AMD-Pro-BenchCommunity

13,812 benchmarks in 32 GPU combinations documented

DeepSeek-V4-Flash on AMD GPUs

20 results · 3 sources · 59 measurements

Explore all results

Grouped by reported model identity. Exact weights, quantizations and software can differ. How results are reviewed

Documented results

20 selected tests
View
ReportLocalMaxxing		DeepSeek-V4-FlashMQ2RModel ref. ↗Token speed—
201.6 tok/sPrefill · median · per request
30.75 tok/sGeneration · median · per request
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Ubuntu 26.04
Engine
hipfire
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 128 tokens
Context
128
Concurrency
1
Date
2026-08-10
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashMQ2RModel ref. ↗Token speed—
389.19 tok/sPrefill · median · per request
54.27 tok/sGeneration · median · per request
4 × R9700Tensor parallel · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Ubuntu 26.04
Engine
hipfire
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
2052 / 512 tokens
Context
2052
Concurrency
1
Date
2026-08-10
Full record →
View
ReportGeneric LLM		DeepSeek-V4-Flash2.875 bpw routed experts; FP8 dense weightsModel ref. ↗Token speed——
~8 tok/sGeneration · per request
1 × R9700Single GPU · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Not reported
Engine
Affinity
Backend
HIP / ROCm 7.x
Harness
Not reported
Prompt / output
? / ? tokens
Context
Not reported
Concurrency
Not reported
Date
Not reported
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashMQ2-LloydModel ref. ↗Token speed—
144.5 tok/sPrefill · unknown
15.2 tok/sGeneration · unknown
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Not reported
Engine
hipfire
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 0 tokens
Context
16384
Concurrency
Not reported
Date
2026-07-12
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
269.68 tok/sPrefill · unknown
30.78 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
View
ReportGeneric LLM		DeepSeek-V4-Flash2.875 bpw routed experts; FP8 dense weightsModel ref. ↗Token speed——
~110 tok/sGeneration · per request
2 × R9700Tensor parallel · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Not reported
Engine
Affinity
Backend
HIP / ROCm 7.x
Harness
Not reported
Prompt / output
? / ? tokens
Context
Not reported
Concurrency
Not reported
Date
Not reported
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashQ2_KModel ref. ↗Token speed—
218.59 tok/sPrefill · unknown
14.13 tok/sGeneration · unknown
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Not reported
Engine
dwarfstar
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 0 tokens
Context
16384
Concurrency
Not reported
Date
2026-07-12
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
274.29 tok/sPrefill · unknown
30.93 tok/sGeneration · peak · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashQ2_KModel ref. ↗Token speed—
178.83 tok/sPrefill · unknown
15.26 tok/sGeneration · unknown
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Not reported
Engine
dwarfstar
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 0 tokens
Context
2048
Concurrency
Not reported
Date
2026-07-12
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashMQ2RModel ref. ↗Conflicting evidenceToken speed—
481.02 tok/sPrefill · median · per request
53.1 tok/sGeneration · median · per request
GPU count unknownTensor parallel · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Ubuntu 26.04
Engine
hipfire
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
2052 / 512 tokens
Context
2052
Concurrency
1
Date
2026-08-10
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashQ2_KModel ref. ↗Token speed—
193.42 tok/sPrefill · unknown
12.39 tok/sGeneration · unknown
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Not reported
Engine
dwarfstar
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 0 tokens
Context
65536
Concurrency
Not reported
Date
2026-07-12
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashUD-IQ1_SModel ref. ↗Token speed—
224.27 tok/sPrefill · unknown
7.15 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
8192
Concurrency
Not reported
Date
2026-07-17
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashQ2_KModel ref. ↗Token speed—
90 tok/sPrefill · unknown
15.6 tok/sGeneration · unknown
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Not reported
Engine
dwarfstar
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
0 / 0 tokens
Context
16384
Concurrency
Not reported
Date
2026-07-11
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashUD-IQ1_SModel ref. ↗Token speed—
265.43 tok/sPrefill · unknown
6.98 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
4183 / 256 tokens
Context
8192
Concurrency
Not reported
Date
2026-07-17
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashMQ2RModel ref. ↗Token speed——
39 tok/sGeneration · median · per request
1 × Ryzen AI Max+ 395Single GPU · Unified memory
Configuration
Memory
Unified memory
OS
Ubuntu 26.04
Engine
hipfire
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
25 / 128 tokens
Context
4096
Concurrency
1
Date
2026-08-10
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashUD-IQ1_SModel ref. ↗Token speed—
242.24 tok/sPrefill · unknown
7.11 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
2099 / 256 tokens
Context
8192
Concurrency
Not reported
Date
2026-07-17
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
49.29 tok/sPrefill · unknown
7.38 tok/sGeneration · peak · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
119.48 tok/sPrefill · unknown
9.21 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
118.6 tok/sPrefill · unknown
9.25 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
vulkan
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
View
ReportLocalMaxxing		DeepSeek-V4-FlashAD-IQ1_MModel ref. ↗Token speed—
278.86 tok/sPrefill · unknown
21.39 tok/sGeneration · unknown
3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPU
Configuration
Memory
32 GB VRAM/GPU
OS
Arch Linux
Engine
llama.cpp
Backend
rocm
Harness
LocalMaxxing community speed test; exact client where reported
Prompt / output
526 / 256 tokens
Context
782
Concurrency
Not reported
Date
2026-08-04
Full record →
Quantizations and software

Quantizations: 2.875 bpw routed experts; FP8 dense weights, AD-IQ1_M, MQ2-Lloyd, MQ2R, Q2_K, UD-IQ1_S
Engines: Affinity, dwarfstar, hipfire, llama.cpp

Explore all 20 results →

This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports

GPU benchmark pages: R9700 · Ryzen AI Max+ 395