{"id":174019,"date":"2026-07-10T17:06:02","date_gmt":"2026-07-10T10:06:02","guid":{"rendered":"https:\/\/phongvu.vn\/cong-nghe\/?p=174019"},"modified":"2026-07-10T17:06:06","modified_gmt":"2026-07-10T10:06:06","slug":"chip-may-tinh-nhanh-gap-478-lan-gpu-nvidia","status":"publish","type":"post","link":"https:\/\/phongvu.vn\/cong-nghe\/chip-may-tinh-nhanh-gap-478-lan-gpu-nvidia\/","title":{"rendered":"Chip m\u00e1y t\u00ednh th\u1ebf h\u1ec7 m\u1edbi g\u00e2y s\u1ed1c v\u1edbi hi\u1ec7u n\u0103ng nhanh g\u1ea5p 478 l\u1ea7n GPU Nvidia"},"content":{"rendered":"\n

PV Tech News \u2013 Nh\u00f3m nghi\u00ean c\u1ee9u Trung Qu\u1ed1c v\u1eeba c\u00f4ng b\u1ed1 chip nh\u1edb ti\u1ebfn tr\u00ecnh 40 nanomet c\u00f3 kh\u1ea3 n\u0103ng m\u00f4 ph\u1ecfng c\u1ea5u tr\u00fac b\u1ed9 n\u00e3o theo th\u1eddi gian th\u1ef1c, \u0111\u1ea1t t\u1ed1c \u0111\u1ed9 x\u1eed l\u00fd nhanh h\u01a1n GPU Nvidia A100 \u0111\u1ebfn 478 l\u1ea7n.<\/strong><\/p>\n\n\n\n

C\u00f4ng b\u1ed1 tr\u00ean t\u1ea1p ch\u00ed Science<\/em> ng\u00e0y 2\/6, nh\u00f3m nghi\u00ean c\u1ee9u thu\u1ed9c \u0110\u1ea1i h\u1ecdc B\u1eafc Kinh v\u00e0 Vi\u1ec7n H\u00e0n l\u00e2m Khoa h\u1ecdc Trung Qu\u1ed1c gi\u1edbi thi\u1ec7u m\u1ed9t lo\u1ea1i chip nh\u1edb t\u00edch h\u1ee3p m\u1ea1ng n\u01a1-ron nh\u00e2n t\u1ea1o ngay tr\u00ean ph\u1ea7n c\u1ee9ng. \u0110i\u1ec3m \u0111\u00e1ng ch\u00fa \u00fd n\u1eb1m \u1edf kh\u1ea3 n\u0103ng t\u00e1i t\u1ea1o b\u1ec1 m\u1eb7t g\u1ea5p kh\u00fac c\u1ee7a v\u1ecf n\u00e3o, c\u1ea5u tr\u00fac v\u1ed1n c\u1ef1c k\u1ef3 ph\u1ee9c t\u1ea1p do ch\u1ee9a h\u00e0ng t\u1ef7 neuron ch\u1ec9 trong ch\u01b0a \u0111\u1ea7y n\u1eeda gi\u00e2y. V\u1edbi c\u00e1c h\u1ec7 th\u1ed1ng m\u00e1y t\u00ednh truy\u1ec1n th\u1ed1ng d\u00f9ng GPU, t\u00e1c v\u1ee5 t\u01b0\u01a1ng t\u1ef1 th\u01b0\u1eddng \u0111\u00f2i h\u1ecfi thi\u1ebft b\u1ecb c\u1ed3ng k\u1ec1nh v\u00e0 th\u1eddi gian t\u00ednh to\u00e1n k\u00e9o d\u00e0i h\u01a1n nhi\u1ec1u.<\/p>\n\n\n\n

