Contributed byMario AlkaMeta Meta-Llama-3.1-8B-Instruct
Performance benchmark · measured on 22.07.2026 18:53
Benchmark-ID
run-20260723-194429-b2114cTimebench 3 - Kombi (Prefill + Generation)Dense8BRuntime: llama.cppQuantisierung: Q4_K_M
Generation3,55tok/s
Prefill35,95tok/s
Time to First Token59.043,50ms
Total duration600,00s
Concurrency1parallel
Ranking in the field
26of 33 systems
Performance benchmark · Primary metric: Generation-Speed (tok/s) · 1× concurrent
This run is better than 22 % of all comparable systems.
Generation
3,6 tok/s
-84 % vs Ø 21,9
Prefill
36,0 tok/s
-94 % vs Ø 605,3
Time to First Token
59.044 ms
+32 % vs Ø 44.665
Distribution in the field1 – 81 tok/s
Wie schlägt sich dieser Benchmark mit anderen Modellen?
Nemotron-3-Nano-4B2× NVIDIA GeForce RTX 2060 · run-20260822-084953-2c6a8a
80,9 tok/s
Nemotron-3-Nano-4B2× NVIDIA GeForce RTX 2060 · run-20260822-084953-c06809
72,4 tok/s
Nemotron-3-Nano-4B2× NVIDIA GeForce RTX 2060 · run-20260820-184209-66eec2
71,7 tok/s
gemma-4-E4B-it2× NVIDIA GeForce RTX 2060 · run-20260820-191511-0f4886
53,4 tok/s
Meta-Llama-3.1-8B-Instruct same model2× NVIDIA GeForce RTX 2060 · run-20260822-084953-8baabe
49,0 tok/s
Meta-Llama-3.1-8B-Instruct same model2× NVIDIA GeForce RTX 2060 · run-20260820-194812-6808ae
47,4 tok/s
Meta-Llama-3.1-8B-Instruct same model2× NVIDIA GeForce RTX 2060 · run-20260822-084953-32cd71
45,7 tok/s
Ornith-1.0-9B2× NVIDIA GeForce RTX 2060 · run-20260820-194812-5e259d
42,4 tok/s
gemma-4-E4B-it2× NVIDIA GeForce RTX 2060 · run-20260820-205402-e2e0f8
38,3 tok/s
Meta-Llama-3.1-8B-Instruct same model2× NVIDIA GeForce RTX 2060 · run-20260820-205403-83fca2
31,8 tok/s
gemma-4-12B-it2× NVIDIA GeForce RTX 2060 · run-20260820-194812-450bed
30,3 tok/s
Ornith-1.0-9B2× NVIDIA GeForce RTX 2060 · run-20260820-212718-3dd8f1
29,6 tok/s
Ministral-3-14B-Reasoning-25122× NVIDIA GeForce RTX 2060 · run-20260820-194812-c4edf3
29,1 tok/s
gemma-4-E2B-it2× NVIDIA GeForce RTX 2060 · run-20260820-184209-7a88fc
15,3 tok/s
Meta-Llama-3.1-8B-Instruct this run2× NVIDIA GeForce RTX 2060 · run-20260723-194429-b2114c
3,6 tok/s
How does this benchmark compare on other GPUs?
