The local Artificial Intelligence is not just a thing for big servers or desktops with huge graphics card. Nowadays, a compact device like the SLIMBOOK ONE can execute AI models locally to create private assistants, analyse documents, summarise texts, help with programming, translate, query knowledge databases or work with data without sending it constantly to the Cloud.
But the question is: Which models can the SLIMBOOK one really run?
What models you can run depending on RAM
| Configuration | Recommended models | Recommended usecase | Commentary |
|---|---|---|---|
| 16 GB RAM | Llama 3.2 1B / 3B Qwen2.5 3B Gemma 2 2B Phi-3 Mini / Phi-3.5 Mini DeepSeek-Coder 1.3B Some quantised 6B / 7B | Basic chatbot Summarise short-length texts Classification of information Lightweight translation Simple personal assistant | Valid configuration to start with local AI or for lightweight tasks. It is not the most recommended if you want to work usually with more capable models |
| 32 GB RAM | Llama 3.1 / 3.2 8B Qwen2.5 7B Mistral 7B Gemma 2 9B DeepSeek-Coder 6.7B Embeddings such as Nomic Embed, BGE o MiniLM | Local office assistant small team Private small chatbot Document queries Lightweight RAG with documentation Text generation Basic help with programming | For most users, 32GB already lets you experience AI locally properly, specially with quantised models and simple tools like Ollama |
| 64 GB RAM | Qwen2.5 14B Qwen2.5-Coder 14B Mistral Nemo 12B Llama 3.1 8B with large context DeepSeek-Coder 14B / 16B quantised Some quantised 20B models | Internal assistants with more capabilities Help with programming Log analysis Technical support Business documentation Technical redaction Local agents with more context | For us, 64GB of RAM is the sweet spot if you want to use the SLIMBOOK ONE as a serious local AI device. It is the most balanced configuration between size, usage, price and real capabilities |
| 96 GB / 128 GB RAM | Qwen 32B quantised DeepSeek 32B quantised Llama 3.1 70B quantised Big models in GGUF format Agents with more context Document base | Local AI lab Tests with big models Research and Development Privacy Heavy documentation analysis Scenarios where local control matters more than maximum speed | Lets you load big models, but does not need to be confused with high range workstations with a dedicated GPU. A model big in size being able to be loaded does not mean that its speed will be great |
| ONE + eGPU | Depends on the VRAM of the external graphics card | Accelered midium inference Artifical vision moderate Heavy models Image generation Multimedia processing Dedicated GPU workflows | The most powerful option if you need accelerated graphics. The ONE can be compact in normal day-to-day usage and grow with an eGPU when needed. |
AMD 7 H 255 and AI 9 HX 370: CPU vs CPU, and CPU vs NPU
The SLIMBOOK ONE can be configured with different AMD processors. For local AI, the two models to compare are the AMD Ryzen 7 H 255 and the AMD Ryzen AI 9 HX 370.
The Ryzen 7 H 255 is a very capable CPU for a miniPC. It contains 8 cores and 16 threads, integrated graphics Radeon 780M and lets you run local models through CPU, RAM and when the software allows it, accelerated graphcis. It is a very interesting option for the ones who want a powerful, compact and balanced device.
The Ryzen AI 9 HX 370, in contrast, goes one step further. It contains 12 cores and 24 threads, integrated graphics 890M, and over all of that, a dedicated NPU for Artificial intelligence, up to 50 TOPS with the NPU and 80 TOPS in total when all the processing blocks are combined.
The NPU provides speed when running AI with the proper software, since the core is designed to be used with algorithms used in Artificial Intelligence.
This is important because we are not just talking of "more CPU". We are talking about a designed architecture ready to be used with the new generations of AI software.
In other words:
| Model | CPU | Integrated GPU | NPU | Recommended use |
|---|---|---|---|---|
| ONE Ryzen 7 H 255 | 8 cores / 16 threads | Radeon 780M | No NPU | Local AI with CPU and RAM, Ollama, llama.cpp, quantised models and general use |
| ONE Ryzen AI 9 HX 370 | 12 cores / 24 threads | Radeon 890M | 50 TOPS NPU / 80 TOPS total | Local AI in higher quality, better CPU, iGPU and NPU with proper software |
Then, is the Ryzen AI 9 HX 370 the better choice for AI?
