Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
Contributed byMario AlkaQwen (Alibaba)

Qwen3-0.6B

Performance benchmark · measured on 25.09.2026 16:36

Benchmark-IDrun-20261020-051233-ef1fce
Timebench 3 - Kombi (Prefill + Generation)Dense0.6BRuntime: vLLMQuantisierung: FP8
Generation503,33tok/s
Prefill112.346,36tok/s
Time to First Token241,50ms
Total duration41,23s
Concurrency10parallel
Ranking in the field
395of 1427 systems

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

This run is better than 72 % of all comparable systems.
Generation 503,3 tok/s
+18 % vs Ø 425,3
Prefill 112.346,4 tok/s
+1.587 % vs Ø 6.659,3
Time to First Token 242 ms
-99 % vs Ø 45.848
Distribution in the field0 – 2.491 tok/s
Ø 425 Median Dieser Lauf

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-3.5-Lightning-30B-A3BNVIDIA GeForce RTX 3090 Ti · run-20261021-093815-dce558
2.263,7 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× 3x 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× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-0aed86
1.937,7 tok/s
Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-8db0fa
1.911,6 tok/s
Nemotron-3-Nano-4B3× 3x 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× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-034052-26063b
1.835,7 tok/s
Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-a77046
1.830,3 tok/s
Qwen3-0.6B this runNVIDIA GeForce RTX 4090 · run-20261020-051233-ef1fce
503,3 tok/s

How does this benchmark compare on other GPUs?

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

Hardware

GPU: NVIDIA GeForce RTX 4090 · 24 GB VRAM
CPU: AMD Ryzen 3 3100 4-Core Processor
RAM: 61 GB
Mainboard: Micro-Star International Co., Ltd. B550M PRO-VDH WIFI (MS-7C95)

Setup

Runtime: vLLM
Quantization: FP8
Model: Qwen3-0.6B

Configuration

benchmark-konfiguration — run-20261020-051233-ef1fce
# LLM-Benchmark Konfiguration # Modell : Qwen3-0.6B # Engine : vLLM # Run-ID : run-20261020-051233-ef1fce # GPU : NVIDIA GeForce RTX 4090 # CPU : AMD Ryzen 3 3100 4-Core Processor # RAM : 61 GB bench@llm-benchmark:~$ /home/godcore/benchmark/vllm-env/bin/python \ -m vllm.entrypoints.openai.api_server \ --model /home/godcore/benchmark/model-cache/2a4df9b2f3602180 \ --served-model-name Qwen3-0.6B \ --host 0.0.0.0 \ --port 8000 \ --max-model-len 4096 \ --max-num-seqs 10 \ --max-num-batched-tokens 4096 \ --gpu-memory-utilization 0.90 \ --enforce-eager
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.Qwen3-0.6B
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.4096
CUDA_VISIBLE_DEVICES0
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird./home/godcore/benchmark/model-cache/2a4df9b2f3602180
VLLM_USE_FLASHINFER_SAMPLER0
Python-Modul?Startet vLLM als Python-Modul (python -m vllm.entrypoints.openai.api_server). Der folgende Wert ist der Modulname.vllm.entrypoints.openai.api_server
Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen.Qwen3-0.6B
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.4096
Max-Sequenzen?Maximale Anzahl gleichzeitig bearbeiteter Anfragen (Batch-Groesse). Hoeher = mehr Durchsatz, braucht aber mehr KV-Cache-Speicher.10
Max-Batch-Tokens?Maximale Anzahl Tokens, die pro Rechenschritt gebuendelt verarbeitet werden. Beeinflusst Prefill-Durchsatz und Latenz.4096
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'.0.90
execution_typelocal

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Qwen3-0.6B 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

7285463641820,0029.58059.16088.740Prefill (tok/s)Generation (tok/s)AMD Radeon AI PRO R9700 - 561,4 tok/s Generation, 20.958 tok/s Prefill, TTFT 573 ms (3 Laufe)AMD Radeon AI PRO R97...AMD Radeon RX 7900 XTX - 495,5 tok/s Generation, 22.637 tok/s Prefill, TTFT 718 ms (8 Laufe)AMD Radeon RX 7900 XTXAMD Oberon (PlayStation 5) - 302,6 tok/s Generation, 2.983 tok/s Prefill, TTFT 2.304 ms (2 Laufe)AMD Oberon (PlayStati...CPU-only - 8,0 tok/s Generation, 74 tok/s Prefill, TTFT 126.155 ms (4 Laufe)CPU-onlyNVIDIA GeForce RTX 4090 - 503,3 tok/s Generation, 71.575 tok/s Prefill, TTFT 174 ms (3 Laufe) | DIESER LAUF★ NVIDIA GeForce RTX 40...
AMD Radeon AI PRO R9700 561,4 tok/s★ NVIDIA GeForce RTX 4090 503,3 tok/s this runAMD Radeon RX 7900 XTX 495,5 tok/sAMD Oberon (PlayStation 5) 302,6 tok/sCPU-only 8,0 tok/s

