• 3 % discount on prepayment
  • Insured shipping from Switzerland
  • Sustainable: refurbished instead of new
  • Warranty on all cards

Which card for which model?

Choose a model and see right away which GPU fits, with CHF price and stock status.

47 model families, 90 models. Data as of: 09.10.2026, source: Hugging Face.

Language models (LLM)

Open alternatives to ChatGPT, Gemini and Copilot for chatting, analyzing and programming.

DeepSeek

DeepSeek
NewMoE

Open language model for chat and text. 3 sizes from 304.2B to 1.7T parameters, partly mixture-of-experts (only a part of the parameters is active per token).

09/2026

Modality
Text → text
Parameters
304.2B – 1.7T
Max. context
1M Tokens
Languages
–
License
MIT

GLM

Z.ai (Zhipu)
NewMoE

Open language model for chat and text. 3 sizes from 31.2B to 753.3B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

08/2026

Modality
Text → text
Parameters
31.2B – 753.3B
Max. context
1M Tokens
Languages
EN, ZH
License
MIT

Qwen3.8

Alibaba (Qwen)
NewMoE

Open language model for chat and text. 3 sizes from 27.8B to 2.4T parameters, partly mixture-of-experts (only a part of the parameters is active per token).

08/2026

Modality
Text → text
Parameters
27.8B – 2.4T
Max. context
256k Tokens
Languages
–
License
Apache 2.0

Granite 4.2

IBM
NewGerman
Choose model

Open language model for chat and text. 3 sizes from 3.7B to 29.3B parameters.

08/2026

Modality
Text → text
Parameters
3.7B – 29.3B
Max. context
128k Tokens
Languages
EN, DE, ES, FR, JA…
License
Apache 2.0

Nemotron

NVIDIA
NewGermanMoE
Choose model

Open language model for chat and text. 31.6B Parameter, partly mixture-of-experts (only a part of the parameters is active per token).

08/2026

Modality
Text → text
Parameters
31.6B
Max. context
256k Tokens
Languages
EN, ES, FR, DE, IT…
License
OpenMDW 1.1

LFM2.5

Liquid AI
NewGermanMoE
Choose model

Open language model for chat and text. 2 sizes from 2.7B to 8.5B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

07/2026

Modality
Text → text
Parameters
2.7B – 8.5B
Max. context
128k Tokens
Languages
Multilingual
License
Lfm1.0

Apertus

Swiss AI
New
Choose model

Open language model for chat and text. 2 sizes from 8.9B to 72B parameters.

07/2026

Modality
Text → text
Parameters
8.9B – 72B
Max. context
–
Languages
–
License
Apache 2.0

Kimi

Moonshot AI
NewMoE
Choose model

Open language model for chat and text. 3 sizes from 49.1B to 2.8T parameters, partly mixture-of-experts (only a part of the parameters is active per token).

06/2026

Modality
Text → text
Parameters
49.1B – 2.8T
Max. context
1M Tokens
Languages
–
License
MIT

Cohere

Cohere
GermanMoE

Open language model for chat and text. 2 sizes from 30.5B to 218.8B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

06/2026

Modality
Text → text
Parameters
30.5B – 218.8B
Max. context
488k Tokens
Languages
Multilingual
License
Apache 2.0

MiniMax

MiniMax
MoE
Choose model

Open language model for chat and text. 2 sizes from 228.7B to 427B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

06/2026

Modality
Text → text
Parameters
228.7B – 427B
Max. context
1M Tokens
Languages
–
License
Eigene Lizenz

Gemma 4

Google
MoE

Open language model for chat and text. 5 sizes from 5.1B to 31.3B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

05/2026

Modality
Text → text
Parameters
5.1B – 31.3B
Max. context
256k Tokens
Languages
–
License
Apache 2.0

Qwen3.6

Alibaba (Qwen)
MoE
Choose model

Open language model for chat and text. 2 sizes from 27.8B to 36B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

