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

DeepSeek-V4-Flash-284B-A13B

Performance benchmark · measured on 06.08.2026 19:03

Benchmark-IDrun-20260807-053423-5b93d0
Timebench 3 - Kombi (Prefill + Generation)MoE284BRuntime: llama.cppQuantisierung: UD-Q4_K_XL
Generation18,80tok/s
Prefill53,04tok/s
Time to First Token260.848,50ms
Total duration1.200,00s
Concurrency5parallel
Ranking in the field
154of 203 systems

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

This run is better than 24 % of all comparable systems.
Generation 18,8 tok/s
-76 % vs Ø 79,5
Prefill 53,0 tok/s
-97 % vs Ø 1.742,0
Time to First Token 260.849 ms
+222 % vs Ø 81.026
Distribution in the field0 – 382 tok/s
Ø 80 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: 3x AMD Radeon AI PRO R9700 · 32 GB VRAM
CPU: AMD Ryzen Threadripper PRO 7955WX 16-Cores
RAM: 184 GB
Mainboard: ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE

Setup

Runtime: llama.cpp
Quantization: UD-Q4_K_XL
Model: DeepSeek-V4-Flash-284B-A13B

Configuration

benchmark-konfiguration — run-20260807-053423-5b93d0
# LLM-Benchmark Konfiguration # Modell : DeepSeek-V4-Flash-284B-A13B # Engine : llama.cpp # Run-ID : run-20260807-053423-5b93d0 # GPU : 3x AMD Radeon AI PRO R9700 # CPU : AMD Ryzen Threadripper PRO 7955WX 16-Cores # RAM : 184 GB bench@llm-benchmark:~$ /home/godcore/llama.cpp/build/bin/llama-server \ -m /home/godcore/.cache/huggingface/hub/models--unsloth--DeepSeek-V4-Flash-GGUF/snapshots/e3aa0d6a5fa4f820d9e132ac1fd1d01e1b2b49e0/UD-Q4_K_XL/DeepSeek-V4-Flash-UD-Q4_K_XL-00001-of-00005.gguf \ --alias DeepSeek-V4-Flash-284B-A13B \ --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.DeepSeek-V4-Flash-284B-A13B
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--DeepSeek-V4-Flash-GGUF/snapshots/e3aa0d6a5fa4f820d9e132ac1fd1d01e1b2b49e0/UD-Q4_K_XL/DeepSeek-V4-Flash-UD-Q4_K_XL-00001-of-00005.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

DeepSeek-V4-Flash-284B-A13B 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

5013762511250,008471.6952.542Prefill (tok/s)Generation (tok/s)NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition - 385,9 tok/s Generation, 2.052 tok/s Prefill, TTFT 13.971 ms (6 Laufe)NVIDIA RTX PRO 6000 B...NVIDIA GeForce RTX 5070 Ti - 10,0 tok/s Generation, 69 tok/s Prefill, TTFT 132.055 ms (3 Laufe)NVIDIA GeForce RTX 50...NVIDIA GeForce RTX 5090 - 3,2 tok/s Generation, 27 tok/s Prefill, TTFT 240.091 ms (3 Laufe)NVIDIA GeForce RTX 50...CPU-only - 1,2 tok/s Generation, 14 tok/s Prefill, TTFT 148.315 ms (1 Lauf)CPU-onlyAMD Radeon AI PRO R9700 - 18,8 tok/s Generation, 58 tok/s Prefill, TTFT 165.244 ms (10 Laufe) | DIESER LAUF★ AMD Radeon AI PRO R97...
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 385,9 tok/s★ AMD Radeon AI PRO R9700 18,8 tok/s this runNVIDIA GeForce RTX 5070 Ti 10,0 tok/sNVIDIA GeForce RTX 5090 3,2 tok/sCPU-only 1,2 tok/s

