IngiaFungua pochi

M3 Max yenye GB 48 inaweza kuendesha nini?

Modeli huria 38 kati ya 73 zinaenda vizuri juu yake. Kwa chaguo-msingi GPU inaweza kutumia takriban GB 36 kati ya GB 48 zake, na husoma kumbukumbu kwa 400 GB/s, jambo linaloamua kasi yake ya kuandika. Inapatikana katika: MacBook Pro (14-inch, M3 Pro or M3 Max, Nov 2023), MacBook Pro (16-inch, Nov 2023).

Yenye akili zaidi inayoenda vizuriQwen3.8 27B4-bit · 15.2 GB · 12–21 tok/sKubwa zaidi inayoenda vizuriQwen3.6 35B A3B6-bit · 28.4 GB · 75–133 tok/sKwa kuandika codeQwen3 Coder 30B A3B Instruct8-bit · 32.4 GB · 51–91 tok/s

Zinaenda vizuri 41

Inatosha na kubakiza nafasi ya context, kwa tokeni 10 au zaidi kwa sekunde.

ModeliToleoKumbukumbuTokeni/s
Qwen3.8 27B27B · Akili 33.74-bit15.2 GB12–21
Qwen3.6 35B A3B35B · 3B zinazotumika · Akili 18.26-bit28.4 GB75–133
Muse Glimmer 30B29.6B · Akili 17.54-bit16.7 GB11–19
Gemma 4 31B30.7B · Akili 14.74-bit17.3 GB10–18
Nemotron 3.5 Lightning 30B A3B30B · 3B zinazotumika · Akili 12.98-bit31.9 GB56–100
Granite 4.2 30B30B · Akili 12.74-bit16.9 GB11–19
MiniCPM5 2B2.52B · Akili 12.58-bit2.7 GB67–119
Qwen3.5 9B9B · Akili 11.28-bit9.6 GB19–33
Granite 4.2 8B8B · Akili 11.18-bit8.5 GB21–38
Granite 4.2 3B3B · Akili 9.18-bit3.2 GB56–100
gpt-oss-20b21B · 3.6B zinazotumika · Akili 94.25-bit asili13.8 GB95–168
Nemotron 3 Nano 30B A3B30B · 3.5B zinazotumika · Akili 8.98-bit31.9 GB49–86
Gemma 4 E4B8B · Akili 8.98-bit8.5 GB38–67
Ministral 3 8B8.9B · Akili 5.58-bit9.5 GB19–34
Nemotron 3 Nano 4B3.97B · Akili 4.88-bit4.2 GB43–76
Qwen3 Coder 30B A3B Instruct30.5B · 3.3B zinazotumika8-bit32.4 GB51–91
GLM-4.7 Flash30B · 3B zinazotumika8-bit31.9 GB56–100
Qwen3.6 27B27B4-bit15.2 GB12–21
Gemma 4 26B A4B25.2B · 3.8B zinazotumika8-bit26.8 GB45–80
Devstral Small 2 (24B)24B4-bit13.5 GB13–24
Magistral Small 1.2 (2509)24B4-bit13.5 GB13–24
Mistral Small 3.2 (24B)24B4-bit13.5 GB13–24
LFM2.5 24B A2B24B · 2B zinazotumika8-bit25.5 GB86–152
Phi-4 Reasoning Vision 15B15B8-bit15.9 GB11–20
Phi-4 (14B)14B8-bit14.9 GB12–21
Gemma 4 12B Unified12B8-bit12.7 GB14–25
Qwen3 8B8.2B8-bit8.7 GB21–37
Llama 3.1 8B Instruct8B8-bit8.5 GB21–38
Gemma 4 E2B5.1B8-bit5.4 GB75–133
Qwen3.5 4B4B8-bit4.3 GB42–74
Phi-4 mini instruct3.8B8-bit4 GB45–80
Llama 3.2 3B Instruct3.21B8-bit3.4 GB53–94
SmolLM3 3B3B8-bit3.2 GB56–100
LFM2.5 2.6B2.69B8-bit2.9 GB62–110
Qwen3.5 2B2B8-bit2.1 GB86–152
Llama 3.2 1B Instruct1.23B8-bit1.3 GB138–246
Qwen3.5 0.8B0.8B8-bit0.9 GB200–356
Qwen3 0.6B0.6B8-bit0.6 GB300–533
Nemotron 3.5 Bolt 74B A9B74B · 9B zinazotumika · Akili 21.43-bit ubora wa chini32.4 GB46–82
Qwen3 Coder Next80B · 3B zinazotumika · Akili 9.22-bit ubora wa chini25 GB200–356
GLM-4.5 Air106B · 12B zinazotumika2-bit ubora wa chini33.1 GB47–84

Zinaenda, lakini polepole 1

Inatosha, lakini chini ya tokeni 10 kwa sekunde: sawa kwa kazi za kundi, polepole kwa mazungumzo.