Yes, if we are talking about potential for local artificial intelligence, the Ryzen AI 9 HX 370 is clearly the most interesting choice.
Having more CPU cores, a more powerful integrated graphics, and also a dedicated NPU. That NPU does not replace a dedicated high range graphics card, but it allows you to run some AI workflows more efficient, with less consumption and without CPU or GPU usage.
The key is in the software. Nowadays, local AI tools still use CPU, RAM, integrated GPU or dedicated GPU. But each day, more tools are integrating NPU usage for modern CPUs.
For this reason, an user that just wants to start with local AI, the Ryzen 7 H 255 can be enough. But for those who want a more futureproof device for efficient AI, our recommendation is the SLIMBOOK ONE AI9 with Ryzen AI 9 HX 370.
Ollama and llama.cpp: the simplest way to start
For most of the users, the most easy option to start with local AI is using Ollama.
Ollama lets you download and run models such as Llama, Qwen, Mistral, Gemma, Phi or DeepSeek with simple commands. For example:
ollama run llama3.2
Or also:
ollama run qwen2.5
Ollama also uses technologies like llama.cpp, one of the most important projects of the local AI ecosystem. llama.cpp lets you run models in optimised formats like GGUF, using quantification and making that several language models work in personal devices without depending on the Cloud.
In other words:
llama.cpp is one of the most important technical bases to run local optimised.
Ollama lets you run these models with ease.
GGUF and the quantised models let you run big models using less memory.
For this reason, to start with local AI with the SLIMBOOK ONE, Ollama and llama.cpp is a recommended combination.
Ollama is ideal for:
Users that want to start fast
Businesses that want to try local assistants
Internal chatbots
RAG about documentation
Automatisations
Tests with different models
Prototype development
Models like Llama, Qwen, Mistral, Gemma, Phi or DeepSeek
FastFlowLM and the NPU: Efficient AI in Ryzen AI
In the models with AMD Ryzen AI, like the SLIMBOOK ONE AI9, another element appears: the NPU.
That NPU does not replace a dedicated high range graphics card, but it allows you to run some AI workflows more efficient, with less consumption and without CPU or GPU usage.
Solutions like FastFlowLM are perfect here, designed to use the NPU of the Ryzen AI processors to run the language models.
While Ollama excels at its easy usage and model variety, FastFlowLM aims to use the NPU to run local AI models more efficient.
This can be interesting for:
Local assistants that are always active
Lightweight inference
Compact devices
Low consumption
Scenarios where we want to reserve the usage of CPU and GPU for other tasks
Tests with the integrated NPU of Ryzen AI
In other words:
Ollama is the easiest option to start.
llama.cpp is the most important technical bases to run local optimised models.
FastFlowLM is an advanced option for those who want to experiment with NPU and the efficency of RyzenAI
What is better: Ollama, llama.cpp or FastFlowLM?
They are not exactly the same, and they do not act in the same areas.
| Tool | Excels at | Use case |
|---|---|---|
| Ollama | Easy usage, lots of models, simple commands | Users, businesses and developers that want to start fast |
| llama.cpp | Technical base, GGUF models, quantisation, advanced control | Technical users, integrators and developers |
| FastFlowLM | Using the NPU of Ryzen AI, efficency, low consumption | Advanced users that want to experiment with the NPU |
Our recommendation would be:
Ollama to start
llama.cpp for those who want more technical control
FastFlowLM to use the NPU of Ryzen AI
What about the Integrated GPU?
The SLIMBOOK ONE AI9 incorporates Radeon 890M integrated graphics, meanwhile the Ryzen 7 H 255 incorporates Radeon 780M. These iGPU are more and more powerful with the time and it helps you in some workloads, depending on the operating system, drivers, backend and software compatibility used.
In any case, when we talk of big models, heavy workloads or maximum speed in inference, a dedicated GPU with VRAM will always go ahead iGPUs.
