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

Seed-OSS-36B-Instruct

Performance benchmark · measured on 23.07.2026 02:20

Benchmark-IDrun-20260723-023424-47e9f4
Timebench 3 - Kombi (Prefill + Generation)36BRuntime: llama.cppQuantisierung: Q4_K_M
Generation0,97tok/s
Prefill7,84tok/s
Time to First Token240.790,00ms
Total duration600,00s
Concurrency1parallel
Ranking in the field
32of 33 systems

Performance benchmark · Primary metric: Generation-Speed (tok/s) · 1× concurrent

This run is better than 3 % of all comparable systems.
Generation 1,0 tok/s
-96 % vs Ø 21,9
Prefill 7,8 tok/s
-99 % vs Ø 605,3
Time to First Token 240.790 ms
+439 % vs Ø 44.665
Distribution in the field1 – 81 tok/s
Ø 22 Median Dieser Lauf

Wie schlägt sich dieser Benchmark mit anderen Modellen?

How does this benchmark compare on other GPUs?

Same model on different hardware · 1× concurrent · Generation (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: Seed-OSS-36B-Instruct

Configuration

benchmark-konfiguration — run-20260723-023424-47e9f4
# LLM-Benchmark Konfiguration # Modell : Seed-OSS-36B-Instruct # Engine : llama.cpp # Run-ID : run-20260723-023424-47e9f4 # 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/Seed-OSS-36B-Instruct-Q4_K_M.gguf \ -a Seed-OSS-36B-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.Seed-OSS-36B-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/Seed-OSS-36B-Instruct-Q4_K_M.gguf
Alias?Anzeigename des Modells nach aussen (served model name), unabhaengig vom Dateinamen.Seed-OSS-36B-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

All benchmarks of this model To leaderboard

Anzeige
Model comparison

Seed-OSS-36B-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

1.0918195462730,001.3192.6393.958Prefill (tok/s)Generation (tok/s)NVIDIA GeForce RTX 5090 - 839,7 tok/s Generation, 2.664 tok/s Prefill, TTFT 12.384 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 460,0 tok/s Generation, 3.141 tok/s Prefill, TTFT 10.624 ms (3 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 383,9 tok/s Generation, 3.193 tok/s Prefill, TTFT 13.087 ms (9 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA RTX A6000 - 349,2 tok/s Generation, 2.724 tok/s Prefill, TTFT 10.630 ms (30 Laufe)NVIDIA RTX A6000NVIDIA GeForce RTX 3090 Ti - 111,6 tok/s Generation, 992 tok/s Prefill, TTFT 41.392 ms (3 Laufe)NVIDIA GeForce RTX 30...AMD Radeon AI PRO R9700 - 107,3 tok/s Generation, 693 tok/s Prefill, TTFT 114.241 ms (16 Laufe)AMD Radeon AI PRO R97...AMD Radeon PRO W7900 Dual Slot - 88,6 tok/s Generation, 680 tok/s Prefill, TTFT 17.557 ms (3 Laufe)AMD Radeon PRO W7900 ...AMD Radeon PRO W7800 48GB - 57,7 tok/s Generation, 1.357 tok/s Prefill, TTFT 10.389 ms (3 Laufe)AMD Radeon PRO W7800 ...NVIDIA GeForce RTX 5070 Ti - 39,2 tok/s Generation, 423 tok/s Prefill, TTFT 95.542 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 2060 - 1,0 tok/s Generation, 8 tok/s Prefill, TTFT 240.790 ms (1 Lauf) | DIESER LAUF★ NVIDIA GeForce RTX 20...
NVIDIA GeForce RTX 5090 839,7 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 460,0 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 383,9 tok/sNVIDIA RTX A6000 349,2 tok/sNVIDIA GeForce RTX 3090 Ti 111,6 tok/sAMD Radeon AI PRO R9700 107,3 tok/sAMD Radeon PRO W7900 Dual Slot 88,6 tok/sAMD Radeon PRO W7800 48GB 57,7 tok/sNVIDIA GeForce RTX 5070 Ti 39,2 tok/s★ NVIDIA GeForce RTX 2060 1,0 tok/s this run

