AMD · workstation

Radeon Pro W7700

Radeon Pro W7700 has 16 GB of VRAM at 576 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1849 of 2118 indexed models fit at 8K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
GDDR6
Bandwidth
576 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
190 W
$999 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1585vision language 161audio asr 39video 15embedding 26image 2audio tts 21

What fits at 8K context

largest quantization that fits, per model · 1849 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
gemma-2-27b-itIQ3_M27.2B11.60 GiB2.25 GiB14.88 GiB0.00 GiB26±26.5%
magnum-v4-27bIQ3_M27.2B11.60 GiB2.25 GiB14.88 GiB0.00 GiB26±26.5%
Nemotron-Mini-4B-InstructQ8_04.2B12.96 GiB1.00 GiB14.88 GiB0.00 GiB26±26.5%
Skyfall-31B-v4.2-hereticI1-IQ3_XS31.4B12.17 GiB1.69 GiB14.88 GiB0.00 GiB26±26.5%
Skyfall-31B-v4.2I1-IQ3_XS31.4B12.17 GiB1.69 GiB14.88 GiB0.00 GiB26±26.5%
GLM-4-32B-0414-Korean-CultureI1-IQ3_S32.6B13.40 GiB0.48 GiB14.86 GiB0.02 GiB26±26.5%
Qwen3-Coder-REAP-25B-A3BMoEIQ4_NL24.9B13.22 GiB0.75 GiB14.86 GiB0.02 GiB76±37%
Seed-OSS-36B-InstructUD-IQ2_M36.2B11.86 GiB2.00 GiB14.85 GiB0.03 GiB26±26.5%
GLM-Z1-32B-0414Q3_K_S32.6B13.38 GiB0.48 GiB14.85 GiB0.03 GiB26±26.5%
GLM-4-32B-0414Q3_K_S32.6B13.38 GiB0.48 GiB14.85 GiB0.03 GiB26±26.5%
TildeOpen-30B-Instruct-LVI1-IQ3_XS30.7B11.99 GiB1.88 GiB14.85 GiB0.03 GiB26±26.5%
grug-27bQ3_K_M27.4B13.38 GiB0.50 GiB14.84 GiB0.04 GiB26±26.5%
Carnice-V2-27bQ3_K_M27.4B13.38 GiB0.50 GiB14.84 GiB0.04 GiB26±26.5%
Fara1.5-27BQ3_K_M27.4B13.38 GiB0.50 GiB14.84 GiB0.04 GiB26±26.5%
granite-20b-code-instruct-8kQ5_K_L20.1B13.86 GiB0.00 GiB14.84 GiB0.04 GiB26±26.5%
Devstral-Small-2-24B-Instruct-2512Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Voxtral-Small-24B-2507Q4_024.3B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Transformed-Journey-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magistry-24B-v1.1I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mergedonia-AETHER-24B-v1aI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mergedonia-AETHER-24B-v1bI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Slimaki-Tavern-24B-v1.3I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Maginum-Cydoms-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Maginum-Cydoms-24B-absolute-heresyI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Dolphin3.0-R1-Mistral-24BQ4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Dolphin3.0-Mistral-24BQ4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-hereticI1-Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Huihui-Mistral-Small-3.2-24B-Instruct-2506-abliterated-llamacppfixedI1-Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Dans-PersonalityEngine-V1.2.0-24bI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mistral-Small-3_2-24B-Instruct-2506-antislop.v2I1-Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia_VistralQ4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mistral-Small-3.2-24B-Instruct-2506Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Dans-PersonalityEngine-V1.3.0-24bI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Devstral-Small-2507Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Goetia-24B-v1.1I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Devstral-Small-2505Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
MS3.2-PaintedFantasy-v3-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
RP-Spectrum-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
MS3.2-PaintedFantasy-v4.1-24B-ultra-uncensored-heretic-v2I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magidonia-24B-v4.3-heretic-v1.2I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magidonia-24B-v4.3-absolute-heresyI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
MagiSeek-Pro-V1I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magistral-Small-2509Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magistral-Small-2507Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cogidonia-v2-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magidonia-24B-v4.3I1-Q4_0—12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Precog-24B-v1I1-Q4_0—12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
experiment024bI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Magidonia-24B-v4.2.0Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Berthier-Mistral-Military-24BI1-Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
MS-2501-DPE-QwQify-v0.1-24BQ4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Mistral-Small-3.2-24B-Instruct-2506-llamacppfixedI1-Q4_024.0B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia-24B-v4.3-absolute-heresyI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia-24B-v4.3-heretic-v2I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia-24B-v4.3-hereticI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia-24B-v4.3-heretic-v4I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Cydonia-24B-v4.2.0I1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Journeys-End-24BI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
sarvam-mQ4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±26.5%
Dolphin-Mistral-GLM-4.7-Flash-24B-Venice-Edition-Thinking-UncensoredI1-Q4_023.6B12.57 GiB1.25 GiB14.84 GiB0.04 GiB26±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 W7700 run?
1849 of 2118 indexed open-weight models fit a Radeon Pro W7700 at 8,192 context with f16 KV cache, the largest being gemma-2-27b-it at IQ3_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon Pro W7700 actually have?
Its nameplate is 16 GB, but about 14.88 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon Pro W7700 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.