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

Qwen3.8-27B

Performance benchmark · measured on 27.08.2026 11:44

Benchmark-IDrun-20260827-101402-605dd8
Timebench 3 - Kombi (Prefill + Generation)Qwen3_5ForConditionalGeneration (multimodal, image-text-to-text)27BRuntime: vLLMQuantisierung: INT4
Generation58,10tok/s
Prefill2.529,39tok/s
Time to First Token779,00ms
Total duration36,81s
Concurrency1parallel
Ranking in the field
553of 1311 systems

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

This run is better than 58 % of all comparable systems.
Generation 58,1 tok/s
-24 % vs Ø 76,5
Prefill 2.529,4 tok/s
-3 % vs Ø 2.615,8
Time to First Token 779 ms
-98 % vs Ø 31.676
Distribution in the field0 – 405 tok/s
Ø 77 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 NVIDIA RTX A6000 · 48 GB VRAM
CPU: AMD Ryzen Threadripper PRO 7955WX 16-Cores
RAM: 184 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE

Setup

Runtime: vLLM
Quantization: INT4
Model: Qwen3.8-27B

Configuration

benchmark-konfiguration — run-20260827-101402-605dd8
# LLM-Benchmark Konfiguration # Modell : Qwen3.8-27B # Engine : vLLM # Run-ID : run-20260827-101402-605dd8 # GPU : 2x NVIDIA RTX A6000 # CPU : AMD Ryzen Threadripper PRO 7955WX 16-Cores # RAM : 184 GB bench@llm-benchmark:~$ vllm \ --tensor-parallel-size 2 \ --max-model-len 16384 '(2x' RTX A6000 '48GB)'
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
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.16384
Tensor-Parallel?Anzahl GPUs, auf die JEDER einzelne Modell-Layer aufgeteilt wird (Tensor-Parallelitaet). Mehr GPUs = mehr VRAM und meist mehr Speed, aber die Anzahl der Attention-Heads muss durch diesen Wert teilbar sein (z.B. 32 Heads -> nur 1, 2, 4, 8 ... moeglich, NICHT 3).2
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.16384

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Qwen3.8-27B 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

4523793062331601.6232.6313.6384.646Prefill (tok/s)Generation (tok/s)AMD Radeon AI PRO R9700 - 229,4 tok/s Generation, 2.202 tok/s Prefill, TTFT 2.623 ms (36 Laufe)AMD Radeon AI PRO R97...NVIDIA RTX A6000 - 383,3 tok/s Generation, 4.067 tok/s Prefill, TTFT 3.195 ms (9 Laufe) | DIESER LAUF★ NVIDIA RTX A6000
★ NVIDIA RTX A6000 383,3 tok/s this runAMD Radeon AI PRO R9700 229,4 tok/s

CPUby processor

4224023833643452.4212.5242.6272.730Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 7955WX 16-Cores - 383,3 tok/s Generation, 2.575 tok/s Prefill, TTFT 2.738 ms (45 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
★ AMD Ryzen Threadripper PRO 7955WX 16-Cores 383,3 tok/s this run

MBby mainboard

4224023833643452.4212.5242.6272.730Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 383,3 tok/s Generation, 2.575 tok/s Prefill, TTFT 2.738 ms (45 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
★ ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 383,3 tok/s this run

ENGby engine

47736124512913,41.2672.1773.0873.997Prefill (tok/s)Generation (tok/s)llama.cpp - 107,0 tok/s Generation, 1.782 tok/s Prefill, TTFT 1.546 ms (24 Laufe)llama.cppvLLM - 383,3 tok/s Generation, 3.482 tok/s Prefill, TTFT 4.099 ms (21 Laufe) | DIESER LAUF★ vLLM
★ vLLM 383,3 tok/s this runllama.cpp 107,0 tok/s

DRVby driver

4224023833643452.4212.5242.6272.730Prefill (tok/s)Generation (tok/s)unbekannt - 383,3 tok/s Generation, 2.575 tok/s Prefill, TTFT 2.738 ms (45 Laufe)unbekannt
unbekannt 383,3 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 95 W
⚡ TDP 671 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)671 W estimated (TDP)GPU 600 + CPU 46 + Board 25 W full load
Avg cost / hourEUR 0.20
Electricity / 1M tokensEUR 0.96
Token / kWh311.71K
Acquisition (system)EUR 11,454 full priceGPU EUR 6,000 · CPU EUR 1,399 · Board EUR 1,299 · RAM EUR 2,576 · PSU EUR 180
Electricity (2 years)
TCO (2 years)EUR 14,981
Output tokens (2 years)3.66B
☁️ 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 (95 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

Qwen3.8-27B2x NVIDIA RTX A6000Qwen3.8-27B2x NVIDIA RTX A6000Qwen3.8-27B4x AMD Radeon AI PRO R9700Qwen3.8-27BNVIDIA RTX A6000
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.