Created bymario-alka.dePowered bygodcore.denoob2claw.detricoma.de
DEEN
Sign in Sign up
Beta
Contributed byMario AlkaGoogle

gemma-4-31B-it

Performance benchmark · measured on 27.07.2026 11:38

Benchmark-IDrun-20260727-121924-2535cc
Timebench 3 - Kombi (Prefill + Generation)Dense31BRuntime: llama.cppQuantisierung: INT8
Generation41,50tok/s
Prefill367,71tok/s
Time to First Token143.109,00ms
Total duration600,00s
Concurrency10parallel
Ranking in the field
This GPUAll GPUs
13of 20 systems

Performance benchmark · Primary metric: Generation-Speed (tok/s) · 10× concurrent

This run is better than 37 % of all comparable systems.
Generation 41,5 tok/s
-91 % vs Ø 464,5
Prefill 367,7 tok/s
-93 % vs Ø 5.437,0
Time to First Token 143.109 ms
+234 % vs Ø 42.890
Distribution in the field3 – 1.743 tok/s
Ø 465 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

Hardware

GPU: NVIDIA GeForce RTX 5070 Ti · 16 GB VRAM
CPU: AMD Ryzen Threadripper PRO 5975WX 32-Cores
RAM: 247 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI

Setup

Runtime: llama.cpp
Quantization: INT8
Model: gemma-4-31B-it

Configuration

benchmark-konfiguration — run-20260727-121924-2535cc
# LLM-Benchmark Konfiguration # Modell : gemma-4-31B-it # Engine : llama.cpp # Run-ID : run-20260727-121924-2535cc # GPU : NVIDIA GeForce RTX 5070 Ti # CPU : AMD Ryzen Threadripper PRO 5975WX 32-Cores # RAM : 247 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build/bin/llama-server \ -m /home/godcore/.cache/huggingface/hub/models--ReadyArt--gemma-4-31B-it-scotoma-GGUF/snapshots/273b4b513d43680abdad00a219bc938032547f12/gemma-4-31B-scotoma-Q4_K_M.gguf \ --alias gemma-4-31b-it-int8 \ --host 0.0.0.0 \ --port 8000 \ -ngl 30 \ -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-4-31b-it-int8
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--ReadyArt--gemma-4-31B-it-scotoma-GGUF/snapshots/273b4b513d43680abdad00a219bc938032547f12/gemma-4-31B-scotoma-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.30
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-4-31B-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

46,244,142,039,937,8342357371386Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 5070 Ti - 42,0 tok/s Generation, 364 tok/s Prefill, TTFT 71.670 ms (6 Laufe) | DIESER LAUF★ NVIDIA GeForce RTX 50...
★ NVIDIA GeForce RTX 5070 Ti 42,0 tok/s this run

CPUby processor

46,244,142,039,937,8342357371386Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 5975WX 32-Cores - 42,0 tok/s Generation, 364 tok/s Prefill, TTFT 71.670 ms (6 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
★ AMD Ryzen Threadripper PRO 5975WX 32-Cores 42,0 tok/s this run

MBby mainboard

46,244,142,039,937,8342357371386Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 42,0 tok/s Generation, 364 tok/s Prefill, TTFT 71.670 ms (6 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
★ ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 42,0 tok/s this run

ENGby engine

46,244,142,039,937,8342357371386Prefill (tok/s)Generation (tok/s)llama.cpp - 42,0 tok/s Generation, 364 tok/s Prefill, TTFT 71.670 ms (6 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 42,0 tok/s this run

DRVby driver

46,244,142,039,937,8342357371386Prefill (tok/s)Generation (tok/s)unbekannt - 42,0 tok/s Generation, 364 tok/s Prefill, TTFT 71.670 ms (6 Laufe)unbekannt
unbekannt 42,0 tok/s
💰 Economics

Economics of this run

Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (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 10 W
⚡ TDP 10 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)10 W missingBoard 10 W full load
Avg cost / hourEUR 0.0030
Electricity / 1M tokensEUR 0.020
Token / kWh14.94M
Acquisition (system)EUR 2,096 missingRAM EUR 1,976 · PSU EUR 120
Electricity (2 years)–
TCO (2 years)EUR 2,149
Output tokens (2 years)2.62B
☁️ External LLM (API) – comparison
External LLM cost (2 years)–
Savings vs. external (2 years)–
No power draw measured – values estimated from GPU TDP + CPU (idle + 15 %) + board.

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 (10 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

NVIDIA GeForce RTX 5070 Ti · this run 42 tok/sNVIDIA GeForce RTX 5070 Ti 42 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.