Laptop Hardware

[MG2]

Hi all, I realize that has probably been asked but have been unable to find. 

I will be upgrading to AC 28 and need a new laptop to run it. I was looking at an MSI Katana 15 HX B14WGK-248CA.

It is a gaming unit but does have an 

• Intel® Core™ i7-14650HX
• Windows 11 Home which I will upgrade to Pro
• NVIDIA® GeForce RTX™ 5070 Laptop GPU w/ 8GB GDDR7 VRAM
• 32 GB RAM

I work on both small and medium sized projects. I can also get a model with the RTX 5060 but not sure if the slightly lesser graphics card will make much of a difference.

Any thoughts would be appreciated.

Operating system used: Windows 11

[mthd]

Hi @MG2, You may have already had a look at these specifications but here is the link for AC28 general system requirements. They might help guide you. 

https://www.graphisoft.com/resources-and-support/system-requirements-28/

[scottjm]

That looks more than adequate I would say for small to medium. 
If the 5060 is a significant savings I’d take it. For only Archicad use the 5070 is probably overkill, but if you are using escape or twinmotion etc the 5070 is probably worth it. 

[Pavel_N]

Hi, I’m looking for some advice regarding hardware requirements for Archicad and Twinmotion, specifically in the context of modern "AI" laptops with shared memory. I’m particularly interested in the AMD Ryzen AI Max+ 388–395 series, with integrated Radeon 8060S graphics and 32GB of shared RAM.

Does anyone have experience with this hardware in terms of speed and stability when working on medium-sized projects?

My main concern isn't necessarily the raw performance of the CPU or GPU, but rather how 32GB of shared RAM behaves on Windows machines—where memory allocation might not be as efficient as it is on Macs (where 32GB can often be sufficient).

I’m aware that laptops with 64GB of shared RAM are available, but given current RAM pricing, they are well beyond what I’m willing to pay, especially for a Windows-based machine.

I also intend to use the computer for running local AI models to generate visualizations; for this reason, I think machines with more accessible RTX GPUs (featuring 8GB or 12GB of VRAM) aren´t the best fit.

[Laszlo Nagy]

The Radeon 8060S integrated GPU seems to be fast enough because it has ~18,000 in PassMark's G3D Mark (Radeon 8060s vs GeForce RTX 4060 PassMark), which passes Twinmotion's minimum (10,000) and Archicad 29's mid-range recommendation (15,000). 

32 GB shared memory on Windows should be enough for Archicad + Twinmotion, but may not be enough for Archicad + Twinmotion + local AI because they all want to use that same shared memory pool. As far as I know, local AI models require a lot of memory - the larger the model, the more memory they need, more advanced models need hundreds of GBs. Models for image generation specifically need less, but still need at least 20-30 GBs, AFAIK. 

Based on my research laptops with these CPUs do not have the possibility of adding a dedicated GPU or adding RAM later. So, you will have to deal with the amount of RAM you choose for as long as you use this laptop.

However more expensive, it seems to me that a CPU with at least 64 GB of shared memory would be a better choice so you have enough RAM for all three uses. Also consider that you buy this laptop and you might want to use it for the next 4 or 5 years so you should buy enough RAM that will be enough in 4 or 5 years' time, not only in the next 1 or 2 years. 

You could go for a CPU with the 8050s, which has ~16,500 in PassMark's G3D Mark, so the GPU performance would still be good, and then it would cost a bit less.

I would trade a higher performance CPU for a lower performance CPU if I could get 64 GB instead of 32 GB RAM because RAM is almost always more of a bottleneck than CPU performance is.

[Pavel_N]

Thanks for the reply. I’ve read that Windows tends to rely on a fixed VRAM allocation for the GPU. If I understand correctly, this means I would manually reserve, say, 8 GB of VRAM for the GPU within the system, leaving 24 GB for Archicad and Twinmotion—though I’m not sure if that’s cutting it too close? In contrast, Mac memory operates in a completely dynamic way. Hopefully, I provided the right prompt to the AI ​​and interpreted the answer correctly! 🙂

I assume I’ll be running local models with all other programs closed. 32 GB of VRAM should be sufficient these days—for instance, to run the realy capable Flux.1 Dev model with Q4 quantization (requiring 12–14 GB of VRAM), leaving plenty of headroom for ControlNets and history. That said, I realize that having 64 GB of RAM opens up a wider range of model options. The Radeon 8050s is an interesting suggestion, but it’s currently a bit hard to come by here.

Does anyone here have any hands-on experience with the Ryzen AI Max+ paired with an 8060s (or perhaps an RTX Spark) in a configuration with lower RAM?

[Laszlo Nagy]

A Google search may help you with your decision (I don't know how many people on this forum will be able to 😞

https://www.google.com/search?q=AMD+Ryzen+AI+Max%2B+Radeon+8060S+32+GB+Flux.1+Dev+Q4+quantization

Based on what I'm reading in the answer, it looks like 32 GB could work.