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

gpt-oss-20b

Performance benchmark · measured on 24.10.2026 16:14

Benchmark-IDrun-20261024-143953-fe7e7e
Timebench 3 - Kombi (Prefill + Generation)MoE20BRuntime: llama.cppQuantisierung: MXFP4
Generation119,73tok/s
Prefill3.481,82tok/s
Time to First Token4.697,70ms
Total duration152,40s
Concurrency5parallel
Ranking in the field
3of 5 systems

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

This run is better than 50 % of all comparable systems.
Generation 119,7 tok/s
+16 % vs Ø 103,5
Prefill 3.481,8 tok/s
+40 % vs Ø 2.481,7
Time to First Token 4.698 ms
-41 % vs Ø 7.933
Distribution in the field72 – 125 tok/s
Ø 103 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 SPARK (N1X)
CPU: NVIDIA RTX Spark N1X (6144-core GPU, 20-core CPU)
RAM: 128 GB
Mainboard: ASUS H7607BA

Setup

Runtime: llama.cpp
Quantization: MXFP4
Model: gpt-oss-20b

Configuration

benchmark-konfiguration — run-20261024-143953-fe7e7e
# LLM-Benchmark Konfiguration # Modell : gpt-oss-20b # Engine : llama.cpp # Run-ID : run-20261024-143953-fe7e7e # GPU : NVIDIA RTX SPARK (N1X) # CPU : NVIDIA RTX Spark N1X (6144-core GPU, 20-core CPU) # RAM : 128 GB bench@llm-benchmark:~$ 'C:\Users\mail\.lmstudio\extensions\backends\llama.cpp-win-arm64-nvidia-cuda13-2.61.0\llama-server.exe' \ --model 'C:\Users\mail\Documents\Codex\2026-10-24\mach\work\models\gpt-oss-20b-MXFP4.gguf' \ --alias benchmark-local \ --host 127.0.0.1 \ --port 18765 \ --device CUDA0 \ --gpu-layers all \ --parallel 10 \ --kv-unified \ --kv-unified-per-slot 4096 \ --cache-type-k q8_0 \ --cache-type-v q8_0 \ --threads 20 \ --threads-batch 20 \ --batch-size 1024 \ --ubatch-size 256 \ --flash-attn on \ --fit off \ --no-cache-prompt \ --no-cache-idle-slots \ --cache-ram 0 \ --spec-type none \ --metrics \ --slots
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.gpt-oss-20b
Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann.4096
Batch-Groesse?Logische Batchgroesse fuer die Prompt-Verarbeitung (Anzahl Token pro Prefill-Durchlauf).1024
cache-ram0
K-Cache-Typ?Datentyp des K-Anteils im KV-Cache (f16/q8_0/q4_0 ...). Quantisiert senkt den VRAM-Bedarf bei leichtem Qualitaetsverlust.q8_0
V-Cache-Typ?Datentyp des V-Anteils im KV-Cache (f16/q8_0/q4_0 ...). Quantisierung braucht meist aktiviertes flash-attn.q8_0
deviceCUDA0
fitoff
Flash Attention?FlashAttention fuer schnellere und speichersparende Attention. Wert on/off/auto je nach Build.on
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.all
kv-unifiedaktiv
kv-unified-per-slot4096
metricsaktiv
Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird.C:\Users\mail\Documents\Codex\2026-10-24\mach\work\models\gpt-oss-20b-MXFP4.gguf
no-cache-idle-slotsaktiv
no-cache-promptaktiv
Parallel?Anzahl paralleler Slots/Sequenzen, die der Server gleichzeitig bedient. Der Kontext wird auf die Slots aufgeteilt.10
slotsaktiv
spec-typenone
Threads?Anzahl CPU-Threads fuer die Token-Generierung (Decode).20
Batch-Threads?Anzahl CPU-Threads fuer Prompt-Verarbeitung und Batch (Prefill).20
Micro-Batch?Physische Micro-Batchgroesse. Begrenzt, wie viele Token gleichzeitig durch die Compute-Pipeline laufen.256
execution_typelocal

