Created bymario-alka.dePowered bynoob2claw.detricoma.de
Contributed byMario AlkaQwen (Alibaba)

Qwen2.5-32B-Instruct-AWQ

Performance benchmark · measured on 09.08.2026 12:21

Benchmark-IDrun-20260809-131004-0c13a0
Timebench 3 - Kombi (Prefill + Generation)Dense32BRuntime: llama.cppQuantisierung: AWQ
Generation40,18tok/s
Prefill1.443,85tok/s
Time to First Token1.704,50ms
Total duration54,39s
Concurrency1parallel
Ranking in the field
57of 115 systems

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

This run is better than 51 % of all comparable systems.
Generation 40,2 tok/s
-45 % vs Ø 72,7
Prefill 1.443,9 tok/s
-36 % vs Ø 2.240,7
Time to First Token 1.705 ms
-99 % vs Ø 124.661
Distribution in the field1 – 246 tok/s
Ø 73 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

Configuration

benchmark-konfiguration — run-20260809-131004-0c13a0
# LLM-Benchmark Konfiguration # Modell : Qwen2.5-32B-Instruct-AWQ # Engine : llama.cpp # Run-ID : run-20260809-131004-0c13a0 # GPU : NVIDIA GeForce RTX 3090 Ti # CPU : AMD Ryzen 9 8945HX with Radeon Graphics # RAM : 92 GB bench@llm-benchmark:~$ /root/llama.cpp/build/bin/llama-server \ -m /root/.cache/huggingface/hub/models--bartowski--Qwen2.5-32B-Instruct-GGUF/snapshots/2116cbb385b8ce3a4d28cf3bf1cd2039a55821a6/Qwen2.5-32B-Instruct-Q4_K_M.gguf \ --alias Qwen2.5-32B-Instruct-AWQ \ --host 0.0.0.0 \ --port 8000 \ -ngl 999 \ -c 16384 \ -np 4 \ --jinja
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.Qwen2.5-32B-Instruct-AWQ
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.16384
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird./root/.cache/huggingface/hub/models--bartowski--Qwen2.5-32B-Instruct-GGUF/snapshots/2116cbb385b8ce3a4d28cf3bf1cd2039a55821a6/Qwen2.5-32B-Instruct-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.16384
np4
execution_typelocal

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Qwen2.5-32B-Instruct-AWQ 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

2442332222112001.9262.0082.0902.172Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 3090 Ti - 222,0 tok/s Generation, 2.049 tok/s Prefill, TTFT 20.300 ms (3 Laufe) | DIESER LAUF★ NVIDIA GeForce RTX 30...
★ NVIDIA GeForce RTX 3090 Ti 222,0 tok/s this run

CPUby processor

2442332222112001.9262.0082.0902.172Prefill (tok/s)Generation (tok/s)AMD Ryzen 9 8945HX with Radeon Graphics - 222,0 tok/s Generation, 2.049 tok/s Prefill, TTFT 20.300 ms (3 Laufe) | DIESER LAUF★ AMD Ryzen 9 8945HX wi...
★ AMD Ryzen 9 8945HX with Radeon Graphics 222,0 tok/s this run

MBby mainboard

2442332222112001.9262.0082.0902.172Prefill (tok/s)Generation (tok/s)Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 222,0 tok/s Generation, 2.049 tok/s Prefill, TTFT 20.300 ms (3 Laufe) | DIESER LAUF★ Meigao Innovation Tec...
★ Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 222,0 tok/s this run

ENGby engine

2442332222112001.9262.0082.0902.172Prefill (tok/s)Generation (tok/s)llama.cpp - 222,0 tok/s Generation, 2.049 tok/s Prefill, TTFT 20.300 ms (3 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 222,0 tok/s this run

DRVby driver

2442332222112001.9262.0082.0902.172Prefill (tok/s)Generation (tok/s)unbekannt - 222,0 tok/s Generation, 2.049 tok/s Prefill, TTFT 20.300 ms (3 Laufe)unbekannt
unbekannt 222,0 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 50 W
⚡ TDP 477 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)477 W estimated (TDP)GPU 450 + CPU 17 + Board 10 W full load
Avg cost / hourEUR 0.14
Electricity / 1M tokensEUR 0.99
Token / kWh303.09K
Acquisition (system)EUR 2,986 partial priceGPU EUR 999 · CPU EUR 549 · RAM EUR 1,288 · PSU EUR 150
Electricity (2 years)–
TCO (2 years)EUR 5,494
Output tokens (2 years)2.53B
☁️ 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 (50 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

Qwen2.5-32B-Instruct-AWQNVIDIA GeForce RTX 3090 Ti
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.