CPUby processor

7295473651820,0019.12238.24457.366Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 3955WX 16-Cores - 495,5 tok/s Generation, 27.231 tok/s Prefill, TTFT 657 ms (5 Laufe)AMD Ryzen Threadrippe...Intel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz - 480,5 tok/s Generation, 14.980 tok/s Prefill, TTFT 819 ms (3 Laufe)Intel(R) Xeon(R) CPU ...AMD Zen2 8-Core (PlayStation 5) - 302,6 tok/s Generation, 2.983 tok/s Prefill, TTFT 2.304 ms (2 Laufe)AMD Zen2 8-Core (Play...4 x Cortex-A78C + 4 x Cortex-A55 - 8,0 tok/s Generation, 90 tok/s Prefill, TTFT 132.604 ms (3 Laufe)4 x Cortex-A78C + 4 x...AMD Jaguar 8-Core (PlayStation 4) - 3,8 tok/s Generation, 23 tok/s Prefill, TTFT 106.805 ms (1 Lauf)AMD Jaguar 8-Core (Pl...AMD Ryzen 3 3100 4-Core Processor - 561,4 tok/s Generation, 46.266 tok/s Prefill, TTFT 374 ms (6 Laufe) | DIESER LAUF★ AMD Ryzen 3 3100 4-Co...
★ AMD Ryzen 3 3100 4-Core Processor 561,4 tok/s this runAMD Ryzen Threadripper PRO 3955WX 16-Cores 495,5 tok/sIntel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz 480,5 tok/sAMD Zen2 8-Core (PlayStation 5) 302,6 tok/s4 x Cortex-A78C + 4 x Cortex-A55 8,0 tok/sAMD Jaguar 8-Core (PlayStation 4) 3,8 tok/s

MBby mainboard

669551432313195029.52259.04488.565Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ROG STRIX X570-F GAMING - 561,4 tok/s Generation, 20.958 tok/s Prefill, TTFT 573 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 495,5 tok/s Generation, 27.231 tok/s Prefill, TTFT 657 ms (5 Laufe)ASUSTeK COMPUTER INC....Dell Inc. PowerEdge R820 - 480,5 tok/s Generation, 14.980 tok/s Prefill, TTFT 819 ms (3 Laufe)Dell Inc. PowerEdge R...unbekannt - 302,6 tok/s Generation, 1.043 tok/s Prefill, TTFT 84.871 ms (6 Laufe)unbekanntMicro-Star International Co., Ltd. B550M PRO-VDH WIFI (MS-7C95) - 503,3 tok/s Generation, 71.575 tok/s Prefill, TTFT 174 ms (3 Laufe) | DIESER LAUF★ Micro-Star Internatio...
ASUSTeK COMPUTER INC. ROG STRIX X570-F GAMING 561,4 tok/s★ Micro-Star International Co., Ltd. B550M PRO-VDH WIFI (MS-7C95) 503,3 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 495,5 tok/sDell Inc. PowerEdge R820 480,5 tok/sunbekannt 302,6 tok/s

ENGby engine

629581532484436021.26142.52263.782Prefill (tok/s)Generation (tok/s)llama.cpp - 561,4 tok/s Generation, 10.512 tok/s Prefill, TTFT 36.729 ms (14 Laufe)llama.cppvLLM - 503,3 tok/s Generation, 52.963 tok/s Prefill, TTFT 499 ms (6 Laufe) | DIESER LAUF★ vLLM
llama.cpp 561,4 tok/s★ vLLM 503,3 tok/s this run

DRVby driver

61859056153350521.85322.78323.71324.643Prefill (tok/s)Generation (tok/s)unbekannt - 561,4 tok/s Generation, 23.248 tok/s Prefill, TTFT 25.860 ms (20 Laufe)unbekannt
unbekannt 561,4 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 0 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 0.075
Token / kWh4.03M
Acquisition (system)EUR 2,150 partial priceGPU EUR 2,000 · PSU EUR 150
Electricity (2 years)–
TCO (2 years)EUR 4,515
Output tokens (2 years)31.75B
☁️ 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 (0 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

Qwen3-0.6BNVIDIA GeForce RTX 4090Qwen3-0.6BAMD Radeon AI PRO R9700Qwen3-0.6B2x AMD Radeon RX 7900 XTXQwen3-0.6B2x AMD Radeon RX 7900 XTX
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.