Contributed byMario AlkaMiniMax MiniMax-M2.5
Performance benchmark · measured on 26.07.2026 09:09
Benchmark-ID
run-20260726-091400-7d7441Timebench 3 - Kombi (Prefill + Generation)MoE230BRuntime: vLLMQuantisierung: AWQ
⚠
Configuration note: This run used 2 devices in distributed inference (2× NVIDIA DGX SPARK (GB10), tensor/pipeline parallel via the vLLM config). Not a single machine – power & acquisition in the economics section are calculated for all 2 devices.
ℹ
For context: Diese Plattform nutzt Unified Memory – die "VRAM" ist gemeinsamer System-RAM (APU/Superchip); das Modell teilt sich den Speicher mit dem System.
Generation77,74tok/s
Prefill5.459,74tok/s
Time to First Token2.641,00ms
Total duration137,32s
Concurrency5parallel
Ranking in the field
1082of 1512 systems
Performance benchmark · Primary metric: Generation-Speed (tok/s) · 5× concurrent
This run is better than 28 % of all comparable systems.
Generation
77,7 tok/s
-70 % vs Ø 256,0
Prefill
5.459,7 tok/s
+1 % vs Ø 5.429,9
Time to First Token
2.641 ms
-91 % vs Ø 29.391
Distribution in the field0 – 1.349 tok/s
Wie schlägt sich dieser Benchmark mit anderen Modellen?
gemma-4-E2B-itNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260727-181606-0adf6d
1.349,3 tok/s
gemma-4-E2B-itNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-071844-e5e309
1.348,4 tok/s
gemma-4-E2B-itNVIDIA GeForce RTX 5090 · run-20260728-184455-5b4937
1.301,9 tok/s
gemma-4-E2B-it3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-184456-4111a9
1.195,7 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260727-181607-30d864
1.188,7 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA GeForce RTX 5090 · run-20260728-194135-5432cd
1.164,8 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-071851-c3c4a5
1.154,4 tok/s
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260728-140954-9e81f9
1.136,3 tok/s
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20260730-035052-09627f
1.135,0 tok/s
NVIDIA-Nemotron-3.5-Lightning-30B-A3BNVIDIA GeForce RTX 3090 Ti · run-20261021-093815-b22926
1.135,0 tok/s
gpt-oss-20bNVIDIA GeForce RTX 5090 · run-20260729-032121-058a31
1.113,5 tok/s
gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Edition · run-20261023-071836-99dfa5
1.105,8 tok/s
NVIDIA-Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-836a91
1.090,6 tok/s
NVIDIA-Nemotron-3-Nano-4B3× 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260727-181606-8b381e
1.077,4 tok/s
MiniMax-M2.5 this runNVIDIA DGX SPARK (GB10) · run-20260726-091400-7d7441
77,7 tok/s
How does this benchmark compare on other GPUs?
Same model on different hardware · 5× concurrent · Generation (tok/s)
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260728-164528-587188
316,9 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032130-ae100e
315,2 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032127-31e6f7
315,0 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260728-164528-0af3f9
313,1 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032129-f8f834
312,8 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032127-c585fb
310,9 tok/s
NVIDIA DGX SPARK (GB10) this runrun-20260726-091400-7d7441
77,7 tok/s
4× AMD Radeon AI PRO R9700run-20260810-033720-3b8e41
56,1 tok/s
NVIDIA GeForce RTX 5070 Tirun-20260728-164528-ab238d
22,5 tok/s
NVIDIA GeForce RTX 5070 Tirun-20260728-164527-ad6faf
21,9 tok/s
3× AMD Radeon AI PRO R9700run-20260806-094437-c0a675
21,2 tok/s
AMD Radeon PRO W7900 Dual Slotrun-20260817-230037-7127ca
19,3 tok/s
3× AMD Radeon AI PRO R9700run-20260806-094437-ba596e
19,1 tok/s
2× AMD Radeon PRO W7900 Dual Slotrun-20260818-094144-b084b5
19,0 tok/s
Hardware
GPU: NVIDIA DGX SPARK (GB10) · 128 GB VRAM
CPU: NVIDIA Grace
RAM: 120 GB
Mainboard: ASUSTeK COMPUTER INC. GX10
Setup
Runtime: vLLM
Quantization: AWQ
Driver: NVIDIA 590.48.01 / CUDA 13.1
Model: MiniMax-M2.5
Configuration
# LLM-Benchmark Konfiguration
# Modell : MiniMax-M2.5
# Engine : vLLM
# Run-ID : run-20260726-091400-7d7441
# GPU : NVIDIA DGX SPARK (GB10)
# CPU : NVIDIA Grace
# RAM : 120 GB
bench@llm-benchmark:~$ bash \
-lc HF_HUB_OFFLINE=1 VLLM_HOST_IP=10.200.0.1 vllm serve cyankiwi/MiniMax-M2.5-AWQ-4bit \
--served-model-name MiniMax-M2.5 \
