Hardware / AMD /Instinct MI300X

AMD · GPU

Instinct MI300X

Runs 176 models fully in memory, 18 more with CPU offload. 2 of those sit close enough to 192 GB that our size estimate is what decides them.

Memory

192 GB

Bandwidth

5300 GB/s

Runs

176

01

Specifications

Memory

192 GB

Decides which models fit at all.

Memory bandwidth

5300 GB/s

How fast it can read the model. Sets the speed of the reply.

Compute

1307 TFLOPS

Trillions of half-precision sums a second. Sets how fast it reads a long prompt, not how fast it replies.

Power draw

750 W

At full load, for sizing a power supply.

MSRP

$20,000

What the maker asked for it new. We hold no used or street prices, so this is not what one costs today.

Released

Dec 6, 2023

Architecture

CDNA 3

02

What it runs

Best quant per model. Text models first, then speech, then what does not fit.

What is quantisation? →

ModelParamsContextSpeedVerdict

Nex-N2-Mini35.1B262K138 tok/sFits in memory Granite 4.1 8B8.8B131K646 tok/sFits in memory Qwen3.6 35B A3B36B262K919 tok/sFits in memory Gemma 4 26B A4B25.8B262K689 tok/sFits in memory Gemma 4 31B31.3B131K182 tok/sFits in memory MiniMax M2.7229B131K483 tok/sFits in memory Qwen3.5-9B9.7B262K499 tok/sFits in memory Qwen3.5-27B27.8B262K204 tok/sFits in memory Qwen3.5-122B-A10B125B262K276 tok/sFits in memory GLM 4.7 Flash31.2B131K155 tok/sFits in memory Nemotron 3 Nano 30B A3B31.6B262K1372 tok/sFits in memory Ministral 3 14B 251213.9B262K233 tok/sFits in memory Ministral 3 8B 25128.9B262K544 tok/sFits in memory Ministral 3 3B 25123.8B131K853 tok/sFits in memory Olmo 3 32B Think32.2B66K150 tok/sFits in memory gpt-oss-safeguard-20b21.5B131K1144 tok/sFits in memory Qwen3 VL 32B Instruct33.4B131K145 tok/sFits in memory Granite 4.0 Micro3.2B66K1514 tok/sFits in memory Qwen3 VL 8B Thinking8.8B131K646 tok/sFits in memory Qwen3 VL 30B A3B Thinking31.1B262K919 tok/sFits in memory Cydonia 24B V4.123.6B131K241 tok/sFits in memory Qwen3 VL 235B A22B Thinking236B131K220 tok/sFits in memory Qwen3 Next 80B A3B Thinking81.3B262K919 tok/sFits in memory gpt-oss-120b120B131K540 tok/sFits in memory GLM 4.5 Air111B131K403 tok/sFits in memory UI-TARS 7B8.3B66K391 tok/sFits in memory Uncensored24B66K202 tok/sFits in memory Hunyuan A13B Instruct80.4B131K212 tok/sFits in memory Mistral Small 3.2 24B24B131K237 tok/sFits in memory Llama Guard 4 12B12B262K404 tok/sFits in memory Qwen3 14B14.8B131K219 tok/sFits in memory Gemma 3 12B12.2B131K466 tok/sFits in memory Gemma 3 27B27.4B262K207 tok/sFits in memory Rocinante 12B12.2B66K266 tok/sFits in memory Llama 3.2 3B Instruct3.2B131K1013 tok/sFits in memory Mistral Nemo12.2B131K266 tok/sFits in memory Mixtral 8x22B Instruct141B66K124 tok/sFits in memory Laguna XS 2.133.4B262K123 tok/sFits in memory Qwen3.6 27B27.8B262K174 tok/sFits in memory Qwen3.5-35B-A3B36B262K1610 tok/sFits in memory MiniMax M2.5229B131K483 tok/sFits in memory MiniMax M2.1229B131K483 tok/sFits in memory Qwen3 VL 8B Instruct8.8B262K368 tok/sFits in memory Qwen3 VL 30B A3B Instruct31.1B262K1372 tok/sFits in memory Qwen3 VL 235B A22B Instruct236B131K220 tok/sFits in memory Qwen3 Next 80B A3B Instruct81.3B262K1372 