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

Ministral-3-14B-Reasoning-2512

Performance benchmark · measured on 18.08.2026 11:16

Benchmark-IDrun-20260818-111824-762c0e
Timebench 3 - Kombi (Prefill + Generation)Dense14BRuntime: llama.cppQuantisierung: Q4_K_M
Generation58,18tok/s
Prefill2.416,66tok/s
Time to First Token945,50ms
Total duration37,09s
Concurrency1parallel
Ranking in the field
26of 81 systems

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

This run is better than 69 % of all comparable systems.
Generation 58,2 tok/s
+9 % vs Ø 53,3
Prefill 2.416,7 tok/s
+75 % vs Ø 1.382,6
Time to First Token 946 ms
-78 % vs Ø 4.232
Distribution in the field10 – 155 tok/s
Ø 53 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)

Hardware

GPU: 2x AMD Radeon PRO W7900 Dual Slot · 48 GB VRAM
CPU: AMD Ryzen Threadripper PRO 5975WX 32-Cores
RAM: 247 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI

Setup

Runtime: llama.cpp
Quantization: Q4_K_M
Model: Ministral-3-14B-Reasoning-2512

Configuration

benchmark-konfiguration — run-20260818-111824-762c0e
# LLM-Benchmark Konfiguration # Modell : Ministral-3-14B-Reasoning-2512 # Engine : llama.cpp # Run-ID : run-20260818-111824-762c0e # GPU : 2x AMD Radeon PRO W7900 Dual Slot # CPU : AMD Ryzen Threadripper PRO 5975WX 32-Cores # RAM : 247 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build_hip/bin/llama-server \ -m /home/godcore/bench_scratch/mistralai_Ministral-3-14B-Reasoning-2512-Q4_K_M.gguf \ --alias Ministral-3-14B-Reasoning-2512 \ -ngl 999 \ -fa on \ -sm layer \ --tensor-split 1,1 \ -c 49152 \ -np 12 \ --jinja \ --host 0.0.0.0 \ --port 8000
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.Ministral-3-14B-Reasoning-2512
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.49152
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird./home/godcore/bench_scratch/mistralai_Ministral-3-14B-Reasoning-2512-Q4_K_M.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.999
faon
Split-Mode?Verteilung ueber mehrere GPUs: none (nur eine GPU), layer (Layer aufteilen) oder row (Tensoren zeilenweise).layer
Tensor-Split?Verhaeltnis, in dem die Modell-Layer auf mehrere GPUs verteilt werden, z.B. 3,1.1,1
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.49152
np12

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Ministral-3-14B-Reasoning-2512 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

1.3701.0276853420,004.2228.44512.667Prefill (tok/s)Generation (tok/s)NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 1.060,0 tok/s Generation, 10.288 tok/s Prefill, TTFT 3.218 ms (9 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5090 - 1.028,1 tok/s Generation, 7.724 tok/s Prefill, TTFT 4.732 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 911,6 tok/s Generation, 8.521 tok/s Prefill, TTFT 5.238 ms (9 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5070 Ti - 590,0 tok/s Generation, 5.214 tok/s Prefill, TTFT 5.398 ms (6 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 3090 Ti - 543,1 tok/s Generation, 4.926 tok/s Prefill, TTFT 5.681 ms (6 Laufe)NVIDIA GeForce RTX 30...AMD Radeon AI PRO R9700 - 329,9 tok/s Generation, 3.853 tok/s Prefill, TTFT 8.485 ms (15 Laufe)AMD Radeon AI PRO R97...AMD Radeon 8060S Graphics - 66,3 tok/s Generation, 1.318 tok/s Prefill, TTFT 13.659 ms (2 Laufe)AMD Radeon 8060S Grap...NVIDIA GeForce RTX 2060 - 60,4 tok/s Generation, 1.290 tok/s Prefill, TTFT 5.121 ms (2 Laufe)NVIDIA GeForce RTX 20...NVIDIA Tesla P100 PCIe 16GB - 41,5 tok/s Generation, 501 tok/s Prefill, TTFT 25.353 ms (3 Laufe)NVIDIA Tesla P100 PCI...AMD Radeon PRO W7900 Dual Slot - 226,7 tok/s Generation, 2.572 tok/s Prefill, TTFT 5.319 ms (12 Laufe) | DIESER LAUF★ AMD Radeon PRO W7900 ...
NVIDIA RTX PRO 6000 Blackwell Workstation Edition 1.060,0 tok/sNVIDIA GeForce RTX 5090 1.028,1 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 911,6 tok/sNVIDIA GeForce RTX 5070 Ti 590,0 tok/sNVIDIA GeForce RTX 3090 Ti 543,1 tok/sAMD Radeon AI PRO R9700 329,9 tok/s★ AMD Radeon PRO W7900 Dual Slot 226,7 tok/s this runAMD Radeon 8060S Graphics 66,3 tok/sNVIDIA GeForce RTX 2060 60,4 tok/sNVIDIA Tesla P100 PCIe 16GB 41,5 tok/s

