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

Qwen3.6-27B

Performance benchmark · measured on 29.08.2026 19:04

Benchmark-IDrun-20260829-173720-b0d023
Timebench 3 - Kombi (Prefill + Generation)Dense27BRuntime: llama.cppQuantisierung: Q8_0
Generation23,27tok/s
Prefill1.218,94tok/s
Time to First Token1.585,50ms
Total duration91,19s
Concurrency1parallel
Ranking in the field
261of 277 systems

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

This run is better than 6 % of all comparable systems.
Generation 23,3 tok/s
-76 % vs Ø 99,0
Prefill 1.218,9 tok/s
-75 % vs Ø 4.970,6
Time to First Token 1.586 ms
+125 % vs Ø 703
Distribution in the field5 – 218 tok/s
Ø 99 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: llama.cpp
Quantization: Q8_0
Model: Qwen3.6-27B

Anmerkung

NVLink installiert: 2x RTX A6000 ueber NVLink-Bridge (4x NVLink) verbunden.

Configuration

benchmark-konfiguration — run-20260829-173720-b0d023
# LLM-Benchmark Konfiguration # Modell : Qwen3.6-27B # Engine : llama.cpp # Run-ID : run-20260829-173720-b0d023 # GPU : 2x NVIDIA RTX A6000 # CPU : AMD Ryzen Threadripper PRO 7955WX 16-Cores # RAM : 184 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build/bin/llama-server \ -m Qwen3.6-27B-Q8_0.gguf \ -ngl 999 \ -fa on \ -c 32768 \ -np 8 \ -sm layer '(2x' RTX A6000 48GB 'NVLink)'
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
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.32768
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird.Qwen3.6-27B-Q8_0.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
faon
Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells.32768
np8
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

Anzeige
Model comparison

Qwen3.6-27B 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

7065293531760,001.1052.2103.315Prefill (tok/s)Generation (tok/s)NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 546,9 tok/s Generation, 2.739 tok/s Prefill, TTFT 9.126 ms (3 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5090 - 532,5 tok/s Generation, 2.539 tok/s Prefill, TTFT 9.812 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 464,1 tok/s Generation, 2.737 tok/s Prefill, TTFT 9.091 ms (11 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 3090 Ti - 264,2 tok/s Generation, 1.458 tok/s Prefill, TTFT 19.358 ms (3 Laufe)NVIDIA GeForce RTX 30...AMD Radeon AI PRO R9700 - 165,4 tok/s Generation, 1.380 tok/s Prefill, TTFT 28.271 ms (9 Laufe)AMD Radeon AI PRO R97...NVIDIA GeForce RTX 5070 Ti - 57,2 tok/s Generation, 447 tok/s Prefill, TTFT 80.318 ms (3 Laufe)NVIDIA GeForce RTX 50...AMD Radeon 8060S Graphics - 36,1 tok/s Generation, 536 tok/s Prefill, TTFT 41.917 ms (1 Lauf)AMD Radeon 8060S Grap...AMD Radeon PRO W7800 48GB - 25,9 tok/s Generation, 570 tok/s Prefill, TTFT 3.308 ms (1 Lauf)AMD Radeon PRO W7800 ...NVIDIA RTX A6000 - 431,7 tok/s Generation, 2.490 tok/s Prefill, TTFT 7.603 ms (24 Laufe) | DIESER LAUF★ NVIDIA RTX A6000
NVIDIA RTX PRO 6000 Blackwell Workstation Edition 546,9 tok/sNVIDIA GeForce RTX 5090 532,5 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 464,1 tok/s★ NVIDIA RTX A6000 431,7 tok/s this runNVIDIA GeForce RTX 3090 Ti 264,2 tok/sAMD Radeon AI PRO R9700 165,4 tok/sNVIDIA GeForce RTX 5070 Ti 57,2 tok/sAMD Radeon 8060S Graphics 36,1 tok/sAMD Radeon PRO W7800 48GB 25,9 tok/s

