AMD · consumer

Radeon RX 6650 XT

Radeon RX 6650 XT has 8 GB of VRAM at 280 GB/s — about 7.44 GiB usable after driver and compositor overhead. 924 of 2118 indexed models fit at 64K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
280 GB/s
128-bit bus
Tensor FP16
—
dense
TDP
180 W
$399 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 749audio asr 38embedding 24vision language 85audio tts 20video 7image 1

What fits at 64K context

largest quantization that fits, per model · 924 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Ministral-3-8B-Instruct-2512-BF16-abliteratedI1-IQ1_S8.9B1.98 GiB4.52 GiB7.43 GiB0.01 GiB27±26.5%
Amaretto-8BI1-IQ1_S8.9B1.98 GiB4.52 GiB7.43 GiB0.01 GiB27±26.5%
Foundation-Sec-8B-InstructI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Foundation-Sec-8B-Instruct-hereticI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Meta-Llama-3-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
llama3.1-heretic-unsensoredI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Anubis-Mini-8B-v1I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
dolphin-2.9-llama3-8bIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Meta-Llama-3-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8B-hereticI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Gluon-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-ReasoningI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama3.3-8B-Instruct-Thinking-Claude-4.5-Opus-High-ReasoningI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Hypnos-i1-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
grok-oss-Apollyon-8B-hereticI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.3-8B-Instruct-128K-JbliteratedI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
PsyCoPref-Llama3-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama3.1-GptDeluxe-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
SelfCite-8B-CC-SFTI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
ai-girlfriend-v2I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.3-8B-Instruct-128K_AbliteratedI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Aisha-Uncensored-v2I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
grok-oss-Apollyon-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
HARC-Llama-3.1-8B-InstructI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Lumen-1.2.5I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.3-8B-Instruct-128KI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Stoicism1_Llama3.1-8b-instructI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
grok-oss-Thanatos-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
OpenElla-NovelWriter-Requiem-AscendedI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
llama-joycaption-beta-one-hf-llavaI1-IQ2_XXS8.5B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.1-Swallow-8B-Instruct-v0.5IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
DarkIdol-Llama-3.1-8B-Instruct-1.3-UncensoredI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
calme-2.3-legalkit-8bI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama3.1-DilemmaDeluxe-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.1-8B-Stheno-v3.4I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
DarkIdol-Llama-3.1-8B-Instruct-1.0-UncensoredI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
SzilviaB-Daredevil-LongWriter-8B_abliteratedI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.1-SuperNova-LiteI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Humanish-Roleplay-Llama-3.1-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.1_OpenScholar-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Meta-Llama-3.1-8B-ClaudeI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.1-Storm-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Meta-Llama-3.1-8B-Instruct-abliteratedI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3.3-8B-InstructI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Infinity-Instruct-7M-Gen-Llama3_1-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
markdownizerI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-SEA-LION-v2-8BI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Meta-Llama-3-8B-Instruct-abliterated-v3I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
llama3-8B-DarkIdol-2.3-Uncensored-32KI1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama-3-Instruct-8B-SPPO-Iter3I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Hathor_Unstable-L3-8B-v0.3I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Hathor_Tahsin-L3-8B-v0.85I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
L3-8B-Stheno-v3.2I1-IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Hermes-3-Llama-3.1-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Llama3-ChatQA-1.5-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
openchat-3.6-8b-20240522IQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
llama-3-8b-Instruct-bnb-4bitIQ2_XXS8.2B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
L3-8B-Stheno-v3.3-32KIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
Hermes-2-Theta-Llama-3-8BIQ2_XXS8.0B2.23 GiB4.25 GiB7.42 GiB0.02 GiB27±26.5%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Image generation3.97 it/s2.63–4.85138
Benchmarked· n=138

Aggregated from community-submitted runs, so the spread is wide by nature — it covers different models, resolutions, step counts and settings, not one controlled configuration. Read the middle 50% rather than the median alone. These figures are reproduced with attribution from vladmandic-sd-data-benchmark, which publishes no licence — so we display and link rather than redistribute them.

Questions people ask

What AI models can a Radeon RX 6650 XT run?
924 of 2118 indexed open-weight models fit a Radeon RX 6650 XT at 65,536 context with q8_0 KV cache, the largest being Ministral-3-8B-Instruct-2512-BF16-abliterated at I1-IQ1_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 6650 XT actually have?
Its nameplate is 8 GB, but about 7.44 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon RX 6650 XT fast for local AI?
Its memory bandwidth is 280 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.