B\u00ed quy\u1ebft gi\u00fap con chip v\u01b0\u1ee3t tr\u1ed9i n\u1eb1m \u1edf ki\u1ebfn tr\u00fac t\u00ednh to\u00e1n trong b\u1ed9 nh\u1edb (computing-in-memory<\/em>). Thay v\u00ec t\u00e1ch ri\u00eang b\u1ed9 nh\u1edb v\u00e0 b\u1ed9 x\u1eed l\u00fd nh\u01b0 c\u1ea5u tr\u00fac m\u00e1y t\u00ednh ph\u1ed5 bi\u1ebfn hi\u1ec7n nay, v\u1ed1n t\u1ea1o ra \u0111\u1ed9 tr\u1ec5 do d\u1eef li\u1ec7u ph\u1ea3i di chuy\u1ec3n qua l\u1ea1i, chip m\u1edbi l\u01b0u tr\u1eef v\u00e0 x\u1eed l\u00fd d\u1eef li\u1ec7u tr\u00ean c\u00f9ng m\u1ed9t m\u1ea3ng b\u1ed9 nh\u1edb. Nh\u00f3m nghi\u00ean c\u1ee9u c\u0169ng t\u1eadn d\u1ee5ng hi\u1ec7n t\u01b0\u1ee3ng “tr\u00f4i d\u1ea1t \u0111\u1ed9 d\u1eabn” trong v\u1eadt li\u1ec7u nh\u1edb \u0111\u1ed5i pha (PCM), m\u1ed9t \u0111\u1eb7c t\u00ednh th\u01b0\u1eddng b\u1ecb xem l\u00e0 nh\u01b0\u1ee3c \u0111i\u1ec3m, \u0111\u1ec3 ph\u1ee5c v\u1ee5 c\u00e1c ph\u00e9p t\u00ednh m\u00f4 ph\u1ecfng \u0111\u1ed9ng h\u1ecdc th\u1ea7n kinh, qua \u0111\u00f3 ti\u1ebft ki\u1ec7m n\u0103ng l\u01b0\u1ee3ng \u0111\u00e1ng k\u1ec3 so v\u1edbi GPU.<\/p>\n\n\n\n

\"\"
Chip nh\u1edb Trung Qu\u1ed1c m\u00f4 ph\u1ecfng b\u1ed9 n\u00e3o, nhanh g\u1ea5p 478 l\u1ea7n GPU Nvidia A100 (Ngu\u1ed3n: Internet)<\/em><\/figcaption><\/figure>\n\n\n\n

So v\u1edbi GPU Nvidia A100, m\u1ed9t trong nh\u1eefng d\u00f2ng chip \u0111\u1ed3 h\u1ecda ph\u1ed5 bi\u1ebfn nh\u1ea5t cho t\u00e1c v\u1ee5 t\u00ednh to\u00e1n AI hi\u1ec7n nay, chip nh\u1edb 40nm c\u1ee7a \u0110\u1ea1i h\u1ecdc B\u1eafc Kinh cho t\u1ed1c \u0111\u1ed9 x\u1eed l\u00fd nhanh h\u01a1n 50 \u0111\u1ebfn 478 l\u1ea7n khi th\u1ef1c hi\u1ec7n ri\u00eang b\u00e0i to\u00e1n t\u00e1i t\u1ea1o b\u1ec1 m\u1eb7t n\u00e3o. \u0110\u00e2y l\u00e0 m\u1ee9c ch\u00eanh l\u1ec7ch l\u1edbn, nh\u01b0ng c\u1ea7n l\u01b0u \u00fd con s\u1ed1 n\u00e0y \u00e1p d\u1ee5ng cho m\u1ed9t t\u00e1c v\u1ee5 chuy\u00ean bi\u1ec7t, kh\u00f4ng ph\u1ea3n \u00e1nh hi\u1ec7u n\u0103ng t\u1ed5ng qu\u00e1t c\u1ee7a chip trong m\u1ecdi lo\u1ea1i ph\u00e9p t\u00ednh nh\u01b0 GPU \u0111a d\u1ee5ng.<\/p>\n\n\n\n