CPUby processor

7045283521760,001.2992.5983.897Prefill (tok/s)Generation (tok/s)AMD Ryzen 9 9950X 16-Core Processor - 546,9 tok/s Generation, 1.883 tok/s Prefill, TTFT 5.524 ms (5 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen 7 5800X3D 8-Core Processor - 532,5 tok/s Generation, 2.539 tok/s Prefill, TTFT 9.812 ms (3 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 464,1 tok/s Generation, 3.212 tok/s Prefill, TTFT 11.084 ms (9 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 264,2 tok/s Generation, 1.458 tok/s Prefill, TTFT 19.358 ms (3 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 57,2 tok/s Generation, 478 tok/s Prefill, TTFT 61.065 ms (4 Laufe)AMD Ryzen Threadrippe...AMD RYZEN AI MAX+ 395 w/ Radeon 8060S - 36,1 tok/s Generation, 536 tok/s Prefill, TTFT 41.917 ms (1 Lauf)AMD RYZEN AI MAX+ 395...AMD Ryzen Threadripper PRO 7955WX 16-Cores - 431,7 tok/s Generation, 2.187 tok/s Prefill, TTFT 13.240 ms (33 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
AMD Ryzen 9 9950X 16-Core Processor 546,9 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 532,5 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 464,1 tok/s★ AMD Ryzen Threadripper PRO 7955WX 16-Cores 431,7 tok/s this runAMD Ryzen 9 8945HX with Radeon Graphics 264,2 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 57,2 tok/sAMD RYZEN AI MAX+ 395 w/ Radeon 8060S 36,1 tok/s

MBby mainboard

7055293531760,001.0212.0413.062Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 546,9 tok/s Generation, 2.204 tok/s Prefill, TTFT 6.875 ms (4 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 532,5 tok/s Generation, 2.539 tok/s Prefill, TTFT 9.812 ms (3 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 264,2 tok/s Generation, 1.458 tok/s Prefill, TTFT 19.358 ms (3 Laufe)Meigao Innovation Tec...ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 57,2 tok/s Generation, 478 tok/s Prefill, TTFT 61.065 ms (4 Laufe)ASUSTeK COMPUTER INC....Bosgame AXB35-02 (BeyondMax Series) - 36,1 tok/s Generation, 536 tok/s Prefill, TTFT 41.917 ms (1 Lauf)Bosgame AXB35-02 (Bey...unbekannt - 28,8 tok/s Generation, 599 tok/s Prefill, TTFT 122 ms (1 Lauf)unbekanntASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 464,1 tok/s Generation, 2.407 tok/s Prefill, TTFT 12.778 ms (42 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 546,9 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 532,5 tok/s★ ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 464,1 tok/s this runMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 264,2 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 57,2 tok/sBosgame AXB35-02 (BeyondMax Series) 36,1 tok/sunbekannt 28,8 tok/s

ENGby engine

7055293531760,001.6393.2784.917Prefill (tok/s)Generation (tok/s)vLLM - 431,7 tok/s Generation, 4.052 tok/s Prefill, TTFT 2.692 ms (10 Laufe)vLLMunbekannt - 28,8 tok/s Generation, 599 tok/s Prefill, TTFT 122 ms (1 Lauf)unbekanntllama.cpp - 546,9 tok/s Generation, 1.783 tok/s Prefill, TTFT 19.382 ms (47 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 546,9 tok/s this runvLLM 431,7 tok/sunbekannt 28,8 tok/s

DRVby driver

7055293531760,01849881.7932.597Prefill (tok/s)Generation (tok/s)unbekannt - 546,9 tok/s Generation, 2.181 tok/s Prefill, TTFT 16.454 ms (57 Laufe)unbekanntNVIDIA 580.159.03 / CUDA 13.0 - 28,8 tok/s Generation, 599 tok/s Prefill, TTFT 122 ms (1 Lauf)NVIDIA 580.159.03 / C...
unbekannt 546,9 tok/sNVIDIA 580.159.03 / CUDA 13.0 28,8 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 2.40
Token / kWh124.85K
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)1.47B
☁️ 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

Qwen3.6-27B2x NVIDIA RTX A6000Qwen3.6-27BNVIDIA GeForce RTX 5090Qwen3.6-27BNVIDIA RTX PRO 6000 Blackwell Workstation EditionQwen3.6-27B3x 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

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.