2-GPU systems including RX 7900 XT LLM benchmarks

51 results · 64 measurements · 3 sources

Explore all results Image benchmarks Video benchmarks Submit a result

Device specifications

Variant	Memory	Memory type	Bus	Bandwidth	FP16 matrix	INT8 matrix
RX 7900 XT ↗	20 GB	GDDR6	320-bit	800 GB/s	103 TFLOPS	103 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 51 results

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


View ReportLocalMaxxing	Nemotron-3.5-Lightning-30B-A3BNVFP4Weights ↗	Token speed	—	1,191.7 tok/sPrefill · per request	116 tok/sGeneration · per request	2 × RX 7900 XTMulti-GPU · split unknown · 20 GB VRAM/GPU Configuration Memory 20 GB VRAM/GPU OS Linux Engine llama.cpp Backend Not reported Harness LocalMaxxing community speed test; exact client where reported Prompt / output 39 / 256 tokens Context 2048 Concurrency 1 Date 2026-08-12 Full record →
View ReportGeneric LLM	Qwen3.5-9BUD-Q4_K_XLModel ref. ↗	Token speed	—	2,781 tok/sPrefill · per request	61.06 tok/sGeneration · per request	Mixed: R9700 + RX7900XT · 2 GPUsPipeline parallel Configuration Memory Not reported OS Debian GNU/Linux13.3 / kernel6.18.2-zen4 Engine llama.cpp Backend Vulkan Harness Author repeated llama-server prompt loop Prompt / output ? / ? tokens Context 262144 Concurrency 1 Date Not reported Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	27.81 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 11 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportLocalMaxxing	Nemotron-3.5-Lightning-30B-A3BNVFP4Weights ↗	Token speed	—	535.3 tok/sPrefill · per request	115.6 tok/sGeneration · per request	2 × RX 7900 XTMulti-GPU · split unknown · 20 GB VRAM/GPU Configuration Memory 20 GB VRAM/GPU OS Linux Engine llama.cpp Backend Not reported Harness LocalMaxxing community speed test; exact client where reported Prompt / output 39 / 128 tokens Context 2048 Concurrency 1 Date 2026-08-12 Full record →
View ReportGeneric LLM	Qwen3.5-9BUD-Q4_K_XLModel ref. ↗	Token speed	—	1,260 tok/sPrefill · mean · per request	60.18 tok/sGeneration · mean · per request	Mixed: R9700 + RX7900XT · 2 GPUsPipeline parallel Configuration Memory Not reported OS Debian GNU/Linux13.3 / kernel6.18.2-zen4 Engine llama.cpp Backend Vulkan Harness Author repeated llama-server prompt loop Prompt / output ? / ? tokens Context 262144 Concurrency 1 Date Not reported Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	27.03 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 7887 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportLocalMaxxing	Nemotron-3.5-Lightning-30B-A3BNVFP4Weights ↗	Token speed	—	477.1 tok/sPrefill · per request	59.9 tok/sGeneration · per request	2 × RX 7900 XTMulti-GPU · split unknown · 20 GB VRAM/GPU Configuration Memory 20 GB VRAM/GPU OS Linux Engine llama.cpp Backend Not reported Harness LocalMaxxing community speed test; exact client where reported Prompt / output 39 / 128 tokens Context 2048 Concurrency 1 Date 2026-08-12 Full record →
View ReportGeneric LLM	Qwen3.5-9BUD-Q4_K_XLModel ref. ↗	Token speed	—	2,559 tok/sPrefill · per request	49.79 tok/sGeneration · per request	Mixed: R9700 + RX7900XT · 2 GPUsPipeline parallel Configuration Memory Not reported OS Debian GNU/Linux13.3 / kernel6.18.2-zen4 Engine llama.cpp Backend ROCm / HIP Harness Author repeated llama-server prompt loop Prompt / output ? / ? tokens Context 262144 Concurrency 1 Date Not reported Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	1,148.54 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 7887 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportGeneric LLM	Qwen3.5-9BUD-Q4_K_XLModel ref. ↗	Token speed	—	839 tok/sPrefill · mean · per request	48.87 tok/sGeneration · mean · per request	Mixed: R9700 + RX7900XT · 2 GPUsPipeline parallel Configuration Memory Not reported OS Debian GNU/Linux13.3 / kernel6.18.2-zen4 Engine llama.cpp Backend ROCm / HIP Harness Author repeated llama-server prompt loop Prompt / output ? / ? tokens Context 262144 Concurrency 1 Date Not reported Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	26.26 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 16050 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	1,090.8 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 16050 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	24.69 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 32460 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	1,028.05 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 32460 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	21.77 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 64787 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	869.68 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 64787 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	18.89 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 96393 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	691.44 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 96393 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	—	15.66 tok/sGeneration · median · per request	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 128266 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedladder	—	591.63 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 128266 / 1000 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedfresh_prefill	—	641.57 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 512 / 64 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedfresh_prefill	—	857.52 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 1978 / 64 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedfresh_prefill	—	1,154.68 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 8077 / 64 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →
View ReportServing benchmark	Huihui-Qwen3.8-27B-abliteratedUD-IQ4_XSModel ref. ↗	Token speedreal_prompts	—	306.22 tok/sPrefill · median · per request	—	Mixed: AMD Radeon RX 7900 XT + AMD Radeon RX 7800 XT · 2 GPUsLayer split Configuration Memory Not reported OS Ubuntu 26.04.1 LTS / Linux 7.0.0-34-generic Engine llama.cpp Backend ROCm / HIP Harness llama-split-bench Prompt / output 73 / 1200 tokens Context 131072 Concurrency 1 Date 2026-09-28 Full record →

Explore all 51 results →

Models and software in this collection

Model families: Qwen3.5

Engines: llama.cpp

Methodology and evidence standards · Download the research dataset

2-GPU systems including RX 7900 XT LLM benchmarks

Selected reported tests

LLMs benchmarked on the RX 7900 XT