⚠

Configuration note: Es sind 3 Grafikkarten installiert, aber nur 2 wurden fuer diesen Lauf genutzt. Mit allen 3 GPUs koennte der Durchsatz deutlich hoeher liegen.

Generation28,90tok/s

Prefill1.318,49tok/s

Time to First Token2.246,00ms

Total duration75,66s

Concurrency1parallel

Ranking in the field

This GPU All GPUs

218of 455 systems

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

This run is better than 52 % of all comparable systems.

Generation 28,9 tok/s

-18 % vs Ø 35,3

Prefill 1.318,5 tok/s

-33 % vs Ø 1.966,4

Time to First Token 2.246 ms

-94 % vs Ø 35.848

Distribution in the field1 – 148 tok/s

Ø 35 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-85f64f

147,8 tok/s

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165908-d95290

147,8 tok/s

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-637af7

143,9 tok/s

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165908-95273d

143,9 tok/s

Nemotron-3-Nano-4B3× AMD Radeon AI PRO R9700 · run-20260805-053357-112090

139,8 tok/s

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-65d35f

137,7 tok/s

gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165908-ff206d

137,7 tok/s

Nemotron-3-Nano-4B3× AMD Radeon AI PRO R9700 · run-20260805-053357-ed67ec

136,2 tok/s

Nemotron-3-Nano-4B3× AMD Radeon AI PRO R9700 · run-20260805-053357-16d611

129,5 tok/s

Nemotron-3-Nano-30B-A3B4× AMD Radeon AI PRO R9700 · run-20260817-111922-dd07f5

125,7 tok/s

Nemotron-3.5-Lightning-30B-A3B4× AMD Radeon AI PRO R9700 · run-20260815-060242-025c3d

121,3 tok/s

Nemotron-3.5-Lightning-30B-A3B4× AMD Radeon AI PRO R9700 · run-20260815-060242-dadb7a

113,6 tok/s

Nemotron-3.5-Lightning-30B-A3B4× AMD Radeon AI PRO R9700 · run-20260815-060243-6a7849

108,1 tok/s

Nemotron-Cascade-2-30B-A3B3× AMD Radeon AI PRO R9700 · run-20260722-181156-a273f2

107,9 tok/s

Magistral-Small-2509 this run3× AMD Radeon AI PRO R9700 · run-20260722-165909-610fd5

28,9 tok/s

How does this benchmark compare on other GPUs?

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

NVIDIA GeForce RTX 5090run-20260729-032123-c1f464

102,5 tok/s

NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260728-140955-2ee65a

100,1 tok/s

3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260728-140955-0f1e40

97,0 tok/s

3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032125-60ebdf

96,1 tok/s

NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260724-004604-d065ac

96,0 tok/s

3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032125-d4a01f

95,9 tok/s

3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032124-f17a2f

84,9 tok/s

2× NVIDIA RTX A6000run-20260829-184321-7dcaa4

79,5 tok/s

2× NVIDIA RTX A6000run-20260827-140320-273ccc

75,0 tok/s

NVIDIA GeForce RTX 3090 Tirun-20260728-143205-8e5198

59,1 tok/s

NVIDIA GeForce RTX 3090 Tirun-20260724-004603-cbe35f

58,0 tok/s

2× NVIDIA RTX A6000run-20260828-084107-a334be

44,5 tok/s

2× NVIDIA RTX A6000run-20260830-011927-6edfca

44,4 tok/s

NVIDIA RTX A6000run-20260827-003159-901bba

40,4 tok/s

Hardware

GPU: 3x AMD Radeon AI PRO R9700 · 32 GB VRAM
CPU: AMD Ryzen Threadripper PRO 7955WX 16-Cores
RAM: 184 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE

Setup

Runtime: llama.cpp
Quantization: Q4_K_M
Model: Magistral-Small-2509

Configuration

benchmark-konfiguration — run-20260722-165909-610fd5

# LLM-Benchmark Konfiguration # Modell : Magistral-Small-2509 # Engine : llama.cpp # Run-ID : run-20260722-165909-610fd5 # GPU : 3x AMD Radeon AI PRO R9700 # CPU : AMD Ryzen Threadripper PRO 7955WX 16-Cores # RAM : 184 GB bench@llm-benchmark:~$ /src/build/bin/llama-server \ -m /root/.cache/huggingface/hub/models--unsloth--Magistral-Small-2509-GGUF/snapshots/77fe877ee8e44eadf97235a1ad77471678eed310/Magistral-Small-2509-Q4_K_M.gguf \ -ngl 999 \ -c 8192 \ -a Magistral-Small-2509 \ -sm layer \ --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.	Magistral-Small-2509
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.	8192
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird.	/root/.cache/huggingface/hub/models--unsloth--Magistral-Small-2509-GGUF/snapshots/77fe877ee8e44eadf97235a1ad77471678eed310/Magistral-Small-2509-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
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.	8192
Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen.	Magistral-Small-2509
Split-Mode?Verteilung ueber mehrere GPUs: none (nur eine GPU), layer (Layer aufteilen) oder row (Tensoren zeilenweise).	layer

All benchmarks of this model To leaderboard

Model comparison

Magistral-Small-2509 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

NVIDIA GeForce RTX 5090 737,0 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 726,9 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 637,2 tok/sNVIDIA RTX A6000 534,8 tok/sNVIDIA GeForce RTX 3090 Ti 353,8 tok/s★ AMD Radeon AI PRO R9700 230,0 tok/s this runNVIDIA GeForce RTX 5070 Ti 159,0 tok/sAMD Radeon PRO W7800 48GB 153,0 tok/sAMD Radeon PRO W7900 Dual Slot 141,7 tok/sIntel Arc Pro B70 37,9 tok/sNVIDIA Tesla P100 PCIe 16GB 24,7 tok/s

CPUby processor

AMD Ryzen 7 5800X3D 8-Core Processor 737,0 tok/sAMD Ryzen 9 9950X 16-Core Processor 726,9 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 637,2 tok/s★ AMD Ryzen Threadripper PRO 7955WX 16-Cores 534,8 tok/s this runAMD Ryzen 9 8945HX with Radeon Graphics 353,8 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 159,0 tok/sAMD Ryzen 9 7945HX with Radeon Graphics 37,9 tok/s

MBby mainboard

ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 737,0 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 726,9 tok/s★ ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 637,2 tok/s this runMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 353,8 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 159,0 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd DRFXI (MotherBoard Series) 37,9 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) 24,7 tok/s

ENGby engine

★ llama.cpp 737,0 tok/s this runvLLM 726,9 tok/sunbekannt 28,9 tok/s

DRVby driver

unbekannt 737,0 tok/sIntel 26.18.38308.4 37,9 tok/sAMD 7.0.0-27-generic 28,9 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 URL Reset

⚡ Electricity price EUR/kWh

⚙️ System utilization 100 %

🖥️ Acquisition EUR

🔌 Idle 125 W

⚡ TDP 971 W

☁️ External LLM (API) show

Input €/1M Output €/1M Input per output token

Electricity0.30 EUR/kWh

Avg power (incl. idle)971 W estimated (TDP)GPU 900 + CPU 46 + Board 25 W full load

Avg cost / hourEUR 0.29

Electricity / 1M tokensEUR 2.80

Token / kWh107.15K

Acquisition (system)EUR 9,674 full priceGPU EUR 4,200 · CPU EUR 1,399 · Board EUR 1,299 · RAM EUR 2,576 · PSU EUR 200

Electricity (2 years)–

TCO (2 years)EUR 14,778

Output tokens (2 years)1.82B

☁️ 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 (125 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

	Magistral-Small-25093x AMD Radeon AI PRO R9700	Magistral-Small-2509NVIDIA GeForce RTX 5090	Magistral-Small-2509NVIDIA RTX PRO 6000 Blackwell Workstation Edition	Magistral-Small-25093x 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)	–	–	–	–

3× AMD Radeon AI PRO R9700 · this run 29 tok/s NVIDIA GeForce RTX 5090 102 tok/s NVIDIA RTX PRO 6000 Blackwell Workstation Edition 100 tok/s 3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 97 tok/s

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.