04/2026

Modality
Text → text
Parameters
27.8B – 36B
Max. context
256k Tokens
Languages
–
License
Apache 2.0

Mistral

Mistral AI
GermanMoE

Open language model for chat and text. 3 sizes from 119.4B to 675B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

03/2026

Modality
Text → text
Parameters
119.4B – 675B
Max. context
1M Tokens
Languages
Multilingual
License
Apache 2.0

Qwen3.5

Alibaba (Qwen)
MoE

Open language model for chat and text. 8 sizes from 900M to 403.4B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

02/2026

Modality
Text → text
Parameters
900M – 403.4B
Max. context
256k Tokens
Languages
–
License
Apache 2.0

Qwen3-Coder

Alibaba (Qwen)
MoE
Choose model

Open model for programmers. 79.7B Parameter, partly mixture-of-experts (only a part of the parameters is active per token).

01/2026

Modality
Text → code
Parameters
79.7B
Max. context
256k Tokens
Languages
–
License
Apache 2.0

Devstral

Mistral AI
Choose model

Open model for programmers. 2 sizes from 24B to 125B parameters.

11/2025

Modality
Text → code
Parameters
24B – 125B
Max. context
384k Tokens
Languages
–
License
Apache 2.0

gpt-oss

OpenAI
MoE
Choose model

Open language model for chat and text. 2 sizes from 20.9B to 116.8B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

08/2025

Modality
Text → text
Parameters
20.9B – 116.8B
Max. context
128k Tokens
Languages
–
License
Apache 2.0

Phi

Microsoft

Open language model for chat and text. 2 sizes from 3.8B to 14.7B parameters.

04/2025

Modality
Text → text
Parameters
3.8B – 14.7B
Max. context
128k Tokens
Languages
EN
License
MIT

Llama 4

Meta
GermanMoE
Choose model

Open language model for chat and text. 2 sizes from 108.6B to 401.6B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

04/2025

Modality
Text → text
Parameters
108.6B – 401.6B
Max. context
10M Tokens
Languages
AR, DE, EN, ES, FR…
License
Llama 4 Community

Llama 3

Meta
German
Choose model

Open language model for chat and text. 2 sizes from 8B to 70.6B parameters.

11/2024

Modality
Text → text
Parameters
8B – 70.6B
Max. context
128k Tokens
Languages
EN, DE, FR, IT, PT…
License
Llama 3.1 Community

Embedding models

For semantic search and RAG over your own documents.

EmbeddingGemma

Google
New
Choose model

Creates vectors for semantic search and RAG. 700M parameters.

09/2026

Modality
Text → vector
Parameters
700M
Memory
from approx. 3.5 GB
Languages
Multilingual
License
Apache 2.0

Nemotron Embed

NVIDIA
NewGerman
Choose model

Creates vectors for semantic search and RAG. 2 sizes from 1.1B to 7.5B parameters.

07/2026

Modality
Text → vector
Parameters
1.1B – 7.5B
Memory
from approx. 4.3 GB
Languages
Multilingual
License
OpenMDW 1.1

Granite Embedding

IBM
German
Choose model

Creates vectors for semantic search and RAG. 2 sizes from 100M to 300M parameters.

04/2026

Modality
Text → vector
Parameters
100M – 300M
Memory
from approx. 2.2 GB
Languages
Multilingual
License
Apache 2.0

Harrier

Microsoft
German
Choose model

Creates vectors for semantic search and RAG. 3 sizes from 300M to 27B parameters.

03/2026

Modality
Text → vector
Parameters
300M – 27B
Memory
from approx. 2.5 GB
Languages
Multilingual
License
MIT

Jina Embeddings v5

Jina AI
Non-commercial
Choose model

Creates vectors for semantic search and RAG. 600M parameters.

01/2026

Modality
Text → vector
Parameters
600M
Memory
from approx. 3.2 GB
Languages
Multilingual
License
CC BY-NC 4.0

pplx-embed

Perplexity
Choose model

Creates vectors for semantic search and RAG. 600M parameters.