Theo \u0111\u1ea1i di\u1ec7n nh\u00f3m nghi\u00ean c\u1ee9u, Gi\u00e1o s\u01b0 Yang Yuchao \u2013 Ph\u00f3 hi\u1ec7u tr\u01b0\u1edfng tr\u01b0\u1eddng K\u1ef9 thu\u1eadt \u0110i\u1ec7n t\u1eed v\u00e0 M\u00e1y t\u00ednh, \u0110\u1ea1i h\u1ecdc B\u1eafc Kinh, \u0111\u1ed3ng th\u1eddi l\u00e0 tr\u01b0\u1edfng nh\u00f3m ph\u00e1t tri\u1ec3n, con chip c\u00f3 th\u1ec3 ph\u1ee5c d\u1ef1ng ch\u00ednh x\u00e1c c\u00e1c n\u1ebfp g\u1ea5p c\u1ee7a v\u1ecf n\u00e3o, m\u1edf \u0111\u01b0\u1eddng cho vi\u1ec7c x\u00e2y d\u1ef1ng b\u1ea3n sao k\u1ef9 thu\u1eadt s\u1ed1 c\u1ee7a n\u00e3o theo h\u01b0\u1edbng c\u00e1 nh\u00e2n h\u00f3a. \u1ee8ng d\u1ee5ng th\u1ef1c t\u1ebf \u0111\u01b0\u1ee3c k\u1ef3 v\u1ecdng s\u1ebd tr\u1ea3i r\u1ed9ng t\u1eeb ch\u1ea9n \u0111o\u00e1n v\u00e0 \u0111i\u1ec1u tr\u1ecb c\u00e1c b\u1ec7nh tho\u00e1i h\u00f3a th\u1ea7n kinh nh\u01b0 Alzheimer, c\u1ea3i thi\u1ec7n hi\u1ec7u su\u1ea5t giao di\u1ec7n n\u00e3o m\u00e1y t\u00ednh (BCI), \u0111\u1ebfn h\u1ed7 tr\u1ee3 \u0111\u1ecbnh v\u1ecb trong ph\u1eabu thu\u1eadt th\u1ea7n kinh.<\/p>\n\n\n\n

Nghi\u00ean c\u1ee9u n\u00e0y \u0111\u01b0\u1ee3c x\u00e2y d\u1ef1ng d\u1ef1a tr\u00ean n\u1ec1n t\u1ea3ng nhi\u1ec1u th\u1eadp k\u1ef7 ph\u00e1t tri\u1ec3n c\u1ee7a l\u0129nh v\u1ef1c t\u00ednh to\u00e1n th\u1ea7n kinh h\u1ecdc (neuromorphic computing), t\u1eeb c\u00e1c th\u1eed nghi\u1ec7m v\u00f5ng m\u1ea1c silicon \u0111\u1eddi \u0111\u1ea7u c\u1ee7a Carver Mead cho \u0111\u1ebfn chip TrueNorth c\u1ee7a IBM v\u00e0 Loihi c\u1ee7a Intel. \u0110i\u1ec3m kh\u00e1c bi\u1ec7t c\u1ee7a nh\u00f3m B\u1eafc Kinh l\u00e0 t\u1eadp trung v\u00e0o b\u00e0i to\u00e1n t\u00e1i t\u1ea1o b\u1ec1 m\u1eb7t v\u1ecf n\u00e3o \u0111\u1ec3 h\u01b0\u1edbng t\u1edbi gi\u00e1 tr\u1ecb \u1ee9ng d\u1ee5ng l\u00e2m s\u00e0ng, thay v\u00ec ch\u1ec9 d\u1eebng \u1edf m\u00f4 ph\u1ecfng nghi\u00ean c\u1ee9u c\u01a1 b\u1ea3n.<\/p>\n\n\n\n

\"\"
M\u00f4 ph\u1ecfng c\u1ea5u tr\u00fac v\u1ecf n\u00e3o b\u1eb1ng chip nh\u1edb AI (Ngu\u1ed3n: Internet)<\/em><\/figcaption><\/figure>\n\n\n\n

