The hardware
behind AI.

The computer's main processor. It handles the operating system, applications, and everything that isn't specialized. For AI, the CPU is secondary — inference itself happens on the GPU — but a weak CPU can bottleneck data processing, data input, and Python code surrounding the model. High-end desktop CPUs such as AMD's Ryzen 9 9950X3D or Intel's Core Ultra series are more than enough for a local AI workstation.

The graphics card. Designed to run thousands of calculations in parallel — which happens to be exactly what AI models need. A modern GPU is now used for both AI and graphics. NVIDIA dominates the AI market thanks to the CUDA ecosystem; AMD and Intel are catching up but still lag in software support.

The built-in memory on a graphics card. The single most important specification for AI. If the model does not fit in VRAM, part of it has to go into the computer's main memory, and then it answers much more slowly. Llama 3.1 70B in 16-bit precision (FP16) needs about 140 GB of VRAM. Quantized to 4-bit, the weights are about 35 GB, and with room for the conversation it needs around 40–42 GB. The RTX 5090 has 32 GB, the RTX 4090 has 24 GB.

NVIDIA's most powerful GeForce card in 2026. 32 GB GDDR7 VRAM, Blackwell architecture. The sweet spot for developers who want to run 30B-class models locally without moving to professional-grade hardware. From SEK 23,990, NVIDIA's recommended price in Sweden. Sufficient for 30B-class models in 4-bit, Flux Dev image generation, Whisper Large-v3 transcription, and Wan video generation.

NVIDIA's AI workstation for developers — a compact "AI computer" with GB10 architecture (Grace Blackwell) and 128 GB unified memory. Designed to run models up to ~200B parameters locally without a data center budget. A step between a powerful PC and a workstation with an RTX 6000. In Swedish shops it costs from about 50,000 SEK excluding VAT (Dustin, 30 October 2026). Check the current price.

NVIDIA's new system architecture powering DGX Spark and similar workstations. Grace is the CPU part (ARM-based, designed for AI workloads), Blackwell is the GPU part. Together they are called GB10. The key feature: unified memory — CPU and GPU share the same memory instead of copying data back and forth. The result is that large models fit into a shared memory pool.

One of AMD's fastest consumer desktop CPUs in 2026. 16 cores, 32 threads, 3D V-Cache. For AI: rarely the bottleneck, but an excellent all-around CPU for a development machine. A good choice if you also want to game or compile heavy projects in parallel.

The standard format for AI models. Each parameter is stored as a 16-bit floating point. Good balance between precision and size — it's what most models are trained in. A 7B model in FP16 takes about 14 GB. For full precision, there is FP32 (32-bit), but it is almost never used in inference anymore — it's twice as large without a noticeable gain in quality.

NVIDIA's new 4-bit format specifically designed for the Blackwell architecture (RTX 50 series and DGX Spark). Cuts model memory by about 3.5x compared with FP16, and Blackwell hardware runs the format directly. Check the quality on your own task.

The general method for compressing models from FP16 to 8-bit, 4-bit, or lower. The GGUF format (llama.cpp) is the de facto standard for distribution. Different variants — Q4_K_M, Q5_K_S, Q8_0 — represent different trade-offs between size, speed, and quality. See also Quantization in the glossary.

The industry standard for large-scale data processing (Big Data). Open source, runs distributed across a cluster. Handles datasets that don't fit on a single machine by distributing the work across many. Used in financial institutions, telecommunications, e-commerce — everywhere where data is measured in terabytes or petabytes.

An open source software suite from NVIDIA that moves classic data processing from CPU to GPU. Designed as a drop-in replacement for well-known Python libraries — you replace import pandas with import cudf and often get large speedups on big datasets. Main modules: cuDF, cuML and cuGraph.

The RAPIDS module that replaces Pandas. Same API — DataFrames, filter, groupby, joins — but runs on the GPU. For data volumes over ~1 GB, the difference is huge. For small tables, Pandas can be faster, because moving data to the GPU takes time.

The RAPIDS module that replaces Scikit-Learn for classic machine learning (random forests, k-means, PCA, regression). Not deep learning — that is the domain of PyTorch and TensorFlow. cuML is for "traditional" ML that you do before or after the LLM step in a pipeline.

The RAPIDS module for network and graph analysis on GPU. Finds patterns in transaction flows, social networks, supply chains. Common use cases: fraud detection (which accounts are linked?), recommendation systems, knowledge graphs.

The classic Python libraries for data processing and traditional machine learning. Pandas (data manipulation) and Scikit-Learn (ML algorithms) are the foundation of the Python ecosystem — they existed long before AI became a consumer product. Runs on CPU. RAPIDS is NVIDIA's GPU-accelerated versions of these — designed as drop-in replacements for large datasets.

An NVIDIA high-speed network card for AI clusters, with QSFP112 ports. It connects DGX nodes so they can share training or inference across several machines without the network becoming the bottleneck. The DGX Station has the newer ConnectX-8 with up to 800 Gb/s.