--host 0.0.0.0 \
--port 8000 \
--tensor-parallel-size 2 \
--distributed-executor-backend ray \
--gpu-memory-utilization 0.85 \
--max-model-len 8192 \
--trust-remote-code \
--enforce-eager '>/tmp/dgxboth_srv.log' '2>&1'
| 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 |
| Modellalias?Der Name, unter dem das Modell ueber die API angesprochen wird. Genau dieser Wert muss im Request-Feld 'model' stehen. | MiniMax-M2.5 |
| Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann. | 8192 |
| lc | HF_HUB_OFFLINE=1 |
| Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen. | MiniMax-M2.5 |
| 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 |
| Executor-Backend?Verfahren fuer Multi-GPU/Multi-Node: mp = Prozesse auf einer Maschine, ray = ueber Ray (auch ueber mehrere Knoten). | ray |
| GPU-Speicher?Anteil des GPU-Speichers (0 bis 1), den vLLM belegen darf. 0.92 = 92 %. Hoeher = mehr Platz fuer den KV-Cache (mehr/laengere parallele Anfragen), aber groesseres Risiko fuer 'Out of Memory'. | 0.85 |
| Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells. | 8192 |
| Enforce-Eager?Schaltet die optimierte Graph-Ausfuehrung (CUDA-/HIP-Graphs) AB und rechnet Schritt fuer Schritt. Startet schneller und spart etwas VRAM, ist im laufenden Betrieb aber meist langsamer als mit Graphs. | >/tmp/dgxboth_srv.log |
| execution_type | local |
Anzeige
Model comparison
MiniMax-M2.5 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 RTX PRO 6000 Blackwell Max-Q Workstation Edition 551,0 tok/s★ NVIDIA DGX SPARK (GB10) 103,5 tok/s this runAMD Radeon AI PRO R9700 66,1 tok/sNVIDIA GeForce RTX 5070 Ti 22,5 tok/sAMD Radeon PRO W7900 Dual Slot 19,3 tok/sNVIDIA GeForce RTX 3090 Ti 15,9 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 7,0 tok/sNVIDIA GeForce RTX 5090 4,4 tok/sAMD Radeon RX 7900 XTX 4,0 tok/sCPU-only 0,8 tok/s
CPUby processor
AMD Ryzen Threadripper PRO 9965WX 24-Cores 551,0 tok/s★ NVIDIA Grace 103,5 tok/s this runAMD Ryzen Threadripper PRO 7955WX 16-Cores 66,1 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 22,5 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 15,9 tok/sAMD Ryzen 9 9950X 16-Core Processor 7,0 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 4,4 tok/sIntel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz 4,0 tok/sIntel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 0,8 tok/s
MBby mainboard
ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 551,0 tok/s★ ASUSTeK COMPUTER INC. GX10 103,5 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 22,5 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 15,9 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 7,0 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 4,4 tok/sDell Inc. PowerEdge R820 4,0 tok/s
DRVby driver
unbekannt 551,0 tok/s★ NVIDIA 590.48.01 / CUDA 13.1 103,5 tok/s this run
💰 Economics
Economics of this run
Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (5× concurrent). Methodology →
🔌
2 devices in distributed operation – the calculation below (power, acquisition, TCO) uses 2× NVIDIA DGX SPARK (GB10) (detected from tensor/pipeline parallel in the vLLM configuration).
⚙️ 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
30 W
⚡ TDP
280 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)280 W estimated (TDP)GPU 280 W full load
Avg cost / hourEUR 0.084
Electricity / 1M tokensEUR 0.30
Token / kWh999.51K
Acquisition (system)EUR 11,600 partial priceGPU EUR 8,000 · RAM EUR 3,360 · PSU EUR 240
Electricity (2 years)–
TCO (2 years)EUR 13,072
Output tokens (2 years)4.90B
☁️ 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 (30 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
| MiniMax-M2.5NVIDIA DGX SPARK (GB10) | MiniMax-M2.53x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | MiniMax-M2.53x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | MiniMax-M2.53x 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) | – | – | – | – |
NVIDIA DGX SPARK (GB10) · this run 78 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 317 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 315 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 315 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.