01/2026

Modality
Text → vector
Parameters
600M
Memory
from approx. 3.2 GB
Languages
Multilingual
License
MIT

Qwen3-VL-Embedding

Alibaba (Qwen)
Choose model

Creates vectors for semantic search and RAG. 2 sizes from 2.1B to 8.1B parameters.

01/2026

Modality
Text → vector
Parameters
2.1B – 8.1B
Memory
from approx. 6.3 GB
Languages
–
License
Apache 2.0

BGE-M3

BAAI
Choose model

Creates vectors for semantic search and RAG. 600M parameters.

01/2024

Modality
Text → vector
Parameters
600M
Memory
from approx. 3.1 GB
Languages
–
License
MIT

Audio models

Convert speech to text and text to speech.

Nemotron ASR

NVIDIA
German
Choose model

Converts speech to text. 600M parameters.

05/2026

Modality
Audio → text
Parameters
600M
Memory
from approx. 3.3 GB
Languages
Multilingual
License
OpenMDW 1.1

Cohere Transcribe

Cohere
German
Choose model

Converts speech to text. 2.1B parameters.

03/2026

Modality
Audio → text
Parameters
2.1B
Memory
from approx. 6.1 GB
Languages
Multilingual
License
Apache 2.0

Fish Audio S2

Fish Audio
GermanNon-commercial
Choose model

Generates speech from text. 4.6B parameters.

03/2026

Modality
Text → audio
Parameters
4.6B
Memory
from approx. 11.1 GB
Languages
Multilingual
License
Fish Audio Research

Granite Speech

IBM
German
Choose model

Converts speech to text. 2.3B parameters.

02/2026

Modality
Audio → text
Parameters
2.3B
Memory
from approx. 6.6 GB
Languages
Multilingual
License
Apache 2.0

Qwen3-ASR

Alibaba (Qwen)
Choose model

Converts speech to text. 2 sizes from 900M to 2.3B parameters.

01/2026

Modality
Audio → text
Parameters
900M – 2.3B
Memory
from approx. 3.9 GB
Languages
–
License
Apache 2.0

Qwen3-TTS

Alibaba (Qwen)
Choose model

Generates speech from text. 1.9B parameters.

01/2026

Modality
Text → audio
Parameters
1.9B
Memory
from approx. 5.8 GB
Languages
–
License
Apache 2.0

VibeVoice

Microsoft
German
Choose model

Converts speech to text. 8.7B parameters.

01/2026

Modality
Audio → text
Parameters
8.7B
Memory
from approx. 19.3 GB
Languages
Multilingual
License
MIT

Voxtral

Mistral AI
German
Choose model

Converts speech to text. 4.4B parameters.

01/2026

Modality
Audio → text
Parameters
4.4B
Memory
from approx. 10.9 GB
Languages
Multilingual
License
Apache 2.0

VibeVoice Realtime

Microsoft
Choose model

Generates speech from text. 1B parameters.

12/2025

Modality
Text → audio
Parameters
1B
Memory
from approx. 4 GB
Languages
EN
License
MIT

Whisper

OpenAI
German

Converts speech to text. 2 sizes from 800M to 1.5B parameters.

10/2024

Modality
Audio → text
Parameters
800M – 1.5B
Memory
from approx. 3.6 GB
Languages
Multilingual
License
MIT

Image models

Generate images from text, on your own hardware.

Qwen-Image

Alibaba (Qwen)
NewNon-commercial
Choose model

Generates images from text. 2 sizes from 7.1B to 20.4B parameters.

09/2026

Modality
Text → image
Parameters
7.1B – 20.4B
Memory
from approx. 16.2 GB
Languages
EN, ZH
License
Qwen Research

Krea 2

Krea
New
Choose model

Generates images from text. 12.8B parameters.

06/2026

Modality
Text → image
Parameters
12.8B
Memory
from approx. 27.6 GB
Languages
EN
License
Krea 2 Community

ERNIE-Image

Baidu
Choose model

Generates images from text. 8B parameters.

