>> Working memory not expandable
As a long-term Mac user, I would like to give a warm welcome to the new members of the club!
Terrible value.
It's true that anyone with a recent Mac or any of the Lunar Lake mini PCs is accustomed to non-expandable RAM, but it"s still not a convincing element when coupled with the rest.
What is something you won't find on any spec sheet is how terrible Minisforum's customer support is and how lousy they are when it comes to honouring warranties. Minisforum's products aren't exactly known for their reliability and long-lasting value, either.
Factor that in when considering handing $2k+ for one of their products - at least I know Apple does a wonderful job in that field.
QuoteWorking memory not expandable
But the advantage is that the memory speed is at 8000 MT/s. If you get the 128 GB RAM version, it doesn't matter if it's not expandable, because Strix Halo doesn't support more than 128 GB RAM anyway.
QuoteMinisforum MS-S1 Max
AMD Ryzen AI Max+ 395, Radeon 8060S
122719 MB/s
Geekom A9 Max, AI 9 HX 370
AMD Ryzen AI 9 HX 370, Radeon 890M
86541 MB/s
(From the Framework Desktop Strix Halo review)
HX 370: 128-bit * 5600 MT/s / 1000 / 8 = 89.6 GB/s (does check out with the measurement)
AI Max+ 395: 256-bit * 8000 MT/s / 1000 / 8 = 256 GB/s (does
NOT check out with the measurement)
Can you measure with the newest version 8.0?
To quote you from your Framework Desktop Strix Halo review:
QuoteUnfortunately, the system is not compatible with the recent AMD FSR4 update as the GPU is based on the same RDNA 3.5 architecture as the Radeon 890M and not the Radeon RX 9000 RDNA4 series.
Imagine paying 2400€ for last gen GPU architecture.
LLMsThe "Strix Halo" APU is a 256-bit chip with a theoretical memory bandwidth of 256 GB/s (256-bit * 8000 MT/s / 1000 / 8) (and ~210 GB/s practically (expected)), comparable to an entry level quad-channel (4 * 64-bit) workstation' memory bandwidth. A normal desktop PC is dual-channel at best. AMD specifically advertises "Strix Halo" for running/inferencing LLMs. You can run the same LLMs on any PC, if you have at least the same amount of RAM (well, running off of a SSD will also work, but the speed will be super slow), ATX sized or not, dual-channel RAM or not, the differences are:
- The size: This is 3.5 Liters.
- The RAM speed at which any LLM will be running at: Strix Halo is a quad-channel chip at 8000 MT/s vs a normal PC, which is dual-channel at 5600 MT/s to 6200 MT/s (2*64-bit*6200/1000/8 = 99,2 GB/s)). A (mini-)PC based on the "Strix Halo" APU will run a LLM about 2.5 times faster: 256 GB/s / 99,2 GB/s = ~2.58.
- The RAM upgradability: The LPDDR5X RAM in "Strix Halo"-based PCs is not upgradable, maybe because it runs at 8000 MT/s vs 5600 MT/s to 6200 MT/s typically seen in DDR5 UDIMMs. A DDR5 UDIMM version with upgradable RAM may appear later, but it's not going to be at 8000 MT/s, like the soldered ones. CUDIMM may reach 8000 MT/s.
Questions to ask yourself:
- Is the LLM speed difference of 2.5 times (150 %) and the price worth it vs simply getting 2x48GB RAM sticks or 2x64GB RAM sticks for a fraction of the price and having then more RAM (although, yes, slower) vs paying 2400 bucks and being stuck with the hardware and no upgrade path (on a desktop you could upgrade to 4x64GB)?
- And, if the size matters, you can still get a mini-ITX case, AM5 mini-ITX motherboard and build a PC of the same size (or get a pre-built mini-ITX PC), with the possibility to:
- Upgrade the RAM.
- Having a dedicated GPU. There are not many choices for 4.0 - 4.5 Liter mini-ITX builds, mostly low profile RTX 4060 or RTX 5060), but this is still better and faster (and harder, stronger, hehe) than the built-in iGPU in Strix Halo:
- You get the ability to upgrade the GPU later, like when/if in 2026 the Refresh GPUs come out, using 3GB, instead of 2GB, GDDR7 chips and you get 50 % more VRAM in the same size.
- A dedicated GPU (4060 / 5060) will also have faster prompt processing (pp).
- The ability to partially or fully offload to the fast VRAM of the GPU (5060: 448 GB/s).
- A dedicated GPU adds additional capacity to the RAM.
- And, not LLM related, but: You can also game with higher FPS if you add a GPU that is faster than Strix Halo's iGPU (between RTX 4060 Laptop (=RTX 4050 desktop, which doesn't even exist, this is how bad it would be (en.wikipedia.org/wiki/GeForce_RTX_40_series)) and RTX 4070 Laptop (=RTX 4060 desktop)).
- Having 24 PCIe 4.0 lanes vs Stix Halo's 16 PCIe 4.0 lanes.