All benchmarks of this model To leaderboard

Anzeige
Model comparison

gpt-oss-20b 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

2.5801.9351.2906450,008.34216.68525.027Prefill (tok/s)Generation (tok/s)NVIDIA RTX PRO 6000 Blackwell Workstation Edition - 1.995,4 tok/s Generation, 20.200 tok/s Prefill, TTFT 2.065 ms (11 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5090 - 1.890,7 tok/s Generation, 11.034 tok/s Prefill, TTFT 4.372 ms (18 Laufe)NVIDIA GeForce RTX 50...NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 1.835,7 tok/s Generation, 13.385 tok/s Prefill, TTFT 2.313 ms (15 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5070 Ti - 1.210,3 tok/s Generation, 4.544 tok/s Prefill, TTFT 11.765 ms (6 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 3090 Ti - 1.113,5 tok/s Generation, 6.760 tok/s Prefill, TTFT 3.631 ms (22 Laufe)NVIDIA GeForce RTX 30...NVIDIA RTX A6000 - 703,8 tok/s Generation, 7.111 tok/s Prefill, TTFT 1.994 ms (21 Laufe)NVIDIA RTX A6000NVIDIA GeForce RTX 3090 - 448,2 tok/s Generation, 4.932 tok/s Prefill, TTFT 2.012 ms (3 Laufe)NVIDIA GeForce RTX 30...AMD Radeon RX 7900 XTX - 357,7 tok/s Generation, 4.266 tok/s Prefill, TTFT 2.323 ms (9 Laufe)AMD Radeon RX 7900 XTXAMD Radeon PRO W7900 Dual Slot - 344,1 tok/s Generation, 2.818 tok/s Prefill, TTFT 3.556 ms (9 Laufe)AMD Radeon PRO W7900 ...AMD Radeon AI PRO R9700 - 323,6 tok/s Generation, 2.908 tok/s Prefill, TTFT 24.282 ms (24 Laufe)AMD Radeon AI PRO R97...AMD Radeon PRO W7800 48GB - 280,0 tok/s Generation, 3.327 tok/s Prefill, TTFT 2.726 ms (6 Laufe)AMD Radeon PRO W7800 ...NVIDIA DGX SPARK (GB10) - 233,3 tok/s Generation, 3.963 tok/s Prefill, TTFT 2.535 ms (6 Laufe)NVIDIA DGX SPARK (GB1...NVIDIA Tesla V100 PCIe 32GB - 206,6 tok/s Generation, 2.660 tok/s Prefill, TTFT 4.402 ms (3 Laufe)NVIDIA Tesla V100 PCI...NVIDIA Tesla P100 PCIe 16GB - 111,1 tok/s Generation, 646 tok/s Prefill, TTFT 16.076 ms (3 Laufe)NVIDIA Tesla P100 PCI...AMD Radeon 8060S Graphics - 74,4 tok/s Generation, 1.168 tok/s Prefill, TTFT 12.711 ms (2 Laufe)AMD Radeon 8060S Grap...CPU-only - 70,9 tok/s Generation, 119 tok/s Prefill, TTFT 95.764 ms (6 Laufe)CPU-onlyNVIDIA RTX SPARK (N1X) - 200,5 tok/s Generation, 3.459 tok/s Prefill, TTFT 3.538 ms (4 Laufe) | DIESER LAUF★ NVIDIA RTX SPARK (N1X)
NVIDIA RTX PRO 6000 Blackwell Workstation Edition 1.995,4 tok/sNVIDIA GeForce RTX 5090 1.890,7 tok/sNVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1.835,7 tok/sNVIDIA GeForce RTX 5070 Ti 1.210,3 tok/sNVIDIA GeForce RTX 3090 Ti 1.113,5 tok/sNVIDIA RTX A6000 703,8 tok/sNVIDIA GeForce RTX 3090 448,2 tok/sAMD Radeon RX 7900 XTX 357,7 tok/sAMD Radeon PRO W7900 Dual Slot 344,1 tok/sAMD Radeon AI PRO R9700 323,6 tok/sAMD Radeon PRO W7800 48GB 280,0 tok/sNVIDIA DGX SPARK (GB10) 233,3 tok/sNVIDIA Tesla V100 PCIe 32GB 206,6 tok/s★ NVIDIA RTX SPARK (N1X) 200,5 tok/s this runNVIDIA Tesla P100 PCIe 16GB 111,1 tok/sAMD Radeon 8060S Graphics 74,4 tok/sCPU-only 70,9 tok/s