Same model on different hardware · 1× concurrent · Generation (tok/s)
NVIDIA GeForce RTX 5090run-20260729-060803-e57274
246,2 tok/s
NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260728-194136-f84bc1
239,3 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-060805-e19fd2
225,0 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-060805-934ac4
222,9 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-060805-db5fff
212,8 tok/s
2× NVIDIA RTX A6000run-20260829-095458-1f7681
181,2 tok/s
NVIDIA GeForce RTX 3090 Tirun-20260729-032126-398aa1
148,2 tok/s
NVIDIA GeForce RTX 5070 Tirun-20260728-194135-5497cd
140,9 tok/s
2× NVIDIA RTX A6000run-20260829-102814-abffa3
127,4 tok/s
2× NVIDIA RTX A6000run-20260829-141918-4769b3
116,2 tok/s
2× NVIDIA RTX A6000run-20260827-104702-83ea6a
116,0 tok/s
NVIDIA RTX A6000run-20260826-090149-1307ca
102,5 tok/s
NVIDIA RTX A6000run-20260826-190954-366c87
102,4 tok/s
3× AMD Radeon AI PRO R9700run-20260730-130613-00135d
87,9 tok/s
Hardware
GPU: 2x NVIDIA GeForce RTX 2060 · 6 GB VRAM
CPU: 4x Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz
RAM: 504 GB
Mainboard: Dell Inc. PowerEdge R820
Setup
Runtime: llama.cpp
Quantization: Q4_K_M
Model: Meta-Llama-3.1-8B-Instruct
Configuration
# LLM-Benchmark Konfiguration
# Modell : Meta-Llama-3.1-8B-Instruct
# Engine : llama.cpp
# Run-ID : run-20260723-194429-b2114c
# GPU : 2x NVIDIA GeForce RTX 2060
# CPU : 4x Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz
# RAM : 504 GB
bench@llm-benchmark:~$ /opt/llama.cpp/build/bin/llama-server \
-m /opt/models/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf \
-a Meta-Llama-3.1-8B-Instruct \
--host 0.0.0.0 \
--port 8080 \
--numa distribute \
-t 64 \
-tb 64 \
-c 8192 \
--parallel 4
| 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. | Meta-Llama-3.1-8B-Instruct |
| Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann. | 8192 |
| Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird. | /opt/models/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf |
| Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen. | Meta-Llama-3.1-8B-Instruct |
| NUMA?NUMA-Optimierung fuer Multi-Socket-CPUs: distribute/isolate/numactl. Verbessert die Speicherlokalitaet. | distribute |
| Threads?Anzahl CPU-Threads fuer die Token-Generierung (Decode). | 64 |
| Batch-Threads?Anzahl CPU-Threads fuer Prompt-Verarbeitung und Batch (Prefill). | 64 |
| Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells. | 8192 |
| Parallel?Anzahl paralleler Slots/Sequenzen, die der Server gleichzeitig bedient. Der Kontext wird auf die Slots aufgeteilt. | 4 |
Anzeige
Model comparison
Meta-Llama-3.1-8B-Instruct 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
NVIDIA GeForce RTX 5090 1.593,1 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 1.533,0 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1.433,0 tok/sNVIDIA RTX A6000 1.026,4 tok/sNVIDIA GeForce RTX 5070 Ti 913,4 tok/sNVIDIA GeForce RTX 3090 Ti 837,9 tok/sAMD Radeon AI PRO R9700 518,3 tok/sAMD Radeon PRO W7800 48GB 323,1 tok/s★ NVIDIA GeForce RTX 2060 130,9 tok/s this runNVIDIA Tesla P100 PCIe 16GB 67,6 tok/sIntel Arc Pro B70 66,4 tok/s
CPUby processor
AMD Ryzen 7 5800X3D 8-Core Processor 1.593,1 tok/sAMD Ryzen 9 9950X 16-Core Processor 1.533,0 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 1.433,0 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 1.026,4 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 913,4 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 837,9 tok/sIntel(R) Core(TM) i5-7400 CPU @ 3.00GHz 130,9 tok/sAMD Ryzen 9 7945HX with Radeon Graphics 67,6 tok/s★ Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 3,6 tok/s this run
MBby mainboard
ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 1.593,1 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.533,0 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 1.433,0 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 913,4 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 837,9 tok/sASRock H110 Pro BTC+ 130,9 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) 67,6 tok/sShenzhen Meigao Electronic Equipment Co.,Ltd DRFXI (MotherBoard Series) 66,4 tok/s★ Dell Inc. PowerEdge R820 3,6 tok/s this run
DRVby driver
unbekannt 1.593,1 tok/sIntel 26.18.38308.4 66,4 tok/s
💰 Economics
Economics of this run
Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (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
55 W
⚡ TDP
359 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)359 W estimated (TDP)GPU 320 + CPU 29 + Board 10 W full load
Avg cost / hourEUR 0.11
Electricity / 1M tokensEUR 8.42
Token / kWh35.62K
Acquisition (system)EUR 2,425 partial priceGPU EUR 220 · CPU EUR 39 · RAM EUR 2,016 · PSU EUR 150
Electricity (2 years)–
TCO (2 years)EUR 4,311
Output tokens (2 years)223.91M
☁️ 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 (55 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
| Meta-Llama-3.1-8B-Instruct2x NVIDIA GeForce RTX 2060 | Meta-Llama-3.1-8B-InstructNVIDIA GeForce RTX 5090 | Meta-Llama-3.1-8B-InstructNVIDIA RTX PRO 6000 Blackwell Workstation Edition | Meta-Llama-3.1-8B-Instruct3x 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) | – | – | – | – |
2× NVIDIA GeForce RTX 2060 · this run 4 tok/sNVIDIA GeForce RTX 5090 246 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 239 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 225 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.