- And, if looks matter, there are many arguably better looking mini-ITX cases, too.
A listed con was "inadequate cooling for the 16x slot"
That is because it isn't designed to take on high heat GPU's. If you want to use one they recommend the Oculink card with an external case.
Besides, the slot is only 4x with a 16x riser.
Once again for the cheap seats, Minis Forum designs "balanced systems" where power usage is intended to be balanced throughout the platform. If you distort the power usage to favor one thing at the expense of others in the chassis, you will get reboots, instability or heaven forbid a block PSU or VRM fail.
If you want to cram you favorite global warming GPU in this, forget it. It's just not designed for that.
If you want to cram in a billion different accessories and load it all up with Proxmox and make it your personal datacenter, go nuts, but you should really be using a mini-tower with a more robust PSU for that kind of use case. Minis Forum has cut much of the power budget for these designs to reduce the overall power profile.
As noted on Reddit, don't let the price distort your technical better judgement.
Ok, but why/who would need 64 GB or 128 GB RAM all of a sudden? The latest hype is AI/running LLMs locally and it's private. Download e.g. LM Studio (wrapper for llama.cpp) and try it out or download llama.cpp directly (recently got their webUI redesigned) and the LLMs can be downloaded from huggingface.co). Open-weight LLMs are getting better and better, see "Artificial Analysis Intelligence Index by Open Weights vs Proprietary": artificialanalysis.ai/?intelligence-tab=openWeights.
Current SOTA models (basically the filesize must fit into your GPU's VRAM and/or system RAM and you can run them, just make sure to maybe not go below a Q4_K_M quant, because then the performance is going to degrade heavily, especially for LLMs below, say, 200B parameters size):
- huggingface.co/openai/gpt-oss-20b (from the makers of ChatGPT) (downloads last month: 6,749,481 and you never heard of it?) (can fully fit on a 16 GB GPU (mentioned on OpenAI's own description) and therefore will run very fast)
- huggingface.co/unsloth/gpt-oss-20b-GGUF (same but the GGUF format allows to run not only on GPU, but if you don't have enough VRAM, to also partially or fully offload to RAM) ("guide : running gpt-oss with llama.cpp": github.com/ggml-org/llama.cpp/discussions/15396)
- huggingface.co/unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF
- huggingface.co/unsloth/GLM-4.5-Air-GGUF (106 B)
- huggingface.co/unsloth/gpt-oss-120b-GGUF (also from the makers of ChatGPT (huggingface.co/openai/gpt-oss-120b))
- huggingface.co/unsloth/Qwen3-235B-A22B-Instruct-2507-GGUF
- huggingface.co/unsloth/GLM-4.5-GGUF (355 B)
and many more, including thinking/reasoning variants (some LLMs, like gpt-oss, have easily configurable reasoning efforts: low, medium, high), like huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507, specialized ones, like Qwen3-Coder-30B-A3B-Instruct-GGU and big ones, like DeepSeek-V3.1-Terminus-GGUF (671 B).
News and infos about AI / LLMs e.g.: reddit.com/r/LocalLLaMA/top, reddit.com/r/LocalLLM/top
NOTEBOOKCHECK, maybe write an article to explain to your readers about the AI LLM self-hosting ability or why 64 GB RAM, 128 GB RAM or even more (AMD, 512-bit Strix Halo successor when?) are a thing all of a sudden now, because normally, currently 32 GB RAM are enough for most people.
What's with all the unreadable comments? [edit: they are messed when displayed under the article, but ok in the forum. Plus obvious AI generated long winded comment that people can generate themselves if they're really that interested, just provide the prompt]
Strix Halo is an incredible notebook chip. It breaks my heart that a Thinkpad isn't available with Strix Halo yet, because it can run 120b LLMs at very usable speeds. The only other way to get close to that is to use a Threadripper or server class system with at least four channel RAM, or multiple GPUs, which is going to be an expensive mammoth beast. Or a higher end Mac, but SH will have better prompt processing time (probably M5 will improve this for Mac) and a Mac with 128GB is much more expensive.
See https://www.reddit.com/r/LocalLLaMA/comments/1nabcek/comment/ncswqmi/ . And it's only going to get better as people are focused on improving AMD GPU performance.
But SH as a "workstation" is compromised, so it's not that compelling. Since hybrid AI is becoming more capable and some applications will be better with CUDA for the foreseeable future, this would be an incredible system with an uncompromised companion dedicated GPU. I think "soldered RAM" is fine given the advantages, but limited PCIe expansion for a desktop is just an insult. The entire category of mini PC kind of misses the point of the architecture except for appliance type uses.
It's stupid that the reviewer lists "Working memory not expandable" as a con when the max RAM space supported by the CPU is already reached. There's also no word regarding the higher 256GB/s memory bandwidth from the integrated RAM, which is the whole point of getting Ryzen AI Max+ 395 vs something like 9955HX which has a much tighter bandwidth. Apparently, the reviewer doesn't understand what the higher memory bandwidth is for at all.