Big cunvas + big brushes = lag in Krita

[Grum999]

@Lesqwe56

some longer random strokes will consume 35~50% cpu and 7~11gb ram during operation (quite slow).

16k x 16k use around ~975MB per layer in RGBA/8bits
Not a big deal when you have a lot of memory, but can be quickly a problem I think

Also tried it on my side, on fresh created document it’s smooth with most brushes at 1000px; didn’t tried above.
I get some lags with more complexes brushes.

Also, hardware is very important; 24CPU doesn’t mean that it will be faster.

Here some posts about case I’ve got:

[Processor] Intel vs AMD, Raw Power x More cores?

Hi After 8 years of services, I finally changed my old Athlon for a new one. Christmas gift from last 2 years Works Council and annual bonus from last 2 years put aside, I was able to buy an AMD Ryzen 9 3900X So I’ve executed benchmarks and I’m a little bit surprised. ******** Start testing of FreehandStrokeBenchmark ********* Config: Using QtTest library 5.11.3, Qt 5.11.3 (x86_64-little_endian-lp64 shared (dynamic) release build; by GCC 8.3.0) PASS : FreehandStrokeBenchmark::init…

[Processor] Intel vs AMD, Raw Power x More cores?

Fixed DDR4-3600 frequency configuration in UEFI Performances are slightly better. ********* Start testing of FreehandStrokeBenchmark ********* Config: Using QtTest library 5.11.3, Qt 5.11.3 (x86_64-little_endian-lp64 shared (dynamic) release build; by GCC 8.3.0) PASS : FreehandStrokeBenchmark::initTestCase() QWARN : FreehandStrokeBenchmark::testDefaultTip() KoColor debug runtime checks are active. QSocketNotifier: Invalid socket 8 and type 'Read', disabling... QDEBUG : FreehandStrokeBenchma…

In synthesis when you have a lot of CPU, you also need to have a very fast memory otherwise there will be a bottleneck and finally you’ll get worst performances than using less CPU.
On my computer optimal number of CPU if 22 of 24…

Grum999

[Lesqwe56]

@Grum999

16k x 16k use around ~975MB per layer in RGBA/8bits

It fallback to near that size when the operation complete, but still need say, 6~10GB headroom for stroke calculation.

@Grum999

you also need to have a very fast memory otherwise there will be a bottleneck and finally you’ll get worst performances than using less CPU

Yah, while I use DDR5-4800 dual channel, the performance is similar when set krita to 16 threads and 32 threads, which are far from saturated.

[Grum999]

@xMaxrayx

this program always lag? especially with big brushes.

Can you provide your computer characteristics?

CPU (model, cores), RAM (DDR model, amount), SSD transfer speed, …

And also some current Kirta settings like:

• General > Cursor
• General > File handling
• General > Miscellaneous
• Display > Canvas acceleration
• Performances > General
• Performances > Advanced

Grum999

[nickgeneratorfailed]

Yep, I’m windows 10. I still think that at sizes like this you need to adjust brush settings to achieve good performance. I only managed to get good performance with a large blending brush after I adjusted the settings for it to better reflect the size.

But I’m not saying there couldn’t be differences on macOS. I believe the fastest krita is on Linux (?).

[CrazyCatBird]

@nickgeneratorfailed

Turns out their performance comes form properly adjusting brush settings

Other than the spacing (turn off auto spacing as well), what else should be done?