Bottleneck when building blueprints

Community Community & Industry Discussion
, ,
1

Hello,

I am considering upgrading my PC or building a new one to decrease build times.

I work on a big project, with a lot of logic done in blueprint. When I want to test a change on a console (a console-specific behaviour, I can’t test it on PC) (debug is also a pain), I have to wait 2 hours. It sucks, it’s not my choice, I don’t have such issues when I edit C++ but I have to deal with it.

My i7-8700k is only at 17%, my M2 SSD in which Windows, the Project and UE are installed is in low%, my other M2 SSD used exclusively as a ridiculously large swap is barely used too. At least that’s what the task manager says.

I already went from 4h to 2h by going from 32GB to 64GB of RAM, adding the above-mentioned ridiculously large swap (I would need 300 GB of RAM to not swap), and tweaking the garbage collection settings.

I think that maybe RAM speed is the bottleneck and faster ram will save me time. I have DDR4 RAM, I don’t know if DDR5 brings significant gains. Clock speed is higher, but latency in μs not so much.

Is RAM really the problem or I am missing something? What is the most cost efficient way to decrease the build time?

1 Like
2

There are no benefits in using DDR5 right now, the price is not justified and the performances are almost the same.

Which console are you referring to? How are you connected to it? How big are the builds?

When i build for oculus quest 2 my CPU reaches 100% usage on all its 24 threads so it’s strange that it uses only the 17% of your 12 threads.

(I’ve never built for Playstation or Xbox because the devkit are not available to everyone so I’m wondering how do you transfer data to them)

2 Likes
3

Ok thank you for avoiding me wasting a lot of money.

The built game is in the 10s of GB. It’s deployed through gigabit ethernet. The deploy itself it not that long. I have UnrealPak enabled. It’s supposed to cook only modified content, package everything in a single file then deploy that file.

1 Like
4

I think I found the problem, apparently the packaging process is not Multi-thread as you can read here, so the bottleneck is in the software.

Today I tried to build again for oculus after I read this and I noticed that it uses 100% of the CPU only for a short period, and it’s kinda disappointing!

Apparently multithreading works for compiling C++ code, Shaders and Lights but not while cooking a build.

3 Likes
5

Ok thank you. I tried changing “Num cookers to spawn”, but it looks like it stays the same.

1 Like
6

ram affects data. so if you migrate or reading prints containing alot of code, maybe will speed it up vs ddr4 normally 3600mhz average people. ddr5 can go 6000 mhz and more, not sure if stable though
ddr5 speed numbers are 3x faster except it has higher latency. so is it actually less time migrating large projects vs ddr4 3600mhz? i cant find any info on this.
and if you got ssd, its top speed is 500-600 mb/s, then you will be limited to ssd, cuz ram is faster than drives.
so try get a nvme and see if ddr5 can go 3-6 gb/s