Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
Contributed byMario AlkaMistral AI

Mistral-Small-4-119B-2603

Performance benchmark · measured on 30.07.2026 18:31

Benchmark-IDrun-20260730-204032-04a9bd
Timebench 3 - Kombi (Prefill + Generation)Dense119BRuntime: llama.cppQuantisierung: UD-Q4_K_M
Generation14,47tok/s
Prefill189,98tok/s
Time to First Token14.242,00ms
Total duration170,07s
Concurrency1parallel
Ranking in the field
48of 67 systems

Performance benchmark · Primary metric: Generation-Speed (tok/s) · 1× concurrent

This run is better than 29 % of all comparable systems.
Generation 14,5 tok/s
-86 % vs Ø 102,4
Prefill 190,0 tok/s
-93 % vs Ø 2.678,5
Time to First Token 14.242 ms
-58 % vs Ø 34.133
Distribution in the field1 – 246 tok/s
Ø 102 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

How does this benchmark compare on other GPUs?

Same model on different hardware · 1× concurrent · Generation (tok/s)

Configuration

benchmark-konfiguration — run-20260730-204032-04a9bd
# LLM-Benchmark Konfiguration # Modell : Mistral-Small-4-119B-2603 # Engine : llama.cpp # Run-ID : run-20260730-204032-04a9bd # GPU : NVIDIA GeForce RTX 3090 Ti # CPU : AMD Ryzen 9 8945HX with Radeon Graphics # RAM : 92 GB bench@llm-benchmark:~$ /root/llama.cpp/build/bin/llama-server \ -m /root/.cache/huggingface/hub/models--unsloth--Mistral-Small-4-119B-2603-GGUF/snapshots/bd93c721735aa32c035c0f19e738cb3371fd56ff/UD-Q4_K_M/Mistral-Small-4-119B-2603-UD-Q4_K_M-00001-of-00003.gguf \ --alias Mistral-Small-4-119B-2603 \ --host 0.0.0.0 \ --port 8000 \ -ngl 12 \ -c 16384 \ -np 4 \ --jinja
Engine?Die Inferenz-Software, die das Modell ausliefert (z.B. vLLM oder llama.cpp). Sie bestimmt Geschwindigkeit, unterstuetzte Modellformate und welche Parameter ueberhaupt verfuegbar sind.llamacpp
Modellalias?Der Name, unter dem das Modell ueber die API angesprochen wird. Genau dieser Wert muss im Request-Feld 'model' stehen.Mistral-Small-4-119B-2603
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.16384
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird./root/.cache/huggingface/hub/models--unsloth--Mistral-Small-4-119B-2603-GGUF/snapshots/bd93c721735aa32c035c0f19e738cb3371fd56ff/UD-Q4_K_M/Mistral-Small-4-119B-2603-UD-Q4_K_M-00001-of-00003.gguf
GPU-Layer?Anzahl der auf die GPU ausgelagerten Modell-Layer. Hoeher = mehr VRAM und schneller; der Rest laeuft auf der CPU. 999 = alles auf GPU.12
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.16384
np4

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Mistral-Small-4-119B-2603 on various hardware

All published performance runs of this model – each bubble a variant: position = prefill (X) × generation (Y), bubble size = number of runs. Closer to the top right = faster. ★ Marked gold = this benchmark.

GPUby graphics card

1.2469356233120,002.6175.2347.851Prefill (tok/s)Generation (tok/s)NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 959,6 tok/s Generation, 6.363 tok/s Prefill, TTFT 5.653 ms (3 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5070 Ti - 62,3 tok/s Generation, 213 tok/s Prefill, TTFT 120.533 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 5090 - 5,8 tok/s Generation, 325 tok/s Prefill, TTFT 5.900 ms (1 Lauf)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 3090 Ti - 58,5 tok/s Generation, 258 tok/s Prefill, TTFT 109.712 ms (3 Laufe) | DIESER LAUF★ NVIDIA GeForce RTX 30...
NVIDIA RTX PRO 6000 Blackwell Workstation Edition 959,6 tok/sNVIDIA GeForce RTX 5070 Ti 62,3 tok/s★ NVIDIA GeForce RTX 3090 Ti 58,5 tok/s this runNVIDIA GeForce RTX 5090 5,8 tok/s

