Guides · 8 min read
AI servers for local LLMs: Tesla P40, V100, L4 or A100?
How much GPU memory a language model needs, which data-centre GPU suits which budget and what matters for cooling, power and software.
Laden … / Loading …
GPU & AI Servers
GPU servers and AI servers are built to supply several accelerators reliably with power, cooling and PCIe bandwidth. Typical workloads are large language model inference with vLLM, Ollama or llama.cpp, training and fine-tuning, rendering, virtual desktops and video transcoding. We offer refurbished systems such as the Dell PowerEdge R740 and R750xa, HPE ProLiant DL380 Gen10 and GPU servers from Supermicro and Gigabyte with two to eight slots for dual-width cards, fitted with NVIDIA GPUs on request. What matters is airflow for passively cooled cards, the correct GPU power cables (usually 8-pin CPU/EPS), adequately sized power supplies and the PCIe generation: Gen3 on 1st and 2nd Gen Xeon Scalable, Gen4 on AMD EPYC 7002/7003 and 3rd Gen Xeon. Every system is checked and tested with the GPUs you ordered.
- Prices from
- €0.00 gross
- Warranty up to
- 36 months
- Dispatch from stock
- within 24 h
0 products
No products found
Adjust the filters or ask us directly. We can source most hardware within a few days.
Buying a GPU server: choosing the right AI server
A GPU server gives several accelerators the power, airflow and PCIe lanes they need – exactly what an ordinary server lacks. Our range covers three classes: a 2U server with two Tesla V100s already fitted and two systems for up to eight dual-slot GPUs. Which platform makes the right AI server depends on model size, number of cards and where the machine will live.
Three GPU servers compared
| Model | Form factor | CPU platform | GPU slots | Power supplies |
|---|---|---|---|---|
| Dell R740 with 2× V100 | 2U | 2× Xeon Gold, PCIe 3.0 | up to 3 dual-slot, 2× V100 fitted | 2× 1600 to 2400 W |
| Supermicro SYS-4029GP-TRT | 4U | 2× Xeon Scalable, PCIe 3.0, dual root | 8 dual-slot, base unit without GPUs | 4× 2000 W Titanium |
| Gigabyte G292-Z20 | 2U | 1× AMD EPYC 7002, PCIe 4.0 | 8 dual-slot, base unit without GPUs | 2× 2200 W Platinum |
The Dell R740 is the entry point: a standard 2U server with a GPU kit that runs in the rack like any other host and can carry VMs or databases on the side. With 32 GB of HBM2 between its two cards, it handles quantised language models up to about 30 billion parameters, or LoRA fine-tuning of small models.