ModeliToleoKumbukumbuTokeni/s
Llama 3.3 70B Instruct70B3-bit ubora wa chini30.6 GB6–10

Zinatosha kwa shida 8

Inahitaji karibu kumbukumbu yote: ongeza kikomo cha kumbukumbu ya GPU na weka context fupi.

ModeliToleoKumbukumbuTokeni/s
Qwen3.5 122B A10B122B · 10B zinazotumika · Akili 15.62-bit ubora wa chini38.1 GB58–103
Mistral Medium 3.5 (128B)128B · Akili 14.22-bit ubora wa chini40 GB5–8
gpt-oss-120b117B · 5.1B zinazotumika · Akili 11.62-bit ubora wa chini36.6 GB113–200
Mistral Small 4 (119B)119B · 6.5B zinazotumika · Akili 11.32-bit ubora wa chini37.2 GB90–160
Llama 4 Scout 17B-16E Instruct109B · 17B zinazotumika · Akili 6.52-bit ubora wa chini34.1 GB34–60
Devstral 2 (123B)123B2-bit ubora wa chini38.4 GB5–8
Ornith 1.5 122B122B · 10B zinazotumika2-bit ubora wa chini38.1 GB58–103
Nemotron 3 Super 120B A12B120B · 12B zinazotumika2-bit ubora wa chini37.5 GB47–84

Kubwa mno 23

Inahitaji kumbukumbu zaidi ya GB 48, hata kwa biti 2.

Mac nyingine

Kila Mac yenye GB 48, ya chipu yoyote.

M3 Max yenye kumbukumbu zaidi au kidogo:

Chipu jirani:

Ana kwa ana:

Ilinganishe na Mac au GPU yoyote · M3 Max kwa AI ya ndani ya kifaa · Mac bora kwa LLM za ndani · Kila Mac

Maswali

LLM ipi ni bora kwa M3 Max yenye GB 48?
Miongoni mwa modeli zenye Intelligence Index iliyochapishwa, Qwen3.8 27B ina alama ya juu zaidi kati ya zile zinazoenda vizuri hapa: 4-bit, takriban tokeni 12–21 kwa sekunde. Kubwa zaidi inayoenda vizuri ni Qwen3.6 35B A3B.
Je, M3 Max yenye GB 48 inaweza kuendesha Llama 3.3 70B?
Ndiyo, lakini polepole: 3-bit (ubora wa chini), kwa takriban tokeni 6–10 kwa sekunde.
gpt-oss-20b ina kasi gani kwenye M3 Max?
Takriban tokeni 95–168 kwa sekunde (4.25-bit asili), kutokana na bandwidth ya kumbukumbu ya chipu ya 400 GB/s.

Ziendeshe kwa MLX Harness (bure, open source) au mlx-lm ya Apple ndani ya Terminal: hatua kwa hatua.

Jinsi inavyokokotolewa

Kumbukumbu: vigezo vya modeli kwa biti 8, 6, 4, 3 au 2 kwa kila weight pamoja na group scales za MLX (nusu biti kwa kila weight), au muundo wake asili, pamoja na nafasi ya context juu yake. Kwa chaguo-msingi macOS huiruhusu GPU kutumia takriban 75% ya unified memory. Kasi: bandwidth ya kumbukumbu ikigawanywa kwa baiti zinazosomwa kwa kila tokeni ndiyo kikomo cha juu kwa mazungumzo moja; katika benchmarks za umma za llama.cpp kwenye Apple Silicon uendeshaji halisi hufikia takriban 45% hadi 80% yake, na huo ndio upana unaoonyeshwa. Haya ni makadirio, si vipimo: pima Mac yako mwenyewe kwenye MLX Bench. Takwimu za modeli ni kutoka kwa kadi rasmi ya kila modeli, takwimu za chipu ni kutoka Tech Specs za Apple. Imekaguliwa 2026-10-19. Data ileile ya mlx.app/compare.

M3 Max yenye GB 48 inaendesha LLM zipi? Modeli na kasi