็™ป้Œฒใ—ใฆๆ‹›ๅพ…ใƒชใƒณใ‚ฏใ‚’ๅ…ฑๆœ‰ใ™ใ‚‹ใจใ€ๅ‹•็”ปๅ†็”Ÿๅ ฑ้…ฌใจ็ดนไป‹ๅ ฑ้…ฌใ‚’็ฒๅพ—ใงใใพใ™ใ€‚

David Hendrickson
@TeksEdge
CEO & Founder | PhD | Startup Advisor | @Columbia | Author Generative Software Engineering | ๐Ÿ”” Follow for AI & Vibe Coding Tips ๐Ÿ‘‡
ๅ‚ๅŠ  July 2023
551 ใƒ•ใ‚ฉใƒญใƒผไธญ    12.3K ใƒ•ใ‚กใƒณ
This iPad Pro + Thunderbolt R9700 is running Qwen3.8-27B at 158.5 tps. Yesโ€ฆ an iPad. The trick ๐ŸŽฎ 32GB Radeon AI PRO R9700 ๐Ÿ”Œ Thunderbolt eGPU enclosure ๐Ÿ“ฑ iPad Pro running LemonSeed Studio ๐Ÿง  Qwen3.8-27B Q4 ๐Ÿ“š 131K context And the speed progression is kind of ridiculous ๐ŸŒ baseline โ†’ 31.2 tps โšก MTP3 โ†’ 108.7 tps ๐Ÿš€ DFlash2 โ†’ 158.5 tps Prefill: ๐Ÿ“ฅ 4K โ†’ 1,636 tps ๐Ÿ“ฅ 32K โ†’ 1,395 tps LemonSeed actually runs the AMD GPU on iPadOS. It embeds the upstream Linux amdgpu + amdkfd driver as an iPad PCIDriverKit extension, then runs its inference engine on the R9700. Same engine for ๐Ÿง Linux | ๐ŸŽ macOS | ๐Ÿ“ฑ iPadOS Same R9700 on macOS reaches 159.4 tps, so the iPadOS result is basically right there with the Mac. โš ๏ธ Obviously, the iPad's own GPU isn't doing the 159 tps. The iPad is hosting a 32GB desktop AMD GPU over Thunderbolt. ๐Ÿ”— Link in ALT.
ใ‚‚ใฃใจ่ฆ‹ใ‚‹