CPUby processor

5013762511250,008471.6952.542Prefill (tok/s)Generation (tok/s)AMD Ryzen Threadripper PRO 9965WX 24-Cores - 385,9 tok/s Generation, 2.052 tok/s Prefill, TTFT 13.971 ms (6 Laufe)AMD Ryzen Threadrippe...AMD Ryzen Threadripper PRO 5975WX 32-Cores - 10,0 tok/s Generation, 69 tok/s Prefill, TTFT 132.055 ms (3 Laufe)AMD Ryzen Threadrippe...AMD Ryzen 7 5800X3D 8-Core Processor - 3,2 tok/s Generation, 27 tok/s Prefill, TTFT 240.091 ms (3 Laufe)AMD Ryzen 7 5800X3D 8...Intel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz - 1,2 tok/s Generation, 14 tok/s Prefill, TTFT 148.315 ms (1 Lauf)Intel(R) Xeon(R) CPU ...AMD Ryzen Threadripper PRO 7955WX 16-Cores - 18,8 tok/s Generation, 58 tok/s Prefill, TTFT 165.244 ms (10 Laufe) | DIESER LAUF★ AMD Ryzen Threadrippe...
AMD Ryzen Threadripper PRO 9965WX 24-Cores 385,9 tok/s★ AMD Ryzen Threadripper PRO 7955WX 16-Cores 18,8 tok/s this runAMD Ryzen Threadripper PRO 5975WX 32-Cores 10,0 tok/sAMD Ryzen 7 5800X3D 8-Core Processor 3,2 tok/sIntel(R) Xeon(R) CPU E5-4620 v2 @ 2.60GHz 1,2 tok/s

MBby mainboard

5013762511250,00332665997Prefill (tok/s)Generation (tok/s)ASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI - 10,0 tok/s Generation, 69 tok/s Prefill, TTFT 132.055 ms (3 Laufe)ASUSTeK COMPUTER INC....ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING - 3,2 tok/s Generation, 27 tok/s Prefill, TTFT 240.091 ms (3 Laufe)ASUSTeK COMPUTER INC....Dell Inc. PowerEdge R820 - 1,2 tok/s Generation, 14 tok/s Prefill, TTFT 148.315 ms (1 Lauf)Dell Inc. PowerEdge R...ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE - 385,9 tok/s Generation, 806 tok/s Prefill, TTFT 108.517 ms (16 Laufe) | DIESER LAUF★ ASUSTeK COMPUTER INC....
★ ASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 385,9 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 10,0 tok/sASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 3,2 tok/sDell Inc. PowerEdge R820 1,2 tok/s

ENGby engine

425405386367347539562585608Prefill (tok/s)Generation (tok/s)llama.cpp - 385,9 tok/s Generation, 574 tok/s Prefill, TTFT 130.479 ms (23 Laufe) | DIESER LAUF★ llama.cpp
★ llama.cpp 385,9 tok/s this run

DRVby driver

425405386367347539562585608Prefill (tok/s)Generation (tok/s)unbekannt - 385,9 tok/s Generation, 574 tok/s Prefill, TTFT 130.479 ms (23 Laufe)unbekannt
unbekannt 385,9 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 125 W
⚡ TDP 971 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)971 W estimated (TDP)GPU 900 + CPU 46 + Board 25 W full load
Avg cost / hourEUR 0.29
Electricity / 1M tokensEUR 4.30
Token / kWh69.70K
Acquisition (system)EUR 9,674 full priceGPU EUR 4,200 · CPU EUR 1,399 · Board EUR 1,299 · RAM EUR 2,576 · PSU EUR 200
Electricity (2 years)
TCO (2 years)EUR 14,778
Output tokens (2 years)1.19B
☁️ 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 (125 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

DeepSeek-V4-Flash-284B-A13B3x AMD Radeon AI PRO R9700DeepSeek-V4-Flash-284B-A13B3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionDeepSeek-V4-Flash-284B-A13B3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation EditionDeepSeek-V4-Flash-284B-A13B3x AMD Radeon AI PRO R9700
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.