tok/sFits in memory Qwen3 30B A3B Thinking 250730.5B66K1372 tok/sFits in memory Qwen3 Coder 30B A3B Instruct30.5B262K1372 tok/sFits in memory Qwen3 30B A3B Instruct 250730.5B262K919 tok/sFits in memory Qwen3 235B A22B Thinking 2507235B131K220 tok/sFits in memory Qwen3 235B A22B Instruct 2507235B131K220 tok/sFits in memory Qwen3 8B8.2B131K395 tok/sFits in memory Qwen3 32B32.8B131K148 tok/sFits in memory Mistral Small 3.1 24B24B66K237 tok/sFits in memory Gemma 3 4B4.3B131K1126 tok/sFits in memory Reka Flash 320.9B66K232 tok/sFits in memory Llama 3.1 8B Instruct8B131K605 tok/sFits in memory MiniMax M2229B131K483 tok/sFits in memory Muse Glimmer 30B29.8B131K163 tok/sFits in memory Nemotron 3.5 Lightning31.6B262K153 tok/sFits in memory Qwen3.8 27B27.8B262K174 tok/sFits in memory Granite 4.2 8B8.8B131K646 tok/sFits in memory Nemotron 3.5 Content Safety4.3B131K1126 tok/sFits in memory Ternary Bonsai 2 27B27B262K179 tok/sFits in memory Nex-N2.5-Mini35.1B262K138 tok/sFits in memory Gemma 3n 4B7.8B33K621 tok/sFits in memory Skyfall 36B V236.9B33K154 tok/sFits in memory Qwen2.5 Coder 32B Instruct32.8B33K148 tok/sFits in memory UnslopNemo 12B12.2B33K397 tok/sFits in memory Llama 3.2 1B Instruct1.2B33K4735 tok/sFits in memory Mistral Small 323.6B33K137 tok/sFits in memory Qwen2.5 7B Instruct7.6B33K427 tok/sFits in memory Reka Edge7.1B16K457 tok/sFits in memory Voxtral Small 24B 250724.3B16K199 tok/sFits in memory Phi 414.7B16K387 tok/sFits in memory Llama 3 8B Lunaris8B8K405 tok/sFits in memory MythoMax 13B13B8K437 tok/sFits in memory Gemma 2 27B27.2B8K209 tok/sFits in memory Hy-MT2-1.8B2B8K2059 tok/sFits in memory Hy-MT2-30B-A3B30.1B8K1610 tok/sFits in memory Hy-MT2-7B8B8K515 tok/sFits in memory Hermes 4 70B70.6B131K81 tok/sFits in memory Llama 3.3 Euryale 70B70.6B131K81 tok/sFits in memory Llama 3.1 Euryale 70B v2.270.6B131K81 tok/sFits in memory Hermes 3 70B Instruct70.6B131K81 tok/sFits in memory Llama 3.1 70B Instruct70.6B131K81 tok/sFits in memory Llama 3.3 70B Instruct70.6B131K81 tok/sFits in memory Qwen2.5 VL 72B Instruct73.4B66K77 tok/sFits in memory ReMM SLERP 13B13B4K373 tok/sFits in memory DeepSeek V4 Flash291B4K131 tok/sFits in memoryest Qwen2.5 72B Instruct72.7B33K78 tok/sFits in memory Qwen3 Coder Next79.7B262K61 tok/sFits in memory Magnum v4 72B72.7B16K78 tok/sFits in memory Command A111B131K51 tok/sFits in memory Llama 4 Scout109B262K52 tok/sFits in memory Mistral Medium 3.5128B262K45 tok/sFits in memory Ling-2.6-flash107B131K45 tok/sFits in memory Devstral 2 2512125B262K46 tok/sFits in memory GLM 4.6V108B131K45 tok/sFits in memory GLM 4.5V108B66K45 tok/sFits in memory R1 Distill Llama 70B70.6B8K81 tok/sFits in memory Laguna S 2.1118B262K41 tok/sFits in memory Mistral Small 4119B131K41 tok/sFits in memory Nemotron 3 Super124B262K39 tok/sFits in memory Ling-3.0-flash128B131K38 tok/sFits in memory Step 3.7 Flash201B131K24 tok/sFits in memory Step 3.5 Flash199B131K24 tok/sFits in memory Laguna M.1226B131K21 tok/sFits in memory Inkling Small266B66K18 tok/sFits in memoryest

Speech models it runs67 modelsTranscription and speech models that fit in memory. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.

