Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
Contributed byMarcel SommerMistral AI

Codestral-22B-v0.1

Performance benchmark · measured on 03.08.2026 08:39

Benchmark-IDrun-20260804-052132-43f414
Timebench 3 - Kombi (Prefill + Generation)Dense22BRuntime: vLLMQuantisierung: AWQ
Generation277,86tok/s
Prefill3.780,55tok/s
Time to First Token10.182,50ms
Total duration93,79s
Concurrency10parallel
Ranking in the field
51of 81 systems

Performance benchmark · Primary metric: Generation-Speed (tok/s) · 10× concurrent

This run is better than 38 % of all comparable systems.
Generation 277,9 tok/s
-43 % vs Ø 486,1
Prefill 3.780,6 tok/s
-18 % vs Ø 4.631,4
Time to First Token 10.183 ms
-84 % vs Ø 63.801
Distribution in the field0 – 1.493 tok/s
Ø 486 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

How does this benchmark compare on other GPUs?

Same model on different hardware · 10× concurrent · Generation (tok/s)

Hardware

GPU: NVIDIA GeForce RTX 3090 Ti · 24 GB VRAM
CPU: AMD Ryzen 5 5600X 6-Core Processor
RAM: 30 GB
Mainboard: ASUSTeK COMPUTER INC. PRIME A520M-K

Setup

Runtime: vLLM
Quantization: AWQ
Model: Codestral-22B-v0.1

Configuration

benchmark-konfiguration — run-20260804-052132-43f414
# LLM-Benchmark Konfiguration # Modell : Codestral-22B-v0.1 # Engine : vLLM # Run-ID : run-20260804-052132-43f414 # GPU : NVIDIA GeForce RTX 3090 Ti # CPU : AMD Ryzen 5 5600X 6-Core Processor # RAM : 30 GB bench@llm-benchmark:~$ /opt/vllm-gemma/venv/bin/python /opt/vllm-gemma/venv/bin/vllm serve /home/godcore/models/codestral22b-awq \ --served-model-name Codestral-22B-v0.1 \ --host 192.168.41.116 \ --port 8000 \ --quantization awq \ --dtype half \ --max-model-len 32768 \ --gpu-memory-utilization 0.92 \ --enforce-eager \ --chat-template /home/godcore/models/codestral22b-awq/chat_template.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.vllm
Modellalias?Der Name, unter dem das Modell ueber die API angesprochen wird. Genau dieser Wert muss im Request-Feld 'model' stehen.Codestral-22B-v0.1
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.32768
Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen.Codestral-22B-v0.1
Quantisierung?Quantisierungsverfahren der Gewichte (z.B. awq, gptq, fp8, bitsandbytes). Verkleinert das Modell und spart VRAM, kann die Genauigkeit leicht senken. Leer = keine zusaetzliche Quantisierung.awq
Dtype?Zahlenformat der Modellgewichte bei der Berechnung (z.B. auto, float16, bfloat16). 'auto' waehlt automatisch das vom Modell empfohlene Format.half
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.32768
GPU-Speicher?Anteil des GPU-Speichers (0 bis 1), den vLLM belegen darf. 0.92 = 92 %. Hoeher = mehr Platz fuer den KV-Cache (mehr/laengere parallele Anfragen), aber groesseres Risiko fuer 'Out of Memory'.0.92
Enforce-Eager?Schaltet die optimierte Graph-Ausfuehrung (CUDA-/HIP-Graphs) AB und rechnet Schritt fuer Schritt. Startet schneller und spart etwas VRAM, ist im laufenden Betrieb aber meist langsamer als mit Graphs.aktiv
chat-template?Jinja-Vorlage, die Chat-Nachrichten in den Prompt-Text des Modells umwandelt. Noetig, wenn das Modell keine eigene mitbringt./home/godcore/models/codestral22b-awq/chat_template.jinja

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Codestral-22B-v0.1 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

8216164112050,002.6115.2217.832Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 5090 - 641,4 tok/s Generation, 5.824 tok/s Prefill, TTFT 7.096 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 629,5 tok/s Generation, 5.700 tok/s Prefill, TTFT 6.905 ms (3 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 559,4 tok/s Generation, 6.460 tok/s Prefill, TTFT 8.205 ms (9 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5070 Ti - 62,9 tok/s Generation, 994 tok/s Prefill, TTFT 64.942 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 3090 Ti - 325,2 tok/s Generation, 2.975 tok/s Prefill, TTFT 8.460 ms (9 Laufe) | DIESER LAUF★ NVIDIA GeForce RTX 30...
NVIDIA GeForce RTX 5090 641,4 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 629,5 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 559,4 tok/s★ NVIDIA GeForce RTX 3090 Ti 325,2 tok/s this runNVIDIA GeForce RTX 5070 Ti 62,9 tok/s