04/2026

Modality
Text → image
Parameters
8B
Memory
from approx. 18.1 GB
Languages
–
License
Apache 2.0

FLUX.2

Black Forest Labs
Choose model

Generates images from text. 2 sizes from 3.9B to 9.1B parameters.

03/2026

Modality
Text → image
Parameters
3.9B – 9.1B
Memory
from approx. 9.8 GB
Languages
EN
License
Apache 2.0

Z-Image

Alibaba (Tongyi)
Choose model

Generates images from text. 2 sizes from 6.2B to 6.2B parameters.

01/2026

Modality
Text → image
Parameters
6.2B
Memory
from approx. 14.3 GB
Languages
EN
License
Apache 2.0

GLM-Image

Z.ai (Zhipu)
Choose model

Generates images from text. 6.9B parameters.

01/2026

Modality
Text → image
Parameters
6.9B
Memory
from approx. 15.9 GB
Languages
ZH, EN
License
MIT

HunyuanImage

Tencent
Choose model

Generates images from text. 83B parameters.

09/2025

Modality
Text → image
Parameters
83B
Memory
from approx. 168 GB
Languages
–
License
Tencent Hunyuan Community

FLUX.1

Black Forest Labs
Non-commercial
Choose model

Generates images from text. 11.9B parameters.

07/2024

Modality
Text → image
Parameters
11.9B
Memory
from approx. 25.8 GB
Languages
EN
License
FLUX.1 dev Non-Commercial

Stable Diffusion XL

Stability AI
Choose model

Generates images from text. 2.6B parameters.

07/2023

Modality
Text → image
Parameters
2.6B
Memory
from approx. 7.1 GB
Languages
–
License
OpenRAIL++
Classic parameter calculator (technical)

Which card do I need?

Choose a model size and we calculate memory, speed and power costs.

32 bn.
Number of parameters in billions, from 1 to 200. Typical: 7, 14, 32, 70.
Precision
4 bit (common: Q4_K_M) needs the least memory and is the most widely used.
Context length
How much text the model keeps in view at once.
Recommendation · 32 bn. parameters, 4 Bit, 8k context
1× Tesla V100 32 GB

The model needs about 23 GB of memory. Passively cooled, fan and 8-pin EPS cable required.

Output: tokens / minute
1'459
Output: tokens / second
24.3
Input: tokens / second
641
Wait time for 4'000 input tokens
6.2 s
Power
275 W
Electricity / month
CHF 18.48
Electricity cost per 1M tokens: CHF 0.88
CardQuantityInput tok./sOutput tok./min.WattCHF / M
Tesla T4 16 GB22374851981.91
Tesla V100 16 GB25931'3514501.55
AMD Instinct MI50 16 GB22051'0365202.34
Tesla V100 32 GB16411'4592750.88
AMD Instinct MI50 32 GB12221'1193101.29
A100 40 GB11'0221'7342750.74
CMP 170HX 64 GB (entsperrt)15691'5982750.80
A100 80 GB11'0222'1573100.67

“Unlocked”: mining card whose memory is enabled to 64 GB by software. No manufacturer warranty, speed estimated cautiously, only recommended when no other card fits.

Estimates for one user generating text. The recommendation uses as few cards as possible, of those the smallest with fluent speed (from 20 tokens/s). Output and input are based on measured values of each card, scaled to your model size. The A100 40 GB is derived from the 80 GB measurement. Several cards share the model layer by layer and work one after another: more memory, but no extra speed. Power consumption: cards at 70 % of their maximum power plus 100 W for the rest of the computer. For “mixture of experts” models (only part of the parameters is active) the output is faster than calculated here. The 120 bn. size uses an assumed architecture. Real values depend on software and model. Purchase price not included. Measurement source: llama.cpp Benchmark-Tabellen (Llama 2 7B, 4 Bit): pp512 = Eingabe, tg128 = Ausgabe. Quellen: github.com/ggml-org/llama.cpp, Diskussionen 15013 (NVIDIA) und 15021 (AMD).