For this reason, it is important to know the different levels:
CPU + RAM
It is the most universal base. Lets you run several quantised models.llama.cpp / Ollama
Eases out the executing of local models and are ideal for startingNPU Ryzen AI
Interesting for efficiency, low consumption and new tools like FastFlowLm.Dedicated GPU or eGPU
Recommended when you want speed, heavier models or vision/generation tasks.
SLIMBOOK ONE + eGPU: when you need more power
One of the advantages of SLIMBOOK ONE is that you can upgrade with the eGPU, using connections like Oculink or USB-C, depending on configuration or neccesities.
This lets you connect an external graphics card and transforms the ONE in a more powerful machine for AI workloads.
An eGPU can be interesting for:
Big models
Accelerated inference
Artificial vision
Image generation
Multimedia processing
CUDA, ROCm or accelerated graphics workflows
Users that want a compact day-to-day use, but want power when needed
The concept is simple: the SLIMBOOK ONE can be an elegant and efficient miniPC for daily usage, and when a project needs power, you can connect an eGPu to multiply your graphic capabilities.
Local AI: privacy, independence and control
Running local AI is not just a performance need, It is also a privacy and control issue.
When we use external Artificial Intelligence services, most of the time we send texts, documents, code, mails, reports or company information to third party infrastructure. For personal usage it might not be a problem, but to businesses, professionals, administrative places or sensitive projects, it can be an important issue.
With Local AI you can:
Analysing documents without uploading to the Cloud.
Creating private assistants
Consulting company internal information.
Working with logs, code or sensitive data
Minimising dependency on external providers
Evading variable costs for using providers
Maintaining more control with your tools
The cloud will stay. There is huge models and specialised services that make sense that are not in your local device. But most of the day-to-day tasks can be resolved perfectly with local models.
Our recommendation
If you want to buy a SLIMBOOK ONE for local artificial intelligence, our recommendation would be:
To start
SLIMBOOK ONE with 32 GB of RAM
A good choice for 7B or 8B models, lightweight assistants, tests with Ollama and moderate personal or professional use
Second level of local AI
SLIMBOOK ONE AI9 with 64GB of RAM
The most balanced option. Lets you run comfortably with 7B, 8B, 12B and 14B models, and also offers a better CPU, iGPU and with integrated NPU.
For bigger LLMs and more data
SLIMBOOK ONE AI9 with 128GB of RAM
Recommended for those who want to experiment with big models, more context, complex agents or document bases.
For more speed
SLIMBOOK ONE + eGPU
The most powerful option when you need acceleration with a dedicated GPU, heavy models, artificial vision or exigent AI workflows
Conclusion
The SLIMBOOK ONE proves that local AI does not need to take up excessive space nor to depend always on the Cloud. In a compact form factor, elegant and efficient, you can run models like Llama, Qwen, Mistral, Gemma, Phi o DeepSeek, create private assistants, analyse documents, automatise tasks and work with artificial intelligence under your control.
To start, Ollama is the easiest and recommended option. For those who want more technical control, llama.cpp is one of the key pieces of the local AI ecosystem. For those who want to go one step further, FastFlowLM lets you take advantage of the NPU of the Ryzen AI processors. And for heavy workloads, the SLIMBOOK ONE ecosystem can grow with the eGPU.
It is not just to promise wonders. It is to choose wisely your configuration, model and tools.
And there the SLIMBOOK ONE shines: a small computer powerful enough for the next generation of local artificial intelligence, private and efficient.
Choose your SLIMBOOK ONE for local AI
If you want to start with local artificial intelligence, you can configurate your SLIMBOOK ONE depending your needs.
SLIMBOOK ONE AMD Ryzen 7 H 255
SLIMBOOK ONE AMD Ryzen AI 9 HX 370
If you need more power for AI, you can upgrade your SLIMBOOK ONE with one of our Dock eGPU USB4 & OCuLink 800W, designed to be used for heavy workloads and accelerated inference, artificial vision, image generation, multimedia processing or heavy models.
Buy Dock eGPU USB4 & OCuLink 800W
Important: the eGPU dock does not include a dedicated graphics card, sold separately, ask us or visit some available here.
And as a larger and more powerful alternative, be sure to check out our desktop computers, the range Nexus, las Workstation para IA.