CPUby processor

1.0918195462730,001.3192.6393.958Prefill (tok/s)Generation (tok/s)AMD Ryzen 7 5800X3D 8-Core Processor - 839,7 tok/s Generation, 2.664 tok/s Prefill, TTFT 12.384 ms (3 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen 9 9950X 16-Core Processor - 460,0 tok/s Generation, 3.141 tok/s Prefill, TTFT 10.624 ms (3 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 383,9 tok/s Generation, 3.193 tok/s Prefill, TTFT 13.087 ms (9 Laufe)AMD Ryzen Threadrippe...AMD Ryzen Threadripper PRO 7955WX 16-Cores - 349,2 tok/s Generation, 2.017 tok/s Prefill, TTFT 46.669 ms (46 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 111,6 tok/s Generation, 992 tok/s Prefill, TTFT 41.392 ms (3 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 88,6 tok/s Generation, 820 tok/s Prefill, TTFT 41.163 ms (9 Laufe)AMD Ryzen Threadrippe...Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz - 1,0 tok/s Generation, 8 tok/s Prefill, TTFT 240.790 ms (1 Lauf) | DIESER LAUF★ Intel(R) Xeon(R) CPU ...
AMD Ryzen 7 5800X3D 8-Core Processor 839,7 tok/sAMD Ryzen 9 9950X 16-Core Processor 460,0 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 383,9 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 349,2 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 111,6 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 88,6 tok/s★ Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 1,0 tok/s this run

MBby mainboard

1.0918195462730,001.2982.5953.893Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 839,7 tok/s Generation, 2.664 tok/s Prefill, TTFT 12.384 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 460,0 tok/s Generation, 3.141 tok/s Prefill, TTFT 10.624 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 383,9 tok/s Generation, 2.210 tok/s Prefill, TTFT 41.174 ms (55 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 111,6 tok/s Generation, 992 tok/s Prefill, TTFT 41.392 ms (3 Laufe)Meigao Innovation Tec...ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 88,6 tok/s Generation, 820 tok/s Prefill, TTFT 41.163 ms (9 Laufe)ASUSTeK COMPUTER INC....Dell Inc. PowerEdge R820 - 1,0 tok/s Generation, 8 tok/s Prefill, TTFT 240.790 ms (1 Lauf) | DIESER LAUF★ Dell Inc. PowerEdge R...
ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 839,7 tok/sASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 460,0 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 383,9 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 111,6 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 88,6 tok/s★ Dell Inc. PowerEdge R820 1,0 tok/s this run

ENGby engine

1.0228085943811678322.0113.1904.369Prefill (tok/s)Generation (tok/s)vLLM - 349,2 tok/s Generation, 3.736 tok/s Prefill, TTFT 4.697 ms (18 Laufe)vLLMllama.cpp - 839,7 tok/s Generation, 1.465 tok/s Prefill, TTFT 53.294 ms (56 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 839,7 tok/s this runvLLM 349,2 tok/s

DRVby driver

9248828407987561.8971.9772.0582.139Prefill (tok/s)Generation (tok/s)unbekannt - 839,7 tok/s Generation, 2.018 tok/s Prefill, TTFT 41.473 ms (74 Laufe)unbekannt
unbekannt 839,7 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 30.82
Token / kWh9.73K
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)61.18M
☁️ 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

Seed-OSS-36B-Instruct2x NVIDIA GeForce RTX 2060Seed-OSS-36B-InstructNVIDIA RTX PRO 6000 Blackwell Workstation EditionSeed-OSS-36B-InstructNVIDIA GeForce RTX 5090Seed-OSS-36B-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)

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.