\u1ede giai \u0111o\u1ea1n ti\u1ebfp theo, nh\u00f3m nghi\u00ean c\u1ee9u d\u1ef1 ki\u1ebfn th\u1eed nghi\u1ec7m chip tr\u00ean \u0111\u1ed9ng v\u1eadt v\u00e0 t\u00edch h\u1ee3p v\u1edbi c\u00e1c h\u1ec7 th\u1ed1ng ch\u1ee5p \u1ea3nh y t\u1ebf hi\u1ec7n c\u00f3, tr\u01b0\u1edbc khi ti\u1ebfn t\u1edbi c\u00e1c \u1ee9ng d\u1ee5ng l\u00e2m s\u00e0ng th\u1ef1c t\u1ebf. N\u1ebfu qu\u00e1 tr\u00ecnh th\u1eed nghi\u1ec7m th\u00e0nh c\u00f4ng, \u0111\u00e2y c\u00f3 th\u1ec3 l\u00e0 m\u1ed9t h\u01b0\u1edbng \u0111i m\u1edbi cho ph\u1ea7n c\u1ee9ng AI chuy\u00ean d\u1ee5ng, b\u00ean c\u1ea1nh xu h\u01b0\u1edbng d\u00f9ng GPU \u0111a d\u1ee5ng nh\u01b0 hi\u1ec7n nay, \u0111\u1eb7c bi\u1ec7t trong c\u00e1c l\u0129nh v\u1ef1c \u0111\u00f2i h\u1ecfi m\u00f4 ph\u1ecfng th\u1eddi gian th\u1ef1c v\u1edbi m\u1ee9c ti\u00eau th\u1ee5 n\u0103ng l\u01b0\u1ee3ng th\u1ea5p nh\u01b0 robot t\u1ef1 h\u00e0nh hay h\u1ec7 th\u1ed1ng \u0111i\u1ec1u khi\u1ec3n th\u00f4ng minh.<\/p>\n\n\n\n

V\u1edbi ng\u01b0\u1eddi d\u00f9ng c\u00f4ng ngh\u1ec7 t\u1ea1i Vi\u1ec7t Nam, tin t\u1ee9c n\u00e0y ch\u01b0a t\u00e1c \u0111\u1ed9ng tr\u1ef1c ti\u1ebfp \u0111\u1ebfn c\u00e1c s\u1ea3n ph\u1ea9m m\u00e1y t\u00ednh, laptop hay linh ki\u1ec7n \u0111ang b\u00e1n tr\u00ean th\u1ecb tr\u01b0\u1eddng, do chip v\u1eabn \u1edf giai \u0111o\u1ea1n nghi\u00ean c\u1ee9u. Tuy nhi\u00ean, \u0111\u00e2y l\u00e0 m\u1ed9t t\u00edn hi\u1ec7u cho th\u1ea5y cu\u1ed9c \u0111ua ph\u00e1t tri\u1ec3n ph\u1ea7n c\u1ee9ng AI chuy\u00ean d\u1ee5ng \u0111ang m\u1edf r\u1ed9ng ra ngo\u00e0i ph\u1ea1m vi GPU truy\u1ec1n th\u1ed1ng, \u0111i\u1ec1u \u0111\u00e1ng theo d\u00f5i v\u1edbi nh\u1eefng ai quan t\u00e2m \u0111\u1ebfn xu h\u01b0\u1edbng c\u00f4ng ngh\u1ec7 d\u00e0i h\u1ea1n. \u0110\u1ec3 kh\u00f4ng b\u1ecf l\u1ee1 nh\u1eefng tin t\u1ee9c c\u00f4ng ngh\u1ec7 n\u00f3ng h\u1ed5i, h\u00e3y th\u01b0\u1eddng xuy\u00ean theo d\u00f5i Phong V\u0169 Tech News<\/a> nh\u00e9!<\/p>\n\n\n\n

Ngu\u1ed3n: VnExpress<\/em><\/p>\n\n\n\n

B\u00e0i vi\u1ebft li\u00ean quan:<\/strong><\/p>\n\n\n\n