CPUby processor

2.5911.9431.2966480,008.07916.15724.236Prefill (tok/s)Generation (tok/s)AMD Ryzen 9 9950X 16-Core Processor - 1.995,4 tok/s Generation, 19.558 tok/s Prefill, TTFT 1.895 ms (8 Laufe)AMD Ryzen 9 9950X 16-...AMD Ryzen 5 5600X 6-Core Processor - 1.981,1 tok/s Generation, 11.581 tok/s Prefill, TTFT 1.800 ms (10 Laufe)AMD Ryzen 5 5600X 6-C...AMD Ryzen 7 5800X3D 8-Core Processor - 1.890,7 tok/s Generation, 11.034 tok/s Prefill, TTFT 4.372 ms (18 Laufe)AMD Ryzen 7 5800X3D 8...AMD Ryzen Threadripper PRO 9965WX 24-Cores - 1.835,7 tok/s Generation, 13.385 tok/s Prefill, TTFT 2.313 ms (15 Laufe)AMD Ryzen Threadrippe...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 1.210,3 tok/s Generation, 3.456 tok/s Prefill, TTFT 5.664 ms (21 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 9 8945HX with Radeon Graphics - 1.113,5 tok/s Generation, 6.303 tok/s Prefill, TTFT 4.194 ms (18 Laufe)AMD Ryzen 9 8945HX wi...AMD Ryzen Threadripper PRO 7955WX 16-Cores - 703,8 tok/s Generation, 4.576 tok/s Prefill, TTFT 18.346 ms (45 Laufe)AMD Ryzen Threadrippe...AMD Ryzen Threadripper PRO 3955WX 16-Cores - 357,7 tok/s Generation, 4.214 tok/s Prefill, TTFT 2.394 ms (6 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 3 3100 4-Core Processor - 323,6 tok/s Generation, 4.545 tok/s Prefill, TTFT 2.204 ms (3 Laufe)AMD Ryzen 3 3100 4-Co...Intel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz - 247,3 tok/s Generation, 4.370 tok/s Prefill, TTFT 2.182 ms (3 Laufe)Intel(R) Xeon(R) CPU ...NVIDIA Grace - 233,3 tok/s Generation, 3.963 tok/s Prefill, TTFT 2.535 ms (6 Laufe)NVIDIA GraceAMD Ryzen 3 PRO 3200GE w/ Radeon Vega Graphics - 206,6 tok/s Generation, 2.660 tok/s Prefill, TTFT 4.402 ms (3 Laufe)AMD Ryzen 3 PRO 3200G...AMD Ryzen 9 7945HX with Radeon Graphics - 111,1 tok/s Generation, 646 tok/s Prefill, TTFT 16.076 ms (3 Laufe)AMD Ryzen 9 7945HX wi...AMD RYZEN AI MAX+ 395 w/ Radeon 8060S - 74,4 tok/s Generation, 1.168 tok/s Prefill, TTFT 12.711 ms (2 Laufe)AMD RYZEN AI MAX+ 395...Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz - 14,5 tok/s Generation, 88 tok/s Prefill, TTFT 122.346 ms (3 Laufe)Intel(R) Xeon(R) CPU ...NVIDIA RTX Spark N1X (6144-core GPU, 20-core CPU) - 200,5 tok/s Generation, 3.459 tok/s Prefill, TTFT 3.538 ms (4 Laufe) | DIESER LAUF★ NVIDIA RTX Spark N1X ...
AMD Ryzen 9 9950X 16-Core Processor 1.995,4 tok/sAMD Ryzen 5 5600X 6-Core Processor 1.981,1 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 1.890,7 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 1.835,7 tok/sAMD Ryzen Threadripper PRO 5975WX 32-Cores 1.210,3 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 1.113,5 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 703,8 tok/sAMD Ryzen Threadripper PRO 3955WX 16-Cores 357,7 tok/sAMD Ryzen 3 3100 4-Core Processor 323,6 tok/sIntel(R) Xeon(R) CPU E5-4657L v2 @ 2.40GHz 247,3 tok/sNVIDIA Grace 233,3 tok/sAMD Ryzen 3 PRO 3200GE w/ Radeon Vega Graphics 206,6 tok/s★ NVIDIA RTX Spark N1X (6144-core GPU, 20-core CPU) 200,5 tok/s this runAMD Ryzen 9 7945HX with Radeon Graphics 111,1 tok/sAMD RYZEN AI MAX+ 395 w/ Radeon 8060S 74,4 tok/sIntel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 14,5 tok/s

