Hi,
I will be moving and possibly traveling in the near future and was thinking about the best mini pc for houdini Sims and rendering that I can get for like $3500
I was looking at the base model of the Mac Studio M2 but don't think it will be faster than if I buy a custom PC with say one 4090
and fit it in a mini itx case?

Any recommendation on specs would be helpful.
thx

the base model of studio will have 32GB of ram, which is okay, but 64 will be that much better. given how much apple wants for the RAM upgrades, might as well get the Ultra studio.

note - current XPU Karma does not support any GPUs other than NVidia. and if you can get the 7950x into your chassis, it will just shred (get it to middle eco mode and you have amazing efficiency as well). Also, the speed of 4090 vs anything else for rendering (or anything else really) is not even close.

cant really recommend Mac in your usecase. if you need FCPX or Logic Pro, mac mini will suffice. I have cheapest M2 mini just for that, so i can do heavy stuff on pc, and video edits/exports on mini.
PS - many supplemental apps are more mature on windows as well: Marvelous, Gaea, UE, sbs Painter, photogrammetry stuff, whatever else.

I'd say build your own and run Linux on it. That way you get it at a good price and have full control over what you put in and how/what to upgrade.

I'd also put emphasis on ram memory and cpu cores, but also I don't find that high ram or cpu frequencies matter much.

With gpu it's a mixed bag for me. They are cool but more limited, at least in Houdini (19.5). Again, memory beats gpu speed. In the end I go cpu almost always.

It's useful to think of bottlenecks in your setup. All the gpu power in the world doesn't help you much it the sim halfway through silently crashes because it is out of memory, or if you find that the cool thing you want to do only works on a cpu. Similarly, what's the point of overclocking by 10% if most of your time is spend reading geo from a disk, or if instead of 16 hours you have to wait 16 hours and 11 minutes for something.

ram, disk space, io, those things can kill efficiency very quickly, much more so than minute gains from extra freqs and such things.

Assuming you have decent workstation and a good internet connection I can highly recommend Parsec to do remote work. You can use any simple laptop, I've got people working in Unreal this way, using full desktop horsepower, and that works great. Workflow wise it's nice to be 'on' the familiar workstation and since all heavy lifting,and heavy asset internet traffic, is done there the laptop has it easy, you can save money on cpu/gpu and spend on larger laptop screen.


https://parsec.app [parsec.app]

Thank you all for the thoughtful response I am going to dive into each comment and learn more about your suggestions!

Hi,

I’m using Linux and VNC all the time.

However if you need Adobe things I’d choose Windows or Mac. However my impression is that with PC:s (Win) you get a broader range of customization.

For productivity I’d also prefer to work wothouy VNC/remote because it will be faster and give a better feeling. Remote access is for people like me who prefer a completely silent client machine and ability to access the same setup from anywhere. However if you do remote access you have to place your hardware somewhere and make sure things run stably. Occasionally you might need to physically access the machine so it is risky if you can not call some technical person to help you.

In your situation I’d look for a machine with very good cooling using big fans and heatsinks. That would require a wide chassi. Maybe I’d look at liquid cooling as well if the noise level af full load goes down.

The regular things like many cores and lots of RAM and good enough GPU you already know about.

I’m a Linux all in guy, but it all depends on your workflow and what you do. Simplicity, software and just having everything running is more important than getting some 10-20% performance boost. It depends of what you do woth your time.

Best!

Thanks Swest,
I was under the impression that VNC is slower than remote desktop.
In a way I kinda would prefer remote access but I always notice a bit of lag. Ive tried TeamViewer, Parsec, Splashtop.
Was looking into Teradici and they have a new software called hp anywhere but haven't seen any good reviews etc.

I would have my big PC in l.a and would be working from Europe I wonder if that is even feasible with remote workflows without lags. I need to play animations without seeing them skip etc.

tsiwt
Thanks Swest,
I was under the impression that VNC is slower than remote desktop.
In a way I kinda would prefer remote access but I always notice a bit of lag. Ive tried TeamViewer, Parsec, Splashtop.
Was looking into Teradici and they have a new software called hp anywhere but haven't seen any good reviews etc.

I would have my big PC in l.a and would be working from Europe I wonder if that is even feasible with remote workflows without lags. I need to play animations without seeing them skip etc.

Hi again! For several years I've been using TurboVNC [turbovnc.org] which is made for 3D and video workloads. That I combine with VirtualGL [virtualgl.org] that will make OpenGL play well with virtual (only remote) displays (which TurboVNC uses). Using SSH tunneling, JPEG compression, detection of updated regions as well as lossless updates it works really well. However, sound is not implemented, but I transfer this will other tools.

Tests with TurboVNC between Estonia and Sweden it can seem as if you are in the same city (appx. 8 ms). Also, according to meter.net [www.meter.net] Stockholm to London with around 26 milliseconds delay I'd consider it acceptable. However, Sweden to L.A. is about 170 milliseconds. This you would notice and likely no software in the world can solve that.

Actually for this you just need a decent laptop to work on, if you are comfortable with it. For VNC I have a 22-inch Cintiq attached to a dead silent Microsoft Surface without dedicated graphics. However, if you have a heavy and powerful machine in L.A. you don't really need VNC or RDP (Remote Desktop) to use it. You could rather set up a computing server as a back-end solution. You would not do much interaction with such a server. Just run a script (or push some button that activates is) that will upload your project changes to Los Angeles. Then simply start a batch job. When it is done you can download the results. Then you review the animation without frame dropping on your local machine. For this you will just need to solve the licensing questions. I've not been using Houdini engine (yet), but my guess is that this is what it is made for i.e. non-interactive batch processing [www.sidefx.com]. A couple of years ago I did experiment with these things including making my own simple render farm with scripting. Eventually I used Tractor on some servers. Such a setup is technically feasible and not very complicated to get things working. The mentioned setup I call "the sync method" (as opposed to VNC/RDP), because you basically keep syncing your project with your server using "get" or "send".

If you like you could even copy rendered images to a web server in L.A. (for example nginx/apache2) where you can check how rendering still images is proceeding.

Additionally, it is possible to schedule or ”loop” sync scripts so that your host and backend machine are continually mirrored.

Thx Swest for the detail response.

I will look into the sync method and see how well it works for me!

I thought it would be interesting to mention that one doesn't even need a VNC for this kind of task.

It is possible to do practically the same with just FFmpeg and USB/IP via SSH tunneling. It requires some configuring, like setting up firewall rules and port forwarding, two SSH keys, determining optimal FFmpeg settings for downscaling and encoding of the captured KMS buffer, collecting usbip indexes, and finally --- a decent network connection to support lightly compressed video stream. Oh, and an extra set of USB keyboard and mouse on the client machine! Because once those devices are bound to USB/IP, they are passed-through to the machine we want to control, and as such can no longer be used to control the computer they are physically connected to, until they're unbound.

Most GPUs support hardware H.264 and/or HEVC video encoding, so encoding should consume minimal resources on the host. Pretty much all CPUs support hardware decoding (even ARM ones), so we can use virtually any machine to display the stream hosted by the workstation. Possibly even ODROID, ROCKPro64 or a Pi.
Heavy duty 3D rendering is done on the host that serves FFmpeg stream, and client only decodes the video and hosts USB controls.