SESvenska
Logga in
Meny

Bästa Lokal LLM-chattserver

En hemmaserver för privat LLM-chatt (Ollama + Open WebUI), lokal kodkomplettering och LoRA-fintuning — mycket VRAM, mycket RAM.

Behovet
  • Lokal AI-inferens
  • Kodassistent (lokal LLM)
  • Modellträning / finjustering
Redigera behov
Översikt · Unified-memory build€3,247

Krav vs det här bygget

behov · använt
CPU-trådar1856%OK
RAM96 GB75%OK
Lagring4 TB99%Fullt
GPUAI-kapabel

Förbrukning

Användning
24/7
Tomgång
13 W
Typisk
14 W
Topp
38 W
Energi / år
≈ €31
Över 3 år
≈ €92

Poäng

Totalt69
Prestanda76
Effektivitet93
Värde51

Varför den här varianten

Välj den om

Balances your priorities with a 3247€ price tag and a 13W idle draw.

Vad den offrar
  • Not specialized: it does not push any single dimension (value, silence, expansion) to the extreme.
Hur delarna valdes
  • GMKtec EVO-X2 — AMD Ryzen AI Max+ 395: 32 threads, 128GB RAM, 12W idle — a complete, efficient prebuilt for this workload.
  • Turnkey: no assembly, smaller footprint, lower energy than a custom build.
  • Unified memory: the integrated GPU can use up to ~96 GB of the 128 GB as graphics memory, so the model fits without a dedicated card. Decode speed follows the 256 GB/s memory bus — several times slower than a high-end card, but a fraction of its price and power.
Varför inte de andra
  • Unified-memory build — the model in shared RAM, no card beats "Smart build — best all-round" on performance -24, value +51, efficiency +93, expandability -52, longevity +16 — that is why it is recommended for your priorities.
  • Unified-memory build — the model in shared RAM, no card beats "Long-term build — expandable & future-proof" on performance -24, value +51, efficiency +93, expandability -61, longevity +7 — that is why it is recommended for your priorities.

Innan du köper

2 kritiska
Kritisk
Check the balance
This build may struggle with the workload.
Kritisk
No storage redundancy
1 drive(s) with no mirror or parity for the data you plan to store.