Niagara 19m Batch.en20M284124 tok/sFits in memory Niagara 38m Batch.en38M85325 tok/sFits in memory Higgs Audio v3 8b STT v28.9B639 tok/sFits in memory Higgs Audio v3 STT2.7B1201 tok/sFits in memory Cohere Transcribe 03 20262.1B1544 tok/sFits in memory Distil Large v3.50.8B4053 tok/sFits in memory Lite Whisper Large v3 Acc1.4B3460 tok/sFits in memory Owsm CTC v3.1 1B1.1B5166 tok/sFits in memory Owsm CTC v3.2 ft 1B1B3242 tok/sFits in memory Owsm CTC v4 1B1B5683 tok/sFits in memory Hubert Large Ls960 ft0.3B10132 tok/sFits in memory MMS 1b All1B4844 tok/sFits in memory Wav2vec2 Large 960h Lv60 Self0.3B10132 tok/sFits in memory Hojo ASR V15.2B935 tok/sFits in memory Granite 4.0 1b Speech2.3B1410 tok/sFits in memory Granite Speech 3.3 2b3B1894 tok/sFits in memory Granite Speech 3.3 8b8.6B661 tok/sFits in memory Granite Speech 4.1 2b2.3B2471 tok/sFits in memory Granite Speech 4.1 2b NAR2.3B2471 tok/sFits in memory STT 2.6b en2.6B2186 tok/sFits in memory Phi 4 Multimodal Instruct5.6B131K1015 tok/sFits in memory VibeVoice ASR HF8.3B391 tok/sFits in memory Voxtral Mini 3B 25074.7B690 tok/sFits in memory Voxtral Mini 4B Realtime 26024.4B737 tok/sFits in memory Canary 180m Flash0.2B26909 tok/sFits in memory Canary 1b1B3242 tok/sFits in memory Canary 1b Flash0.8B4053 tok/sFits in memory Canary 1b v21B3242 tok/sFits in memory Canary Qwen 2.5b2.6B1863 tok/sFits in memory Nemotron 3.5 ASR Streaming 0.6b0.6B5404 tok/sFits in memory Nemotron Speech Streaming en 0.6b0.6B8073 tok/sFits in memory Parakeet CTC 0.6b0.6B9471 tok/sFits in memory Parakeet CTC 1.1b1.1B5166 tok/sFits in memory Parakeet RNNT 0.6b0.6B5404 tok/sFits in memory Parakeet RNNT 1.1b1.1B5166 tok/sFits in memory Parakeet TDT CTC 110m0.1B29476 tok/sFits in memory Parakeet TDT 0.6b v20.6B8073 tok/sFits in memory Parakeet TDT 0.6b v30.6B9471 tok/sFits in memory Parakeet TDT 1.1b1.1B5166 tok/sFits in memory Nyra Health CrisperWhisper1.6B3552 tok/sFits in memory Whisper Large v31.5B3788 tok/sFits in memory Whisper Large v3 Turbo0.8B4053 tok/sFits in memory MOSS Transcribe Diarize0.9B5382 tok/sFits in memory Qwen3 ASR 0.6B0.9B3603 tok/sFits in memory Qwen3 ASR 0.6B HF0.8B6055 tok/sFits in memory Qwen3 ASR 1.7B2.3B1410 tok/sFits in memory Qwen3 ASR 1.7B HF2B2422 tok/sFits in memory Zipformer cr CTC Transducer XL 290M0.3B11181 tok/sFits in memory Zipformer Transducer XL 290M0.3B19595 tok/sFits in memory ASR Conformer Largescaleasr0.5B10091 tok/sFits in memory GLM ASR Nano 25122.3B2471 tok/sFits in memory Granite Speech 5.0 470m Turboctc0.5B9687 tok/sFits in memory Granite Speech 5.0 470m Turboctc nc0.5B11365 tok/sFits in memory Fun ASR Nano 2512 HF0.8B7103 tok/sFits in memory Gemma 4 12B it12B270 tok/sFits in memory Gemma 4 E2B it5.1B1114 tok/sFits in memory Gemma 4 E4B it8B131K710 tok/sFits in memory Niagara 9m Batch.en9M360261 tok/sFits in memory Niagara 84m Batch.en84M38599 tok/sFits in memory ARK ASR 0.6B1.3B33K3726 tok/sFits in memory ARK ASR 3B4.1B33K1386 tok/sFits in memory Audio8 ASR 0.1B0.3B33K10808 tok/sFits in memory MOSS Transcribe Preview 2B2.4B33K1351 tok/sFits in memory Moonshine Streaming Medium0.3B4K10808 tok/sFits in memory Moonshine Streaming Small0.1B4K32424 tok/sFits in memory Moonshine Streaming Tiny0B4Kno speed estimateFits in memory Moonshine Tiny0B194no speed estimateFits in memory