MBby mainboard

2.5791.9341.2906450,008.04516.09024.135Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI - 1.995,4 tok/s Generation, 19.558 tok/s Prefill, TTFT 1.895 ms (8 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 1.981,1 tok/s Generation, 11.631 tok/s Prefill, TTFT 3.845 ms (24 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 1.835,7 tok/s Generation, 6.778 tok/s Prefill, TTFT 14.338 ms (60 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 1.210,3 tok/s Generation, 3.625 tok/s Prefill, TTFT 4.938 ms (27 Laufe)ASUSTeK COMPUTER INC....Meigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) - 1.113,5 tok/s Generation, 6.303 tok/s Prefill, TTFT 4.194 ms (18 Laufe)Meigao Innovation Tec...ASUSTeK COMPUTER INC. PRIME A520M-K - 572,1 tok/s Generation, 8.817 tok/s Prefill, TTFT 1.101 ms (4 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX X570-F GAMING - 323,6 tok/s Generation, 3.603 tok/s Prefill, TTFT 3.303 ms (6 Laufe)ASUSTeK COMPUTER INC....Dell Inc. PowerEdge R820 - 247,3 tok/s Generation, 2.229 tok/s Prefill, TTFT 62.264 ms (6 Laufe)Dell Inc. PowerEdge R...ASUSTeK COMPUTER INC. GX10 - 233,3 tok/s Generation, 3.963 tok/s Prefill, TTFT 2.535 ms (6 Laufe)ASUSTeK COMPUTER INC....Shenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) - 111,1 tok/s Generation, 646 tok/s Prefill, TTFT 16.076 ms (3 Laufe)Shenzhen Meigao Elect...Bosgame AXB35-02 (BeyondMax Series) - 74,4 tok/s Generation, 1.168 tok/s Prefill, TTFT 12.711 ms (2 Laufe)Bosgame AXB35-02 (Bey...ASUS H7607BA - 200,5 tok/s Generation, 3.459 tok/s Prefill, TTFT 3.538 ms (4 Laufe) | DIESER LAUF★ ASUS H7607BA
ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.995,4 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 1.981,1 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 1.835,7 tok/sASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 1.210,3 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 1.113,5 tok/sASUSTeK COMPUTER INC. PRIME A520M-K 572,1 tok/sASUSTeK COMPUTER INC. ROG STRIX X570-F GAMING 323,6 tok/sDell Inc. PowerEdge R820 247,3 tok/sASUSTeK COMPUTER INC. GX10 233,3 tok/s★ ASUS H7607BA 200,5 tok/s this runShenzhen Meigao Electronic Equipment Co.,Ltd F1FXM (DeskMini Series) 111,1 tok/sBosgame AXB35-02 (BeyondMax Series) 74,4 tok/s

ENGby engine

2.5641.9231.2826410,02.7815.9719.16212.352Prefill (tok/s)Generation (tok/s)vLLM - 1.443,9 tok/s Generation, 10.617 tok/s Prefill, TTFT 5.296 ms (35 Laufe)vLLMunbekannt - 147,8 tok/s Generation, 4.516 tok/s Prefill, TTFT 616 ms (6 Laufe)unbekanntllama.cpp - 1.995,4 tok/s Generation, 6.040 tok/s Prefill, TTFT 11.718 ms (127 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 1.995,4 tok/s this runvLLM 1.443,9 tok/sunbekannt 147,8 tok/s

DRVby driver

2.5641.9231.2826410,02.9614.6906.4208.149Prefill (tok/s)Generation (tok/s)unbekannt - 1.995,4 tok/s Generation, 7.147 tok/s Prefill, TTFT 10.631 ms (156 Laufe)unbekanntNVIDIA 590.48.01 / CUDA 13.1 - 233,3 tok/s Generation, 3.963 tok/s Prefill, TTFT 2.535 ms (6 Laufe)NVIDIA 590.48.01 / CU...AMD 7.0.0-27-generic - 147,8 tok/s Generation, 4.516 tok/s Prefill, TTFT 616 ms (6 Laufe)AMD 7.0.0-27-generic
unbekannt 1.995,4 tok/sNVIDIA 590.48.01 / CUDA 13.1 233,3 tok/sAMD 7.0.0-27-generic 147,8 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 %
🔌 Idle 0 W
⚡ TDP 10 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)0 W missing
Avg cost / hour–
Electricity / 1M tokens–
Token / kWh–
Acquisition (system)– missing
Electricity (2 years)–
TCO (2 years)–
Output tokens (2 years)7.55B
☁️ 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 (0 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

gpt-oss-20bNVIDIA RTX SPARK (N1X)gpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Editiongpt-oss-20bNVIDIA RTX PRO 6000 Blackwell Workstation Editiongpt-oss-20bNVIDIA GeForce RTX 5090
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.