Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
Contributed byMario AlkaGoogle

gemma-4-E4B-it

Performance benchmark · measured on 27.07.2026 18:06

Benchmark-IDrun-20260727-181607-bf4219
Timebench 3 - Kombi (Prefill + Generation)Dense8BRuntime: llama.cppQuantisierung: Q4_K_M
Generation914,25tok/s
Prefill12.654,62tok/s
Time to First Token1.885,00ms
Total duration22,06s
Concurrency5parallel
Ranking in the field
6of 155 systems

Performance benchmark · Primary metric: Generation-Speed (tok/s) · 5× concurrent

This run is better than 97 % of all comparable systems.
Generation 914,3 tok/s
+279 % vs Ø 241,1
Prefill 12.654,6 tok/s
+137 % vs Ø 5.336,0
Time to First Token 1.885 ms
-93 % vs Ø 26.930
Distribution in the field0 – 1.349 tok/s
Ø 240 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

How does this benchmark compare on other GPUs?

Same model on different hardware · 5× concurrent · Generation (tok/s)

Hardware

GPU: NVIDIA RTX PRO 6000 Blackwell Workstation Edition · 96 GB VRAM
CPU: AMD Ryzen 9 9950X 16-Core Processor
RAM: 92 GB
Mainboard: ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI

Setup

Runtime: llama.cpp
Quantization: Q4_K_M
Model: gemma-4-E4B-it

Configuration

benchmark-konfiguration — run-20260727-181607-bf4219
# LLM-Benchmark Konfiguration # Modell : gemma-4-E4B-it # Engine : llama.cpp # Run-ID : run-20260727-181607-bf4219 # GPU : NVIDIA RTX PRO 6000 Blackwell Workstation Edition # CPU : AMD Ryzen 9 9950X 16-Core Processor # RAM : 92 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build/bin/llama-server \ -m /home/godcore/.cache/huggingface/hub/models--unsloth--gemma-4-E4B-it-GGUF/snapshots/bfc15c382204943c3a8fff0c750b94ae2364d7a3/gemma-4-E4B-it-Q4_K_M.gguf \ --alias gemma-4-E4B-it \ --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.gemma-4-E4B-it
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./home/godcore/.cache/huggingface/hub/models--unsloth--gemma-4-E4B-it-GGUF/snapshots/bfc15c382204943c3a8fff0c750b94ae2364d7a3/gemma-4-E4B-it-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

All benchmarks of this model To leaderboard

Model comparison

gemma-4-E4B-it 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

2.2591.6941.1295650,004.8749.74814.622Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 5070 Ti - 1.082,4 tok/s Generation, 9.183 tok/s Prefill, TTFT 2.840 ms (6 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 3090 Ti - 983,7 tok/s Generation, 6.659 tok/s Prefill, TTFT 4.857 ms (3 Laufe)NVIDIA GeForce RTX 30...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 137,2 tok/s Generation, 2.017 tok/s Prefill, TTFT 31.339 ms (3 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 1.758,6 tok/s Generation, 12.085 tok/s Prefill, TTFT 2.684 ms (3 Laufe) | DIESER LAUF★ NVIDIA RTX PRO 6000 B...
★ NVIDIA RTX PRO 6000 Blackwell Workstation Edition 1.758,6 tok/s this runNVIDIA GeForce RTX 5070 Ti 1.082,4 tok/sNVIDIA GeForce RTX 3090 Ti 983,7 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 137,2 tok/s

CPUby processor

2.2591.6941.1295650,004.8749.74814.622Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 5975WX 32-Cores - 1.082,4 tok/s Generation, 9.183 tok/s Prefill, TTFT 2.840 ms (6 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 983,7 tok/s Generation, 6.659 tok/s Prefill, TTFT 4.857 ms (3 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 137,2 tok/s Generation, 2.017 tok/s Prefill, TTFT 31.339 ms (3 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 9950X 16-Core Processor - 1.758,6 tok/s Generation, 12.085 tok/s Prefill, TTFT 2.684 ms (3 Laufe) | DIESER LAUF★ AMD Ryzen 9 9950X 16-...
★ AMD Ryzen 9 9950X 16-Core Processor 1.758,6 tok/s this runAMD Ryzen Threadripper PRO 5975WX 32-Cores 1.082,4 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 983,7 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 137,2 tok/s

MBby mainboard

2.2591.6941.1295650,004.8749.74814.622Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 1.082,4 tok/s Generation, 9.183 tok/s Prefill, TTFT 2.840 ms (6 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 983,7 tok/s Generation, 6.659 tok/s Prefill, TTFT 4.857 ms (3 Laufe)Meigao Innovation Tec...ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 137,2 tok/s Generation, 2.017 tok/s Prefill, TTFT 31.339 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 1.758,6 tok/s Generation, 12.085 tok/s Prefill, TTFT 2.684 ms (3 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
★ ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.758,6 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 1.082,4 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 983,7 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 137,2 tok/s

ENGby engine

2.0821.7361.3911.0457004.4818.05411.62815.201Prefill (tok/s)Generation (tok/s)vLLM - 1.022,8 tok/s Generation, 13.200 tok/s Prefill, TTFT 1.096 ms (3 Laufe)vLLMllama.cpp - 1.758,6 tok/s Generation, 6.482 tok/s Prefill, TTFT 10.866 ms (12 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 1.758,6 tok/s this runvLLM 1.022,8 tok/s

DRVby driver

1.9341.8461.7591.6711.5837.3567.6697.9828.295Prefill (tok/s)Generation (tok/s)unbekannt - 1.758,6 tok/s Generation, 7.825 tok/s Prefill, TTFT 8.912 ms (15 Laufe)unbekannt
unbekannt 1.758,6 tok/s
💰 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 →

⚙️ 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 70 W
⚡ TDP 644 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)644 W estimated (TDP)GPU 600 + CPU 29 + Board 15 W full load
Avg cost / hourEUR 0.19
Electricity / 1M tokensEUR 0.059
Token / kWh5.11M
Acquisition (system)EUR 15,248 full priceGPU EUR 13,000 · CPU EUR 649 · Board EUR 499 · RAM EUR 920 · PSU EUR 180
Electricity (2 years)
TCO (2 years)EUR 18,632
Output tokens (2 years)57.66B
☁️ 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 (70 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

gemma-4-E4B-itNVIDIA RTX PRO 6000 Blackwell Workstation Editiongemma-4-E4B-itNVIDIA GeForce RTX 5070 Tigemma-4-E4B-itNVIDIA GeForce RTX 5070 Tigemma-4-E4B-itNVIDIA 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.