General Category > Hardware

Ryzen 5995XW Build & some issues

(1/4) > >>

hi guys

ive had a monster PC built with the below specs.

5995WX CPU
Asus WRX80E mobo
256gb RAM (32GB crucial 3200MHZ x 8)
2TB NVME drives

Its great while rendering, and for the most part a monster of a PC>  However - im getting a lot of lags in MAX still.  like the material editor.  I feel like its even more laggy than my previous PC (3970x).  I feel like in heavy scenes, it chokes a bit and the IR takes some time to start.  But when it starts, it flies.  Im wondering if this is just a MAX issue or is it an AMD thing?  Does anyone else have this PC and if so, is there anything that can be done? perhaps in BIOS?  i had my PC set up by the guys who supplied it so i assume everything is default.


Nejc Kilar:
I have a 5995wx on that same motherboard just with a bit less RAM and more storage, lol. 3ds Max seems to work fine on my end (as much as 3ds Max can) and I can't really say I've noticed any choking although it might be that I haven't really used as heavy scenes as yours. The UI, while doing IR, seems fine to me although based on anecdotal evidence it appears the UI is snappier on faster single threaded CPUs. I would definitely say that 3ds Max could be snappier overall but that was the case with my previous CPU as well.

As the old saying goes though, the more cores you have the less high they can boost so I'd say it is very much expected that a 5995wx is a bit slower at single threaded tasks than say a 5950x. Just the nature of the beast and on the flip side the 5995wx will render more than twice as fast.

I too upgraded from a 3970x and overall I would say that my impression (subjective) is that this CPU is just faster at everything including the single threaded stuff.

So it could be I'm just not dealing with the scenes you are and you actually might be on to something - and that is what makes this very hard to test I'd say. By and large, the above is my impression with the scenes and plugins I'm working with.

We do have an issue on the C4D side where these 64 core CPUs are running into a bit of a wall but that seems like a completely separate issue to yours. And we are working on it :)

Maybe what you could try doing is disabling SMT (hyper-threading) and see if there's any improvements there? Some users have reported certain things work better without SMT but it does take away a chunk of your performance. Might be worth testing just to see - especially if you use all kinds of other plugins where some of them might struggle with so many threads.

This to me, still remains a mystery. At the moment, there are two Corona daily builds for Max, one to test fluidity/smoothness/responsivness of UI. I have yet to test it.. I don't really have much time or patience for doing comparison tests. Imho Corona Devs should have just bought one big Threadripper for internal testing if they don't have one, as they keep saying they can't reproduce it.

Nejc above also can't reproduce it..  But I don't agree with his observation about this difference being minor, or only result of lower-clocks. To me, it's definitely very very perceptible. Magnitude difference esp. in Material Editor, which in Corona is generally very slow, even with Material Caching. And it's nothing to do with Custom Previews because Corona Mat Editor is slow even with them turned Off. It's definitely IO problem, just who knows if it's software or hardware side.

But I, with my 3990X, definitely can. SMT OFF helped...why, not sure. And I can't really put my finger on how much this issue is with 3dsMax which dedicates single thread for almost everything (but people with Cinema4D report similar issues), or how much is it with high-core architecture, or with AMD specifically (the multi-die architecture).

Few of my observations:

1) I haven't noticed degradation of performance between Windows 10 or Windows 11. I find 11 to be rather annoying and buggy at some times, but I stick to it since it's necessary for any new hardware from now on.
2) AMD doesn't lose multi-threaded performance on Balanced Power Plan (Intel does..a lot!) but Ultimate Power Plan is still the go-to for instant turbo-boost and helps a lot in multi-tasking.
3) 3dsMax from my observation, still suffers from extended running. I.e, multiple-days opened scene, etc.. So I make sure to avoid doing that.
4) But I also started doing this with Windows. I remember never having to restart PC few years ago, at one point, my PC was running 6 months without shutdown! But I stopped doing this with late Win10 and Win11 because I find some performance degradation stemming from... who knows what.

Thank you both for your responses.

Firstly - iim 100% in agreeance with all your points Juraj.. Ive felt as though the last few years, things have gotten worse (ie having to turn the PC off/restart to give it some freshness - it suffers from lag from being on so long).

In terms of my scenes, some arent THAT heavy.  Yes, some of my exterior scenes, with 40M polygons (forest packs/anima/vegetation/high res textures etc) there is a bit of lag.  HOWEVER, when i first open the scenes, its usually fairly zippy.  The material editor is somewhat zippy, then at some point it slows right down.

@Juraj - if you turn off SMT (how do you do this BTW, in BIOS?) then whats the point of having a TR?  I mean, do you just do it for quick max/viewport performance and then when youw ant to go full resolution you turn it back on?
Also, im not that technically gifted with tech, but, this SMT, wouldnt it be amazing if you can turn it off within windows (not BIOS) for a single thread boost and then when you want to do IR or final render, then turn it on.  Wondering if its possible to create an application for this?

I also agree, the Corona team I hope are using a TR to also do testing on?  For instance, i use ANIMA a lot.  And the most recent update, was crashing for TR/AMD users for some reason.  The devs quickly jumped on it and send me a daily right away which fixed it.

Overall very happy with the machine, but there is just something something that is just bugging it, it could even be windows itself (im on win 10).

There is actually application that has considerable power over how threads are used and allocated, called "Project Lasso"
But.. it's quite complicated, and it can cause even more problems :- ) So I tried it some time ago...but decided not to use it later. It's quite a power-user tool and I am not power-user. Requires lot of experimentation for which I don't have patience sadly.
But the biggest issue is that Corona and 3dsMax just share the same process so it's not like you can make some priority between the two, Windows sees it as one. But you can use it to anchor that process to single thread so it doesn't jump around (which Windows Scheduler does to best allocate resources but it can do both good and bad).

64Cores/64Threads (Threadripper with SMT(AMD's Hyperthreading) OFF) is still faster than 32Cores/64Threades but yes, true, you are wasting the money you have paid :- (.
I avoided this solution for very long time, I only tried it I think last October, and kept it since. My theory why it helps is that remaining threads, each time full-core this time are distributed more power and compete less for IO/Memory resources. You would think quad-channel or 8-channel memory would help a lot, but to me it seems like "mixed" workloads (rendering and using at same time, like IR) can still get bogged down somehow.

You have to turn it on/off in bios only yes, or at least I don't know of any other way.

It's also "good" idea (good for productivity in our case) to disable Virtualization support while you're already in bios and while in Windows, pretty much every single Defender feature. Which will leave your PC vulnerable but so at least some of the strictest things (like code injection) you can just make exception for 3dsMax and Photoshop.

It's definitely not Windows 10. I am on Windows 11 for a year at least and it's the same, if not little bit worse :- ). Because I find Windows 11 to be generally sluggish in some other aspects, there is lot of bloat.


[0] Message Index

[#] Next page

Go to full version