2 × RX 9070 XT LLM benchmarks

40 results · 120 measurements · 2 sources

Explore all results Image benchmarks Video benchmarks Submit a result

Device specifications

Variant	Memory	Memory type	Bus	Bandwidth	FP16 matrix	INT8 matrix
RX 9070 XT ↗	16 GB	GDDR6	256-bit	640 GB/s	195 TFLOPS	389 TOPS

Manufacturer specifications. Shared system RAM is not dedicated VRAM. Unreported values remain —; measured LLM results retain their reported configuration.

Selected reported tests

24 of 40 results

Source-reported results, selected across reports by documentation quality. Match configurations before comparing values.


View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	17.14 tok/sThroughput · per request	1,592.45 tok/sPrefill · per request	23.5 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 6393 / 256 tokens Context 12288 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	14.2 tok/sThroughput · per request	2,061.11 tok/sPrefill · per request	24.7 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 8386 / 128 tokens Context 16384 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	25.46 tok/sThroughput · per request	1,545.58 tok/sPrefill · per request	43.56 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 6393 / 256 tokens Context 12288 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	16.34 tok/sThroughput · per request	1,817.81 tok/sPrefill · per request	36.73 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 8386 / 128 tokens Context 16384 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	12.73 tok/sThroughput · per request	1,229.31 tok/sPrefill · per request	26.55 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 6393 / 128 tokens Context 12288 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	9.76 tok/sThroughput · per request	1,879.06 tok/sPrefill · per request	25.74 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 32996 / 256 tokens Context 49152 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	17.26 tok/sThroughput · per request	1,320.21 tok/sPrefill · per request	50.56 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 6393 / 128 tokens Context 12288 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	11.2 tok/sThroughput · per request	1,799.29 tok/sPrefill · per request	43.75 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 32996 / 256 tokens Context 49152 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	5.43 tok/sThroughput · mean · per request	1,715.65 tok/sPrefill · mean · per request	20.35 tok/sGeneration · mean · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 29561 / 128 tokens Context 49152 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	4.96 tok/sThroughput · per request	1,555.55 tok/sPrefill · per request	24.85 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 64287 / 256 tokens Context 98304 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	6.06 tok/sThroughput · per request	1,656.03 tok/sPrefill · per request	39.9 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 29561 / 128 tokens Context 49152 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	5.17 tok/sThroughput · per request	1,495.92 tok/sPrefill · per request	39.09 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 64287 / 256 tokens Context 98304 Concurrency 1 Date 2026-09-26 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	5.01 tok/sThroughput · per request	1,432.13 tok/sPrefill · per request	26.36 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 29561 / 128 tokens Context 49152 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	12.33 tok/sThroughput · per request	1,569.17 tok/sPrefill · per request	25.42 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 8386 / 128 tokens Context 16384 Concurrency 1 Date 2026-09-23 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	5.32 tok/sThroughput · per request	1,389.59 tok/sPrefill · per request	47.21 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 29561 / 128 tokens Context 49152 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	15.06 tok/sThroughput · per request	1,606.98 tok/sPrefill · per request	39 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 8386 / 128 tokens Context 16384 Concurrency 1 Date 2026-09-23 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	5.43 tok/sThroughput · per request	1,716.66 tok/sPrefill · per request	20.28 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 29561 / 128 tokens Context 49152 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	7.69 tok/sThroughput · per request	1,470.8 tok/sPrefill · per request	23.61 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 32996 / 256 tokens Context 49152 Concurrency 1 Date 2026-09-23 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	2.71 tok/sThroughput · per request	1,466.27 tok/sPrefill · per request	18.77 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 58982 / 128 tokens Context 98304 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	9.1 tok/sThroughput · per request	1,580.02 tok/sPrefill · per request	35.36 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 32996 / 256 tokens Context 49152 Concurrency 1 Date 2026-09-23 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	2.29 tok/sThroughput · per request	1,171.17 tok/sPrefill · per request	24.18 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 58982 / 128 tokens Context 98304 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	4.05 tok/sThroughput · per request	1,243.99 tok/sPrefill · per request	22.19 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 64287 / 256 tokens Context 98304 Concurrency 1 Date 2026-09-23 Full record →
View ReportCoding session	Qwen3.6-27BQ4_K_MModel ref. ↗	Token speed	0.72 tok/sThroughput · per request	447.59 tok/sPrefill · per request	17.19 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend ROCm / HIP Harness agent_workload_bench.py Prompt / output 75979 / 128 tokens Context 131072 Concurrency 1 Date 2026-07-15 Full record →
View ReportCoding session	Qwen3.8-27BUD-Q4_K_MModel ref. ↗	Token speed	4.69 tok/sThroughput · per request	1,357.37 tok/sPrefill · per request	35.14 tok/sGeneration · per request	2 × RX 9070 XTLayer split · 16 GB VRAM/GPU Configuration Memory 16 GB VRAM/GPU OS Windows 11 Engine llama.cpp RDNA lab fork Backend Vulkan Harness agent_workload_bench.py Prompt / output 64287 / 256 tokens Context 98304 Concurrency 1 Date 2026-09-23 Full record →

Explore all 40 results →

Models and software in this collection

Model families: Qwen3.6, Qwen3.8

Engines: llama.cpp RDNA lab fork

Methodology and evidence standards · Download the research dataset

2 × RX 9070 XT LLM benchmarks

Selected reported tests

LLMs benchmarked on the RX 9070 XT