AMD · workstation

Radeon Pro W7900 Dual Slot

Radeon Pro W7900 Dual Slot has 48 GB of VRAM at 864 GB/s — about 44.64 GiB usable after driver and compositor overhead. 1975 of 2118 indexed models fit at 128K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
48 GB
GDDR6
Bandwidth
864 GB/s
384-bit bus
Tensor FP16
—
dense
TDP
295 W
$3499 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1690vision language 181image 2audio asr 39audio tts 21video 16embedding 26

What fits at 128K context

largest quantization that fits, per model · 1975 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
IQuest-Coder-V1-40B-InstructI1-Q4_K_M39.8B22.38 GiB21.25 GiB44.63 GiB0.01 GiB13±26.5%
Kimi-Dev-72BUD-IQ1_M72.7B22.35 GiB21.25 GiB44.63 GiB0.01 GiB13±26.5%
Melody1437-27BQ5_K_M27.8B39.34 GiB4.25 GiB44.55 GiB0.09 GiB13±26.5%
Huihui-Qwen3-Coder-Next-abliteratedMoEIQ4_NL79.7B42.04 GiB1.59 GiB44.52 GiB0.12 GiB60±37%
Rombo-LLM-V3.0-Qwen-72bI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-72B-Instruct-abliteratedI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-72B-Instruct-abliterated-v2I1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
HuatuoGPT-o1-72BIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
MiroThinker-v1.0-72BI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
EVA-Qwen2.5-72B-v0.2IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-Math-72B-InstructIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-72B-InstructIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Malaysian-Qwen2.5-72B-InstructI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-72BI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
magnum-v4-72bI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
KAT-Dev-72B-ExpIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Homer-v1.0-Qwen2.5-72BIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Tower-Plus-72B-ultra-uncensored-hereticI1-IQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Qwen2.5-VL-72B-InstructIQ1_M73.4B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Chronos-Platinum-72BIQ1_M72.7B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
UI-TARS-72B-DPOIQ1_M73.4B22.11 GiB21.25 GiB44.39 GiB0.25 GiB13±26.5%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedIQ2_XS109B30.68 GiB12.75 GiB44.36 GiB0.28 GiB19±37%
Hypernova-60B-2605MoEI1-Q5_K_M58.7B41.29 GiB2.14 GiB44.32 GiB0.32 GiB46±37%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPQ5_K_M27.8B39.03 GiB4.25 GiB44.25 GiB0.39 GiB13±26.5%
Midnight-Miqu-70B-v1.5I1-Q2_K_S69.0B21.94 GiB21.25 GiB44.21 GiB0.43 GiB13±26.5%
solar-pro-preview-instructKV unresolvedQ8_022.1B21.91 GiB21.25 GiB44.12 GiB0.52 GiB13±26.5%
NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16MoEQ4_K_S75.4B43.15 GiB0.00 GiB44.12 GiB0.52 GiB102±37%
Hunyuan-A13B-InstructMoEIQ3_M80.4B34.72 GiB8.50 GiB44.11 GiB0.53 GiB13±26.5%
llm-surgery-dark-arts-gpt-oss-60b-96a12MoEI1-Q5_K_S60.9B41.56 GiB1.60 GiB44.06 GiB0.58 GiB34±37%
Qwen3.5-122B-A10B-hereticMoEI1-Q2_K123B41.52 GiB1.59 GiB44.04 GiB0.60 GiB55±37%
Gemma-4-31B-Isometry-RPQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Prosopon-31BQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Gemma-4-Novelist-Eclipse-31BQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Giftige-Blume-31B-v1-StyleSwapQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
G4-MeroMero-31B-StyleSwapQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Gemma-4-31B-StyleTune-heretic-araQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Pantheon-Reasoning-31B-1.1Q8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Gemma-4-31B-StyleTuneQ8_032.7B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
Barcenas-StyleTune-31B-FableQ8_032.1B31.79 GiB11.25 GiB44.02 GiB0.62 GiB13±26.5%
GLM-4.5-Air-REAP-82B-A12BMoEIQ2_S81.9B30.78 GiB12.22 GiB43.92 GiB0.72 GiB19±37%
Qwen3.5-122B-A10BMoEIQ2_M125B41.39 GiB1.59 GiB43.91 GiB0.73 GiB56±37%
CodeLlama-70b-Instruct-hfI1-IQ2_M69.0B21.64 GiB21.25 GiB43.91 GiB0.73 GiB13±26.5%
CodeLlama-70b-Python-hfI1-IQ2_M69.0B21.64 GiB21.25 GiB43.91 GiB0.73 GiB13±26.5%
Nous-Hermes-Llama2-70bI1-IQ2_M69.0B21.64 GiB21.25 GiB43.91 GiB0.73 GiB13±26.5%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB16.47 GiB43.90 GiB0.74 GiB16±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB16.47 GiB43.90 GiB0.74 GiB16±37%
Llama-2-7b-chat-hfQ5_K_M6.7B8.91 GiB34.00 GiB43.83 GiB0.81 GiB13±26.5%
Qwen3-Coder-NextMoEUD-IQ4_NL79.7B36.54 GiB6.38 GiB43.80 GiB0.84 GiB33±37%
Mistral-Small-4-119B-2603MoEUD-IQ3_S119B41.36 GiB1.49 GiB43.78 GiB0.86 GiB57±37%
Llama-4-Scout-17B-16E-Instruct-abliterated-v2MoEKV unresolvedI1-IQ2_S109B30.07 GiB12.75 GiB43.74 GiB0.90 GiB19±37%
Qwen3-48B-A4B-Savant-Commander-Distill-12X-Closed-Open-Heretic-UncensoredMoEQ8_033.6B33.27 GiB9.56 GiB43.74 GiB0.90 GiB19±37%
Yi-34B-200K-DARE-megamerge-v8Q6_K34.4B26.78 GiB15.94 GiB43.70 GiB0.94 GiB13±26.5%
Nous-Hermes-2-Yi-34BI1-Q6_K34.4B26.78 GiB15.94 GiB43.70 GiB0.94 GiB13±26.5%
Nous-Capybara-limarpv3-34BI1-Q6_K34.4B26.78 GiB15.94 GiB43.70 GiB0.94 GiB13±26.5%
c4ai-command-r-08-2024Q8_032.3B31.97 GiB10.63 GiB43.61 GiB1.03 GiB13±26.5%
Mixtral-8x22B-Instruct-v0.1MoEIQ1_S141B27.61 GiB14.88 GiB43.45 GiB1.19 GiB14±37%
Mixtral-8x22B-v0.1MoEIQ1_S141B27.61 GiB14.88 GiB43.44 GiB1.20 GiB14±37%
CallerQ6_K_L32.8B25.39 GiB17.00 GiB43.39 GiB1.25 GiB13±26.5%
Dumpling-Qwen2.5-32BQ6_K_L32.8B25.39 GiB17.00 GiB43.39 GiB1.25 GiB13±26.5%
OREAL-32BQ6_K_L32.8B25.39 GiB17.00 GiB43.39 GiB1.25 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 W7900 Dual Slot run?
1975 of 2118 indexed open-weight models fit a Radeon Pro W7900 Dual Slot at 131,072 context with q8_0 KV cache, the largest being IQuest-Coder-V1-40B-Instruct at I1-Q4_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon Pro W7900 Dual Slot actually have?
Its nameplate is 48 GB, but about 44.64 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon Pro W7900 Dual Slot fast for local AI?
Its memory bandwidth is 864 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.