CPUby processor

1.3701.0276853420,004.2228.44512.667Prefill (tok/s)Generation (tok/s)AMD Ryzen 9 9950X 16-Core Processor - 1.060,0 tok/s Generation, 10.288 tok/s Prefill, TTFT 3.218 ms (9 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen 7 5800X3D 8-Core Processor - 1.028,1 tok/s Generation, 7.724 tok/s Prefill, TTFT 4.732 ms (3 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 911,6 tok/s Generation, 8.521 tok/s Prefill, TTFT 5.238 ms (9 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 543,1 tok/s Generation, 4.926 tok/s Prefill, TTFT 5.681 ms (6 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 7955WX 16-Cores - 329,9 tok/s Generation, 3.853 tok/s Prefill, TTFT 8.485 ms (15 Laufe)AMD Ryzen Threadrippe...AMD RYZEN AI MAX+ 395 w/ Radeon 8060S - 66,3 tok/s Generation, 1.318 tok/s Prefill, TTFT 13.659 ms (2 Laufe)AMD RYZEN AI MAX+ 395...Intel(R) Core(TM) i5-7400 CPU @ 3.00GHz - 60,4 tok/s Generation, 1.290 tok/s Prefill, TTFT 5.121 ms (2 Laufe)Intel(R) Core(TM) i5-...AMD Ryzen 9 7945HX with Radeon Graphics - 41,5 tok/s Generation, 501 tok/s Prefill, TTFT 25.353 ms (3 Laufe)AMD Ryzen 9 7945HX wi...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 590,0 tok/s Generation, 3.452 tok/s Prefill, TTFT 5.346 ms (18 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
AMD Ryzen 9 9950X 16-Core Processor 1.060,0 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 1.028,1 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 911,6 tok/s★ AMD Ryzen Threadripper PRO 5975WX 32-Cores 590,0 tok/s this runAMD Ryzen 9 8945HX with Radeon Graphics 543,1 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 329,9 tok/sAMD RYZEN AI MAX+ 395 w/ Radeon 8060S 66,3 tok/sIntel(R) Core(TM) i5-7400 CPU @ 3.00GHz 60,4 tok/sAMD Ryzen 9 7945HX with Radeon Graphics 41,5 tok/s

MBby mainboard

1.3701.0276853420,004.2228.44512.667Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 1.060,0 tok/s Generation, 10.288 tok/s Prefill, TTFT 3.218 ms (9 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 1.028,1 tok/s Generation, 7.724 tok/s Prefill, TTFT 4.732 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 911,6 tok/s Generation, 5.604 tok/s Prefill, TTFT 7.268 ms (24 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 543,1 tok/s Generation, 4.926 tok/s Prefill, TTFT 5.681 ms (6 Laufe)Meigao Innovation Tec...Bosgame AXB35-02 (BeyondMax Series) - 66,3 tok/s Generation, 1.318 tok/s Prefill, TTFT 13.659 ms (2 Laufe)Bosgame AXB35-02 (Bey...ASRock H110 Pro BTC+ - 60,4 tok/s Generation, 1.290 tok/s Prefill, TTFT 5.121 ms (2 Laufe)ASRock H110 Pro BTC+Shenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) - 41,5 tok/s Generation, 501 tok/s Prefill, TTFT 25.353 ms (3 Laufe)Shenzhen Meigao Elect...ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 590,0 tok/s Generation, 3.452 tok/s Prefill, TTFT 5.346 ms (18 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.060,0 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 1.028,1 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 911,6 tok/s★ ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 590,0 tok/s this runMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 543,1 tok/sBosgame AXB35-02 (BeyondMax Series) 66,3 tok/sASRock H110 Pro BTC+ 60,4 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) 41,5 tok/s

ENGby engine

1.3681.0266843420,01.5724.4407.30810.176Prefill (tok/s)Generation (tok/s)vLLM - 1.015,7 tok/s Generation, 8.655 tok/s Prefill, TTFT 1.956 ms (9 Laufe)vLLMunbekannt - 51,0 tok/s Generation, 3.092 tok/s Prefill, TTFT 775 ms (3 Laufe)unbekanntllama.cpp - 1.060,0 tok/s Generation, 4.754 tok/s Prefill, TTFT 8.029 ms (55 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 1.060,0 tok/s this runvLLM 1.015,7 tok/sunbekannt 51,0 tok/s

DRVby driver

1.3681.0266843420,02.3763.5914.8056.019Prefill (tok/s)Generation (tok/s)unbekannt - 1.060,0 tok/s Generation, 5.303 tok/s Prefill, TTFT 7.175 ms (64 Laufe)unbekanntAMD 7.0.0-27-generic - 51,0 tok/s Generation, 3.092 tok/s Prefill, TTFT 775 ms (3 Laufe)AMD 7.0.0-27-generic
unbekannt 1.060,0 tok/sAMD 7.0.0-27-generic 51,0 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 10 W
⚡ TDP 600 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)600 W estimated (TDP)GPU 590 + Board 10 W full load
Avg cost / hourEUR 0.18
Electricity / 1M tokensEUR 0.86
Token / kWh349.08K
Acquisition (system)EUR 2,156 missingRAM EUR 1,976 · PSU EUR 180
Electricity (2 years)
TCO (2 years)EUR 5,310
Output tokens (2 years)3.67B
☁️ 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 (10 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

Ministral-3-14B-Reasoning-25122x AMD Radeon PRO W7900 Dual SlotMinistral-3-14B-Reasoning-2512NVIDIA RTX PRO 6000 Blackwell Workstation EditionMinistral-3-14B-Reasoning-2512NVIDIA RTX PRO 6000 Blackwell Workstation EditionMinistral-3-14B-Reasoning-2512NVIDIA 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.