Runs only by spilling into system memory18 modelsThese load, but part of the weights sits in ordinary system memory, which is far slower than the chip.

Hy3299B—no speed estimateSpills to system RAM MiniMax M3427B—no speed estimateSpills to system RAMest Trinity Large Thinking399B—no speed estimateSpills to system RAMest Hermes 4 405B406B—no speed estimateSpills to system RAMest Jamba Large 1.7399B—no speed estimateSpills to system RAMest ERNIE 4.5 VL 424B A47B424B—no speed estimateSpills to system RAMest Hermes 3 405B Instruct406B—no speed estimateSpills to system RAMest Nex-N2-Pro397B—no speed estimateSpills to system RAM Hy3 preview299B—no speed estimateSpills to system RAM MiMo-V2.5311B—no speed estimateSpills to system RAM Qwen3.5 397B A17B403B—no speed estimateSpills to system RAMest GLM 4.7358B—no speed estimateSpills to system RAMest GLM 4.6357B—no speed estimateSpills to system RAMest GLM 4.5358B—no speed estimateSpills to system RAM Llama 4 Maverick402B—no speed estimateSpills to system RAMest DeepSeek V4 Flash Vision Exp305B—no speed estimateSpills to system RAM GLM 5.3 Flash321B—no speed estimateSpills to system RAM Nex-N2.5-Pro397B—no speed estimateSpills to system RAM

Too large for this device30 modelsThe weights do not fit even with part of them offloaded to system memory. Closest calls first.

Qwen3 Coder 480B A35B480B—no speed estimateToo largeest MiniMax M1456B—no speed estimateToo large R1 0528685B—no speed estimateToo large DeepSeek V3 0324685B—no speed estimateToo large GLM 5.2753B—no speed estimateToo large MiniMax-01456B—no speed estimateToo large DeepSeek V3685B—no speed estimateToo large GLM 5.1754B—no speed estimateToo large GLM 5754B—no speed estimateToo large Inkling952B—no speed estimateToo large Nemotron 3 Ultra561B—no speed estimateToo large MiMo-V2.5-Pro1T—no speed estimateToo large Ring-2.6-1T1T—no speed estimateToo large DeepSeek V3.1 Terminus685B—no speed estimateToo large DeepSeek V3.1685B—no speed estimateToo large DeepSeek V3.2685B—no speed estimateToo large DeepSeek V3.2 Exp685B—no speed estimateToo large Ling-2.6-1T1T—no speed estimateToo large Kimi K2 Thinking1.1T—no speed estimateToo large Kimi K2.7 Code1.1T—no speed estimateToo large Kimi K2.51.1T—no speed estimateToo large DeepSeek V4.1 Flash763B—no speed estimateToo large Hy4 preview780B—no speed estimateToo large DeepSeek V4 Pro1.6T—no speed estimateToo large Kimi K2 09051T—no speed estimateToo large Kimi K2 07111T—no speed estimateToo large Kimi K2.61.1T—no speed estimateToo large LongCat 2.01.8T—no speed estimateToo large Kimi K32.8T—no speed estimateToo large Qwen3.8 2.4T A95B2.4T—no speed estimateToo large

224 models have a verdict on this device: 176 fit in memory, 18 spill into system memory and 30 are too large. Every one of them is on this page.

Something wrong on this page? Tell us