Contributed byMario AlkaQwen (Alibaba) Qwen3-30B-A3B-Instruct-2507
Performance benchmark · measured on 18.08.2026 13:35
Benchmark-ID
run-20260818-134249-787b11Timebench 3 - Kombi (Prefill + Generation)MoE30BRuntime: llama.cppQuantisierung: Q4_K_M
Generation264,15tok/s
Prefill7.431,65tok/s
Time to First Token4.406,00ms
Total duration87,38s
Concurrency10parallel
Ranking in the field
364of 820 systems
Performance benchmark · Primary metric: Generation-Speed (tok/s) · 10× concurrent
This run is better than 56 % of all comparable systems.
Generation
264,2 tok/s
-46 % vs Ø 485,2
Prefill
7.431,7 tok/s
+59 % vs Ø 4.661,8
Time to First Token
4.406 ms
-93 % vs Ø 63.970
Distribution in the field0 – 2.491 tok/s
Wie schlägt sich dieser Benchmark mit anderen Modellen?
gemma-4-E2B-itNVIDIA GeForce RTX 5090 · run-20260728-184455-e28775
2.491,2 tok/s
gemma-4-E2B-itNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260727-181606-b37ed9
2.414,4 tok/s
Nemotron-3-Nano-4BNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260727-181607-4cca42
2.182,7 tok/s
gemma-4-E2B-it3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-184456-d87c73
2.143,6 tok/s
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260730-035052-9c966e
1.995,4 tok/s
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260728-140954-8007c4
1.993,5 tok/s
Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-0aed86
1.937,7 tok/s
Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-8db0fa
1.911,6 tok/s
Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260727-181606-e0eafd
1.897,0 tok/s
gpt-oss-20bNVIDIA GeForce RTX 5090 · run-20260729-032121-7c961d
1.890,7 tok/s
Nemotron-3-Nano-4BNVIDIA GeForce RTX 5090 · run-20260728-194135-2ed9dd
1.878,7 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-034052-26063b
1.835,7 tok/s
Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-a77046
1.830,3 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-032121-c6e7e0
1.819,5 tok/s
Qwen3-30B-A3B-Instruct-2507 this run2× AMD Radeon PRO W7900 Dual Slot · run-20260818-134249-787b11
264,2 tok/s
How does this benchmark compare on other GPUs?
Same model on different hardware · 10× concurrent · Generation (tok/s)
NVIDIA GeForce RTX 5090run-20260729-105335-2c186b
1.671,0 tok/s
NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260729-032122-3d6d17
1.658,1 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-105342-54b04e
1.515,7 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-105341-5cfa7f
1.507,5 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-105341-7806d7
1.436,0 tok/s
NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260724-004604-860045
961,5 tok/s
NVIDIA GeForce RTX 5090run-20260723-045450-e8df4d
909,3 tok/s
NVIDIA GeForce RTX 3090 Tirun-20260729-060803-9dc48b
905,1 tok/s
NVIDIA GeForce RTX 3090 Tirun-20260724-004603-3aec22
901,9 tok/s
2× AMD Radeon PRO W7900 Dual Slot this runrun-20260818-134249-787b11
264,2 tok/s
2× AMD Radeon PRO W7900 Dual Slotrun-20260818-141454-d2fa2d
251,1 tok/s
AMD Radeon PRO W7900 Dual Slotrun-20260817-133237-ca9a7a
246,8 tok/s
NVIDIA GB10run-20260820-194812-a67646
169,9 tok/s
NVIDIA GeForce RTX 5070 Tirun-20260729-032127-c9386f
152,7 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: Qwen3-30B-A3B-Instruct-2507
Configuration
# LLM-Benchmark Konfiguration
# Modell : Qwen3-30B-A3B-Instruct-2507
# Engine : llama.cpp
# Run-ID : run-20260818-134249-787b11
# 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/Qwen_Qwen3-30B-A3B-Instruct-2507-Q4_K_M.gguf \
--alias Qwen3-30B-A3B-Instruct-2507 \
-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. | Qwen3-30B-A3B-Instruct-2507 |
| 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/Qwen_Qwen3-30B-A3B-Instruct-2507-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 |
| fa | on |
| 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 |
| np | 12 |
Anzeige
Model comparison
Qwen3-30B-A3B-Instruct-2507 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 1.671,0 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 1.658,1 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1.515,7 tok/sNVIDIA GeForce RTX 3090 Ti 905,1 tok/s★ AMD Radeon PRO W7900 Dual Slot 264,2 tok/s this runNVIDIA GB10 (DGX Spark) 169,9 tok/sNVIDIA GeForce RTX 5070 Ti 152,7 tok/sIntel Arc Pro B70 143,6 tok/sAMD Radeon 8060S Graphics 107,6 tok/sAMD Radeon AI PRO R9700 97,6 tok/sCPU-only 7,1 tok/s
CPUby processor
AMD Ryzen 7 5800X3D 8-Core Processor 1.671,0 tok/sAMD Ryzen 9 9950X 16-Core Processor 1.658,1 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 1.515,7 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 905,1 tok/s★ AMD Ryzen Threadripper PRO 5975WX 32-Cores 264,2 tok/s this runNVIDIA Grace 169,9 tok/sAMD Ryzen 9 7945HX with Radeon Graphics 143,6 tok/sAMD RYZEN AI MAX+ 395 w/ Radeon 8060S 107,6 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 97,6 tok/sIntel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 7,1 tok/s
MBby mainboard
ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 1.671,0 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.658,1 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 1.515,7 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 905,1 tok/s★ ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 264,2 tok/s this runASUSTeK COMPUTER INC. GX10 169,9 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd DRFXI (MotherBoard Series) 143,6 tok/sBosgame AXB35-02 (BeyondMax Series) 107,6 tok/sDell Inc. PowerEdge R820 7,1 tok/s
DRVby driver
unbekannt 1.671,0 tok/sNVIDIA 590.48.01 / CUDA 13.1 169,9 tok/sIntel 26.18.38308.4 143,6 tok/sAMD 7.0.0-27-generic 97,6 tok/s
💰 Economics
Economics of this run
Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (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
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 0.19
Token / kWh1.58M
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)16.66B
☁️ 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
| Qwen3-30B-A3B-Instruct-25072x AMD Radeon PRO W7900 Dual Slot | Qwen3-30B-A3B-Instruct-2507NVIDIA GeForce RTX 5090 | Qwen3-30B-A3B-Instruct-2507NVIDIA RTX PRO 6000 Blackwell Workstation Edition | Qwen3-30B-A3B-Instruct-25073x 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) | – | – | – | – |
2× AMD Radeon PRO W7900 Dual Slot · this run 264 tok/sNVIDIA GeForce RTX 5090 1,671 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 1,658 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1,516 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.