CPUby processor

8216164112050,002.6115.2217.832Prefill (tok/s)Generation (tok/s)AMD Ryzen 7 5800X3D 8-Core Processor - 641,4 tok/s Generation, 5.824 tok/s Prefill, TTFT 7.096 ms (3 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen 9 9950X 16-Core Processor - 629,5 tok/s Generation, 5.700 tok/s Prefill, TTFT 6.905 ms (3 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 559,4 tok/s Generation, 6.460 tok/s Prefill, TTFT 8.205 ms (9 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 325,2 tok/s Generation, 2.907 tok/s Prefill, TTFT 14.580 ms (3 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 62,9 tok/s Generation, 994 tok/s Prefill, TTFT 64.942 ms (3 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 5 5600X 6-Core Processor - 277,9 tok/s Generation, 3.010 tok/s Prefill, TTFT 5.401 ms (6 Laufe) | DIESER LAUF★ AMD Ryzen 5 5600X 6-C...
AMD Ryzen 7 5800X3D 8-Core Processor 641,4 tok/sAMD Ryzen 9 9950X 16-Core Processor 629,5 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 559,4 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 325,2 tok/s★ AMD Ryzen 5 5600X 6-Core Processor 277,9 tok/s this runAMD Ryzen Threadripper PRO 5975WX 32-Cores 62,9 tok/s

MBby mainboard

8216164112050,002.6115.2217.832Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 641,4 tok/s Generation, 5.824 tok/s Prefill, TTFT 7.096 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 629,5 tok/s Generation, 5.700 tok/s Prefill, TTFT 6.905 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 559,4 tok/s Generation, 6.460 tok/s Prefill, TTFT 8.205 ms (9 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 325,2 tok/s Generation, 2.907 tok/s Prefill, TTFT 14.580 ms (3 Laufe)Meigao Innovation Tec...ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 62,9 tok/s Generation, 994 tok/s Prefill, TTFT 64.942 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. PRIME A520M-K - 277,9 tok/s Generation, 3.010 tok/s Prefill, TTFT 5.401 ms (6 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 641,4 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 629,5 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 559,4 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 325,2 tok/s★ ASUSTeK COMPUTER INC. PRIME A520M-K 277,9 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 62,9 tok/s

ENGby engine

79160942824664,52.1763.3344.4935.651Prefill (tok/s)Generation (tok/s)llama.cpp - 641,4 tok/s Generation, 4.972 tok/s Prefill, TTFT 16.877 ms (21 Laufe)llama.cppunbekannt - 214,1 tok/s Generation, 2.855 tok/s Prefill, TTFT 4.137 ms (4 Laufe)unbekanntvLLM - 277,9 tok/s Generation, 3.319 tok/s Prefill, TTFT 7.929 ms (2 Laufe) | DIESER LAUF★ vLLM
llama.cpp 641,4 tok/s★ vLLM 277,9 tok/s this rununbekannt 214,1 tok/s

DRVby driver

7066736416095774.2644.4454.6274.808Prefill (tok/s)Generation (tok/s)unbekannt - 641,4 tok/s Generation, 4.536 tok/s Prefill, TTFT 14.327 ms (27 Laufe)unbekannt
unbekannt 641,4 tok/s
💰 Economics

Economics of this run

Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (10× 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 25 W
⚡ TDP 450 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)450 W estimated (TDP)GPU 450 W full load
Avg cost / hourEUR 0.14
Electricity / 1M tokensEUR 0.13
Token / kWh2.22M
Acquisition (system)EUR 1,149 partial priceGPU EUR 999 · PSU EUR 150
Electricity (2 years)
TCO (2 years)EUR 3,514
Output tokens (2 years)17.53B
☁️ 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 (25 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

Codestral-22B-v0.1NVIDIA GeForce RTX 3090 TiCodestral-22B-v0.1NVIDIA GeForce RTX 5090Codestral-22B-v0.1NVIDIA RTX PRO 6000 Blackwell Workstation EditionCodestral-22B-v0.13x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
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

Marcel Sommer

@marcelsommer