CPUby processor

1.2469356233120,002.6175.2347.851Prefill (tok/s)Generation (tok/s)AMD Ryzen 9 9950X 16-Core Processor - 959,6 tok/s Generation, 6.363 tok/s Prefill, TTFT 5.653 ms (3 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 62,3 tok/s Generation, 213 tok/s Prefill, TTFT 120.533 ms (3 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 7 5800X3D 8-Core Processor - 5,8 tok/s Generation, 325 tok/s Prefill, TTFT 5.900 ms (1 Lauf)AMD Ryzen 7 5800X3D 8...AMD Ryzen 9 8945HX with Radeon Graphics - 58,5 tok/s Generation, 258 tok/s Prefill, TTFT 109.712 ms (3 Laufe) | DIESER LAUF★ AMD Ryzen 9 8945HX wi...
AMD Ryzen 9 9950X 16-Core Processor 959,6 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 62,3 tok/s★ AMD Ryzen 9 8945HX with Radeon Graphics 58,5 tok/s this runAMD Ryzen 7 5800X3D 8-Core Processor 5,8 tok/s

MBby mainboard

1.2469356233120,002.6175.2347.851Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 959,6 tok/s Generation, 6.363 tok/s Prefill, TTFT 5.653 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 62,3 tok/s Generation, 213 tok/s Prefill, TTFT 120.533 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 5,8 tok/s Generation, 325 tok/s Prefill, TTFT 5.900 ms (1 Lauf)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 58,5 tok/s Generation, 258 tok/s Prefill, TTFT 109.712 ms (3 Laufe) | DIESER LAUF★ Meigao Innovation Tec...
ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 959,6 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 62,3 tok/s★ Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 58,5 tok/s this runASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 5,8 tok/s

ENGby engine

1.2469356233120,009221.8442.766Prefill (tok/s)Generation (tok/s)vLLM - 5,8 tok/s Generation, 325 tok/s Prefill, TTFT 5.900 ms (1 Lauf)vLLMllama.cpp - 959,6 tok/s Generation, 2.278 tok/s Prefill, TTFT 78.633 ms (9 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 959,6 tok/s this runvLLM 5,8 tok/s

DRVby driver

1.0561.0089609128641.9582.0412.1242.208Prefill (tok/s)Generation (tok/s)unbekannt - 959,6 tok/s Generation, 2.083 tok/s Prefill, TTFT 71.359 ms (10 Laufe)unbekannt
unbekannt 959,6 tok/s
💰 Economics

Economics of this run

Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (1× concurrent). Methodology →

⚙️ ConfigurationAll metrics and charts below follow these settings – based on a 24-month runtime.Save to URLReset
⚡ Electricity price EUR/kWh
⚙️ System utilization 100 %
🖥️ Acquisition EUR
🔌 Idle 50 W
⚡ TDP 477 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)477 W estimated (TDP)GPU 450 + CPU 17 + Board 10 W full load
Avg cost / hourEUR 0.14
Electricity / 1M tokensEUR 2.75
Token / kWh109.15K
Acquisition (system)EUR 2,986 partial priceGPU EUR 999 · CPU EUR 549 · RAM EUR 1,288 · PSU EUR 150
Electricity (2 years)
TCO (2 years)EUR 5,494
Output tokens (2 years)912.65M
☁️ External LLM (API) – comparison
External LLM cost (2 years)
Savings vs. external (2 years)

All values above and the charts below take the configured system utilization into account: at X% the system generates only X% of the time, the rest it idles (50 W). Cost per hour drops (more idle), cost per token rises.

Cost over 2 years – electricity only

Cost over 2 years – incl. acquisition (TCO)

Speed vs. tokens per euro

Euro per 1M tokens

Comparison vs. API – economics per benchmark

Mistral-Small-4-119B-2603NVIDIA GeForce RTX 3090 TiMistral-Small-4-119B-2603NVIDIA RTX PRO 6000 Blackwell Workstation EditionMistral-Small-4-119B-2603NVIDIA GeForce RTX 5070 TiMistral-Small-4-119B-2603NVIDIA GeForce RTX 5090
Electricity cost (24 mo.)
Acquisition cost
Total cost (TCO)
Generated tokens (24 mo.)
Token price via API
Break-even point (days)
Result (savings / extra cost)

Comparison with up to 3 next-best runs of this model at the same concurrency (at least one on different hardware). Power = GPU TDP + CPU (idle + 15 %) + board (estimated), acquisition = full system (GPU + CPU + board + RAM + PSU), prices = stored market prices.

Contributed by

Mario Alka Administrator

@marioalka

Ich bin Unternehmer, Softwareentwickler und KI-Enthusiast. Seit vielen Jahren entwickle ich Unternehmenssoftware und beschäftige mich inzwischen fast täglich mit lokalen LLMs, KI-Agenten und leistungsfähiger KI-Hardware.

Mit LLM-Benchmark.de möchte ich eine Plattform schaffen, auf der Modelle, GPUs und Agenten objektiv und reproduzierbar miteinander verglichen werden.