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

Llama-3_1-Nemotron-Ultra-253B-v1

Performance benchmark · measured on 31.07.2026 09:17

Benchmark-IDrun-20260801-072826-fc5022
Timebench 3 - Kombi (Prefill + Generation)Dense253BRuntime: llama.cppQuantisierung: Q4_K_M
Generation31,63tok/s
Prefill811,78tok/s
Time to First Token50.633,00ms
Total duration618,19s
Concurrency5parallel
Ranking in the field
185of 193 systems

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

This run is better than 4 % of all comparable systems.
Generation 31,6 tok/s
-93 % vs Ø 482,8
Prefill 811,8 tok/s
-88 % vs Ø 6.573,8
Time to First Token 50.633 ms
+322 % vs Ø 11.989
Distribution in the field3 – 1.196 tok/s
Ø 483 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

gemma-4-E2B-it3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-184456-4111a9
1.195,7 tok/s
NVIDIA-Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-836a91
1.090,6 tok/s
NVIDIA-Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260727-181606-8b381e
1.077,4 tok/s
NVIDIA-Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-b8a818
1.071,0 tok/s
NVIDIA-Nemotron-3-Nano-4B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-194136-038e92
1.051,1 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-032121-ffb18a
1.015,5 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260728-034052-437388
989,7 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-032120-8b6146
970,4 tok/s
gpt-oss-20b3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-032120-310458
946,5 tok/s
Nemotron-3-Nano-Omni-30B-A3B-Reasoning3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-105340-c65107
931,3 tok/s
NVIDIA-Nemotron-3-Nano-30B-A3B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-105341-60b672
930,9 tok/s
NVIDIA-Nemotron-3-Nano-30B-A3B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-105341-88bf57
929,1 tok/s
Nemotron-Cascade-2-30B-A3B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-105340-86d336
928,5 tok/s
Nemotron-Cascade-2-30B-A3B3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260729-105340-9ec82f
927,0 tok/s
Llama-3_1-Nemotron-Ultra-253B-v1 this run3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260801-072826-fc5022
31,6 tok/s

How does this benchmark compare on other GPUs?

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

Configuration

benchmark-konfiguration — run-20260801-072826-fc5022
# LLM-Benchmark Konfiguration # Modell : Llama-3_1-Nemotron-Ultra-253B-v1 # Engine : llama.cpp # Run-ID : run-20260801-072826-fc5022 # GPU : 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition # CPU : AMD Ryzen Threadripper PRO 9965WX 24-Cores # RAM : 125 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build/bin/llama-server \ -m /home/godcore/.cache/huggingface/hub/models--DevQuasar--nvidia.Llama-3_1-Nemotron-Ultra-253B-v1-GGUF/snapshots/fa034db97ae9c144e5d5a8fd8fc4fefdd363b8db/nvidia.Llama-3_1-Nemotron-Ultra-253B-v1.Q4_K_M-00001-of-00012.gguf \ --alias Llama-3_1-Nemotron-Ultra-253B-v1 \ --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.Llama-3_1-Nemotron-Ultra-253B-v1
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--DevQuasar--nvidia.Llama-3_1-Nemotron-Ultra-253B-v1-GGUF/snapshots/fa034db97ae9c144e5d5a8fd8fc4fefdd363b8db/nvidia.Llama-3_1-Nemotron-Ultra-253B-v1.Q4_K_M-00001-of-00012.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

Anzeige
Model comparison

Llama-3_1-Nemotron-Ultra-253B-v1 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

80,760,540,320,20,00229457686Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 5070 Ti - 1,7 tok/s Generation, 45 tok/s Prefill, TTFT 141.946 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 1,0 tok/s Generation, 30 tok/s Prefill, TTFT 185.560 ms (8 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5090 - 0,1 tok/s Generation, 8 tok/s Prefill, TTFT 285.808 ms (2 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 62,1 tok/s Generation, 554 tok/s Prefill, TTFT 82.459 ms (9 Laufe) | DIESER LAUF★ NVIDIA RTX PRO 6000 B...
★ NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 62,1 tok/s this runNVIDIA GeForce RTX 5070 Ti 1,7 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 1,0 tok/sNVIDIA GeForce RTX 5090 0,1 tok/s

CPUby processor

80,760,540,320,20,00229457686Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 5975WX 32-Cores - 1,7 tok/s Generation, 45 tok/s Prefill, TTFT 141.946 ms (3 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 9950X 16-Core Processor - 1,0 tok/s Generation, 30 tok/s Prefill, TTFT 185.560 ms (8 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen 7 5800X3D 8-Core Processor - 0,1 tok/s Generation, 8 tok/s Prefill, TTFT 285.808 ms (2 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 62,1 tok/s Generation, 554 tok/s Prefill, TTFT 82.459 ms (9 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
★ AMD Ryzen Threadripper PRO 9965WX 24-Cores 62,1 tok/s this runAMD Ryzen Threadripper PRO 5975WX 32-Cores 1,7 tok/sAMD Ryzen 9 9950X 16-Core Processor 1,0 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 0,1 tok/s

MBby mainboard

80,760,540,320,20,00229457686Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 1,7 tok/s Generation, 45 tok/s Prefill, TTFT 141.946 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 1,0 tok/s Generation, 30 tok/s Prefill, TTFT 185.560 ms (8 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 0,1 tok/s Generation, 8 tok/s Prefill, TTFT 285.808 ms (2 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 62,1 tok/s Generation, 554 tok/s Prefill, TTFT 82.459 ms (9 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
★ ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 62,1 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 1,7 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1,0 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 0,1 tok/s

ENGby engine

68,365,262,159,055,9230240249259Prefill (tok/s)Generation (tok/s)llama.cpp - 62,1 tok/s Generation, 245 tok/s Prefill, TTFT 146.548 ms (22 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 62,1 tok/s this run

DRVby driver

68,365,262,159,055,9230240249259Prefill (tok/s)Generation (tok/s)unbekannt - 62,1 tok/s Generation, 245 tok/s Prefill, TTFT 146.548 ms (22 Laufe)unbekannt
unbekannt 62,1 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 165 W
⚡ TDP 983 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)983 W estimated (TDP)GPU 900 + CPU 57 + Board 25 W full load
Avg cost / hourEUR 0.29
Electricity / 1M tokensEUR 2.59
Token / kWh115.90K
Acquisition (system)EUR 45,748 full priceGPU EUR 39,000 · CPU EUR 3,499 · Board EUR 1,299 · RAM EUR 1,750 · PSU EUR 200
Electricity (2 years)
TCO (2 years)EUR 50,912
Output tokens (2 years)1.99B
☁️ 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 (165 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

Llama-3_1-Nemotron-Ultra-253B-v13x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionLlama-3_1-Nemotron-Ultra-253B-v13x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionLlama-3_1-Nemotron-Ultra-253B-v13x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionLlama-3_1-Nemotron-Ultra-253B-v1NVIDIA GeForce RTX 5070 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.