Contributed byMario AlkainclusionAI Ling-3.0-flash
Performance benchmark · measured on 22.09.2026 17:38
Benchmark-ID
run-20260922-155544-996c21Timebench 3 - Kombi (Prefill + Generation)MoE127.5BRuntime: llama.cppQuantisierung: Q4_K_M
Generation92,62tok/s
Prefill1.184,57tok/s
Time to First Token24.873,00ms
Total duration275,96s
Concurrency10parallel
Ranking in the field
1196of 1491 systems
Performance benchmark · Primary metric: Generation-Speed (tok/s) · 10× concurrent
This run is better than 20 % of all comparable systems.
Generation
92,6 tok/s
-79 % vs Ø 448,4
Prefill
1.184,6 tok/s
-83 % vs Ø 6.851,4
Time to First Token
24.873 ms
-44 % vs Ø 44.166
Distribution in the field0 – 3.484 tok/s
Wie schlägt sich dieser Benchmark mit anderen Modellen?
NVIDIA-Nemotron-3.5-Lightning-30B-A3BNVIDIA GeForce RTX 3090 · run-20261026-180644-063abd
3.484,3 tok/s
NVIDIA-Nemotron-3.5-Lightning-30B-A3BNVIDIA GeForce RTX 3090 · run-20261026-183943-bbe82c
2.867,2 tok/s
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
gemma-4-E2B-itNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-065752-48370d
2.412,3 tok/s
NVIDIA-Nemotron-3.5-Lightning-30B-A3BNVIDIA GeForce RTX 3090 Ti · run-20261021-093815-dce558
2.263,7 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260727-181607-4cca42
2.182,7 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-065758-96fe6a
2.144,3 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
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-065808-332d1f
1.981,1 tok/s
NVIDIA-Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-0aed86
1.937,7 tok/s
NVIDIA-Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-8db0fa
1.911,6 tok/s
Ling-3.0-flash this run2× 2x NVIDIA RTX A6000 · run-20260922-155544-996c21
92,6 tok/s
How does this benchmark compare on other GPUs?
Same model on different hardware · 10× concurrent · Generation (tok/s)
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20261024-140625-14bd58
697,4 tok/s
2× NVIDIA RTX A6000run-20260930-145235-98a0bb
388,1 tok/s
NVIDIA DGX SPARK (GB10)run-20260820-173706-52a2df
253,7 tok/s
NVIDIA DGX SPARK (GB10)run-20260820-131510-16a6a0
250,7 tok/s
NVIDIA DGX SPARK (GB10)run-20260820-131510-6c0f64
247,8 tok/s
2× NVIDIA RTX A6000run-20261001-120303-00c4f3
201,3 tok/s
2× NVIDIA RTX A6000run-20261001-081040-84009f
197,9 tok/s
2× NVIDIA RTX A6000run-20260925-162606-50993f
194,7 tok/s
2× AMD Radeon PRO W7900 Dual Slotrun-20260820-035952-59e76b
165,7 tok/s
2× AMD Radeon PRO W7900 Dual Slotrun-20260820-061102-35573a
165,4 tok/s
2× AMD Radeon PRO W7900 Dual Slotrun-20260820-050509-adfd8e
164,7 tok/s
4× AMD Radeon AI PRO R9700run-20260820-035952-e63c91
160,2 tok/s
4× AMD Radeon AI PRO R9700run-20260820-050509-b12bbc
159,6 tok/s
4× AMD Radeon AI PRO R9700run-20260818-071252-5ea375
141,1 tok/s
Hardware
GPU: 2x NVIDIA RTX A6000 · 48 GB VRAM
CPU: AMD EPYC 7203P 8-Core Processor
RAM: 31 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI
Setup
Runtime: llama.cpp
Quantization: Q4_K_M
Model: Ling-3.0-flash
Configuration
# LLM-Benchmark Konfiguration
# Modell : Ling-3.0-flash
# Engine : llama.cpp
# Run-ID : run-20260922-155544-996c21
# GPU : 2x NVIDIA RTX A6000
# CPU : AMD EPYC 7203P 8-Core Processor
# RAM : 31 GB
bench@llm-benchmark:~$ llama-server \
-m /home/godcore/models/Ling-3.0-flash-Q4_K_M.gguf \
--alias Ling-3.0-flash \
-ngl 999 \
-sm layer \
-fa on \
-c 49152 \
-np 12 '(2x' NVIDIA RTX A6000 48GB, 'CUDA)'
| 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. | Ling-3.0-flash |
| 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/models/Ling-3.0-flash-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 |
| Split-Mode?Verteilung ueber mehrere GPUs: none (nur eine GPU), layer (Layer aufteilen) oder row (Tensoren zeilenweise). | layer |
| fa | on |
| Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells. | 49152 |
| np | 12 |
| execution_type | local |
Anzeige
Model comparison
Ling-3.0-flash 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
CPUby processor
AMD Ryzen 9 7945HX with Radeon Graphics 697,4 tok/s★ AMD EPYC 7203P 8-Core Processor 388,1 tok/s this runNVIDIA Grace 253,7 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 165,7 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 160,2 tok/sAMD Ryzen Threadripper PRO 3955WX 16-Cores 133,0 tok/sIntel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz 14,9 tok/s
MBby mainboard
Shenzhen Meigao Electronic Equipment Co.,Ltd DRFXI (MotherBoard Series) 697,4 tok/sASRockRack ROMED8-2T/BCM 388,1 tok/sASUSTeK COMPUTER INC. GX10 253,7 tok/s★ ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 194,7 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 160,2 tok/sDell Inc. PowerEdge R820 14,9 tok/s
DRVby driver
unbekannt 697,4 tok/sNVIDIA 590.48.01 / CUDA 13.1 253,7 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
40 W
⚡ TDP
610 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)610 W estimated (TDP)GPU 600 + Board 10 W full load
Avg cost / hourEUR 0.18
Electricity / 1M tokensEUR 0.55
Token / kWh546.61K
Acquisition (system)EUR 6,428 partial priceGPU EUR 6,000 · RAM EUR 248 · PSU EUR 180
Electricity (2 years)–
TCO (2 years)EUR 9,634
Output tokens (2 years)5.84B
☁️ 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 (40 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
| Ling-3.0-flash2x NVIDIA RTX A6000 | Ling-3.0-flashNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | Ling-3.0-flash2x NVIDIA RTX A6000 | Ling-3.0-flashNVIDIA DGX SPARK (GB10) | |
|---|---|---|---|---|
| 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× NVIDIA RTX A6000 · this run 93 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 697 tok/s2× NVIDIA RTX A6000 388 tok/sNVIDIA DGX SPARK (GB10) 254 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.