is pushing it as a minimum but its workable, and in some cases Unreal will threaten to purge if you don’t free up resources - this makes it a pain for a fully functional art workstation where I have 1.6GB photoshop documents loaded, Visual Studio for the website work, many browser windows for the documentation and forums etc. My dual Xeon has 24Gb of ram and it has not run into any problems.
When you are finally done with the major revamps that take extended full map scan cooks, then GPU is your friend where you can enable every visual effect your high end consumers will be able to see, go into immersion mode (full screen) and place your foliage, cut and carve, and paint your heart out. High frequency CPU is best for the single Unrealengine process which will in turn give you a less laggy experience, if you notice that your “smooth” tool is pausing then your not quite there - you should experience as soon as you click the terrain is smoothing and you can drag the mouse around and nicely paint with the terrain editing tools as if it was clay. (For a time I was stuck single clicking because things were so overburdened.)
Turn off Windows Defender real-time monitoring.
Watch processes in windows task manager for anything that you don’t need that consumes memory, Adobe Cloud, Autodesk, GitHub, and anything else unrelated like Curse and stuff you are not using - free the CPU cycles to be dedicated to the engine.
Disk is important for constantly updating state to disk and retrieving textures that are not cached in memory, IE when a shader recompiles the textures this is done with CPU as a pre-game asset, once its a game asset its moved into GPU so you can see it in all its glory (specular maps, lighting, and all).
So the idea is, if you are waiting - and your CPU is not bound, check Disk - if Disk is not 100% utilized (queued) then it is finally most likely the GPU that cannot keep the world in memory and keep up with computations - this can be monitored from the show stats show fps option of the viewport.
Search for High Performance CPU Benchmark graph to see where your processor ranks, in this type of workload high frequency is desirable but more cores will ultimately save you lots of time. Solid state is really the only way to go as you save out 1.6Gb PDF’s or extract the latest updates to ARK Dev Kit, load it up, save it - all of that is directly bound by the disk and your mean time to load/update/publish will be slowly drained by a lack in disk availability.
I hope some of this info helps people troubleshooting their performance problems, will try to keep the thread updated with other considerations that can improve or hamper performance so we can all “enjoy” painting on a canvas as is possible 