AMD · workstation

Radeon Pro W7800

Radeon Pro W7800 has 32 GB of VRAM at 576 GB/s — about 29.76 GiB usable after driver and compositor overhead. 1949 of 2118 indexed models fit at 128K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
GDDR6
Bandwidth
576 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
260 W
$2499 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1668video 16vision language 177embedding 26image 2audio tts 21audio asr 39

What fits at 128K context

largest quantization that fits, per model · 1949 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
glm-4-9b-chat-1mQ4_K_L9.5B6.30 GiB22.50 GiB29.75 GiB0.01 GiB13±26.5%
Gemma-4-Novelist-Eclipse-31BQ5_K_L32.7B22.77 GiB5.95 GiB29.70 GiB0.06 GiB13±26.5%
Gemma-4-31B-StyleTuneQ5_K_L32.7B22.77 GiB5.95 GiB29.70 GiB0.06 GiB13±26.5%
glm-4-9b-chat-abliteratedQ4_K_L9.4B6.25 GiB22.50 GiB29.70 GiB0.06 GiB13±26.5%
glm-4-9b-chatQ4_K_L9.4B6.25 GiB22.50 GiB29.70 GiB0.06 GiB13±26.5%
codegeex4-all-9bQ5_K_S9.4B6.23 GiB22.50 GiB29.68 GiB0.08 GiB13±26.5%
TildeOpen-30B-Instruct-LVI1-Q5_K_M30.7B20.26 GiB8.44 GiB29.68 GiB0.08 GiB13±26.5%
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.66 GiB0.10 GiB13±26.5%
Qwen3.6-28BMoEQ8_028.2B28.00 GiB0.70 GiB29.61 GiB0.15 GiB57±37%
ALIA-40b-fc-2606I1-Q4_K_S40.4B21.84 GiB6.75 GiB29.60 GiB0.16 GiB13±26.5%
ALIA-40b-instruct-2606I1-Q4_K_S40.4B21.84 GiB6.75 GiB29.60 GiB0.16 GiB13±26.5%
Darwin-35B-A3B-OpusMoEQ6_K36.0B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
Aurora-Code-1MoEQ6_K34.7B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
grug-35b-v2MoEQ6_K35.1B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
grug-35bMoEQ6_K35.1B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
WorldSim-Opus-3.6-35B-A3BMoEQ6_K35.1B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
Qwen3.6-35B-A3B-AnkoMoEQ6_K35.1B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
KAT-Coder-V2.5-DevMoEQ6_K34.7B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
Ornith-1.0-35BMoEQ6_K34.7B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
Nex-N2-miniMoEQ6_K35.1B27.99 GiB0.70 GiB29.60 GiB0.16 GiB61±37%
Skyfall-31B-v4.2Q5_K_M31.4B20.97 GiB7.59 GiB29.58 GiB0.18 GiB13±26.5%
Qwen3.5-99BMoEI1-IQ2_S99.0B27.80 GiB0.84 GiB29.57 GiB0.19 GiB56±37%
Qwen3.6-35B-A3BMoEUD-Q6_K36.0B27.95 GiB0.70 GiB29.56 GiB0.20 GiB61±37%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q4_K_M42.4B23.94 GiB4.71 GiB29.54 GiB0.22 GiB27±37%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-Q5_K_S39.5B25.20 GiB3.38 GiB29.54 GiB0.22 GiB13±26.5%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-Q5_K_S39.5B25.20 GiB3.38 GiB29.54 GiB0.22 GiB13±26.5%
Qwen3.5-40B-Claude-4.5-Opus-High-Reasoning-Thinking-uncensored-hereticQ5_K_S39.5B25.20 GiB3.38 GiB29.54 GiB0.22 GiB13±26.5%
Qwen3.6-27B-NVFP4NVFP421.2B26.29 GiB2.25 GiB29.50 GiB0.26 GiB13±26.5%
Laguna-XS-2.1MoEQ6_K_L33.4B27.14 GiB1.44 GiB29.47 GiB0.29 GiB51±37%
Noromaid-v0.4-Mixtral-Instruct-8x7b-ZlossMoEIQ4_XS46.7B24.02 GiB4.50 GiB29.46 GiB0.30 GiB18±37%
gemma-4-E4B-it-Uncensored-MAXF328.0B28.02 GiB0.51 GiB29.45 GiB0.31 GiB13±26.5%
Hypernova-60B-2605MoEI1-Q2_K_S58.7B27.42 GiB1.13 GiB29.44 GiB0.32 GiB49±37%
KrakenSakura-Maelstrom-12B-v1F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-Base-2407F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Vikhr-Nemo-12B-Instruct-R-21-09-24F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Wayfarer-2-12BBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Wayfarer-12BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
writing-roleplay-20k-context-nemo-12b-v1.0F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Muse-12BBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Lumimaid-v0.2-12BBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
mini-magnum-12b-v1.1BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Dans-PersonalityEngine-V1.3.0-12bBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Lumimaid-Magnum-v4-12BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Captain-Eris_Violet-V0.420-12BBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-Gutenberg-Doppel-12BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-Instruct-2407F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
MN-12b-RP-InkF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
magnum-v4-12bF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-12B-ArliAI-RPMax-v1.2BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-12B-ArliAI-RPMax-v1.1F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Rocinante-X-12B-v1BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Dans-SakuraKaze-V1.0.0-12bBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
MN-12B-Mag-Mell-R1F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
pixtral-12bBF1612.7B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Violet_Twilight-v0.2BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Crimson_Dawn-v0.2F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Magnum-Picaro-0.7-v2-12bBF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Mistral-Nemo-Prism-12BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Nera_Noctis-12BF1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±26.5%
Chronos-Gold-12B-1.0F1612.2B22.82 GiB5.63 GiB29.39 GiB0.37 GiB13±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.

Questions people ask

What AI models can a Radeon Pro W7800 run?
1949 of 2118 indexed open-weight models fit a Radeon Pro W7800 at 131,072 context with q4_0 KV cache, the largest being glm-4-9b-chat-1m at Q4_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon Pro W7800 actually have?
Its nameplate is 32 GB, but about 29.76 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon Pro W7800 fast for local AI?
Its memory bandwidth is 576 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.