{"id":156738,"date":"2025-11-11T10:00:00","date_gmt":"2025-11-11T03:00:00","guid":{"rendered":"https:\/\/phongvu.vn\/cong-nghe\/?p=156738"},"modified":"2025-11-11T09:22:32","modified_gmt":"2025-11-11T02:22:32","slug":"nvidia-bat-dau-san-xuat-gpu-rubin-voi-cong-nghe-hbm4-tien-tien","status":"publish","type":"post","link":"https:\/\/phongvu.vn\/cong-nghe\/nvidia-bat-dau-san-xuat-gpu-rubin-voi-cong-nghe-hbm4-tien-tien\/","title":{"rendered":"NVIDIA b\u1eaft \u0111\u1ea7u s\u1ea3n xu\u1ea5t GPU Rubin v\u1edbi c\u00f4ng ngh\u1ec7 HBM4 ti\u00ean ti\u1ebfn"},"content":{"rendered":"\n

PV Tech News \u2013 NVIDIA \u0111\u00e3 ch\u00ednh th\u1ee9c \u0111\u01b0a d\u00f2ng GPU th\u1ebf h\u1ec7 ti\u1ebfp theo \u201cRubin\u201d v\u00e0o d\u00e2y chuy\u1ec1n s\u1ea3n xu\u1ea5t v\u00e0 \u0111\u1ed3ng th\u1eddi nh\u1eadn m\u1eabu b\u1ed9 nh\u1edb HBM4 t\u1eeb t\u1ea5t c\u1ea3 c\u00e1c nh\u00e0 cung c\u1ea5p l\u1edbn. Theo \u0111\u00f3, c\u00f4ng ty \u0111ang s\u1eb5n s\u00e0ng cho b\u01b0\u1edbc nh\u1ea3y v\u1ecdt trong h\u1ea1 t\u1ea7ng AI n\u0103m 2026.<\/strong><\/p>\n\n\n\n

Nvidia \u0111\u00e3 b\u01b0\u1edbc v\u00e0o giai \u0111o\u1ea1n s\u1ea3n xu\u1ea5t s\u1edbm cho d\u00f2ng GPU th\u1ebf h\u1ec7 m\u1edbi c\u00f3 t\u00ean m\u00e3 Rubin<\/strong>, k\u1ebf nhi\u1ec7m cho d\u00f2ng Blackwell<\/strong> hi\u1ec7n t\u1ea1i. G\u1ea7n \u0111\u00e2y, Nvidia cung c\u1ea5p 50.000 chip cho H\u00e0n Qu\u1ed1c x\u00e2y si\u00eau nh\u00e0 m\u00e1y AI<\/a> \u0111i\u1ec1u n\u00e0y \u0111\u01b0\u1ee3c xem l\u00e0 m\u1ed9t c\u1ed9t m\u1ed1c quan tr\u1ecdng trong chi\u1ebfn l\u01b0\u1ee3c m\u1edf r\u1ed9ng h\u1ea1 t\u1ea7ng AI c\u1ee7a h\u00e3ng. \u0110\u1ed3ng th\u1eddi, Nvidia x\u00e1c nh\u1eadn \u0111\u00e3 nh\u1eadn \u0111\u01b0\u1ee3c c\u00e1c m\u1eabu b\u1ed9 nh\u1edb HBM4<\/strong> t\u1eeb t\u1ea5t c\u1ea3 nh\u00e0 s\u1ea3n xu\u1ea5t DRAM l\u1edbn tr\u00ean th\u1ebf gi\u1edbi, bao g\u1ed3m Samsung, SK Hynix v\u00e0 Micron.<\/p>\n\n\n\n

Th\u00f4ng tin n\u00e0y \u0111\u01b0\u1ee3c CEO Jensen Huang h\u00e9 l\u1ed9 trong chuy\u1ebfn th\u0103m nh\u00e0 m\u00e1y TSMC t\u1ea1i \u0110\u00e0i Loan. \u00d4ng cho bi\u1ebft \u201cRubin \u0111\u00e3 ch\u00ednh th\u1ee9c c\u00f3 m\u1eb7t tr\u00ean d\u00e2y chuy\u1ec1n s\u1ea3n xu\u1ea5t\u201d, kh\u1eb3ng \u0111\u1ecbnh s\u1ef1 chu\u1ea9n b\u1ecb s\u1edbm cho th\u1ebf h\u1ec7 GPU<\/a> ph\u1ee5c v\u1ee5 c\u00e1c t\u00e1c v\u1ee5 AI ng\u00e0y c\u00e0ng n\u1eb7ng.<\/p>\n\n\n\n

\"NVIDIA
NVIDIA b\u1eaft \u0111\u1ea7u s\u1ea3n xu\u1ea5t GPU Rubin v\u1edbi c\u00f4ng ngh\u1ec7 HBM4 ti\u00ean ti\u1ebfn. (Ngu\u1ed3n: wccftech)<\/figcaption><\/figure>\n\n\n\n

Tham kh\u1ea3o m\u1ed9t s\u1ed1 d\u00f2ng card m\u00e0n h\u00ecnh Nvidia<\/a> t\u1ea1i Phong V\u0169<\/a>: <\/p>\n\n\n\n

<\/div><\/div>
<\/div><\/div>\n\n\n\n

T\u1ea1i s\u1ef1 ki\u1ec7n GTC 2025 tr\u01b0\u1edbc \u0111\u00f3, Nvidia t\u1eebng gi\u1edbi thi\u1ec7u c\u1ea5u tr\u00fac \u201cVera Rubin Superchip\u201d<\/strong> \u2013 m\u1ed9t thi\u1ebft k\u1ebf k\u1ebft h\u1ee3p gi\u1eefa CPU<\/a> Vera v\u00e0 hai GPU Rubin trong c\u00f9ng m\u1ed9t n\u1ec1n t\u1ea3ng, h\u01b0\u1edbng \u0111\u1ebfn vi\u1ec7c t\u0103ng hi\u1ec7u su\u1ea5t x\u1eed l\u00fd cho c\u00e1c trung t\u00e2m d\u1eef li\u1ec7u th\u1ebf h\u1ec7 m\u1edbi. Vi\u1ec7c Rubin \u0111i v\u00e0o s\u1ea3n xu\u1ea5t s\u1edbm cho th\u1ea5y Nvidia \u0111ang theo \u0111\u00fang l\u1ed9 tr\u00ecnh \u0111\u1ec3 ra m\u1eaft th\u1ebf h\u1ec7 si\u00eau chip n\u00e0y v\u00e0o kho\u1ea3ng n\u0103m 2026.<\/p>\n\n\n\n

V\u1ec1 ph\u00eda \u0111\u1ed1i t\u00e1c s\u1ea3n xu\u1ea5t, TSMC \u0111\u00e3 n\u00e2ng c\u00f4ng su\u1ea5t ti\u1ebfn tr\u00ecnh 3 nm (N3) l\u00ean kho\u1ea3ng 50 % d\u00e0nh ri\u00eang cho Nvidia, nh\u1eb1m \u0111\u00e1p \u1ee9ng nhu c\u1ea7u kh\u1ed5ng l\u1ed3 c\u1ee7a d\u00f2ng Blackwell hi\u1ec7n t\u1ea1i c\u0169ng nh\u01b0 chu\u1ea9n b\u1ecb cho Rubin. Vi\u1ec7c m\u1edf r\u1ed9ng quy m\u00f4 s\u1edbm gi\u00fap Nvidia duy tr\u00ec v\u1ecb th\u1ebf th\u1ed1ng tr\u1ecb trong th\u1ecb tr\u01b0\u1eddng GPU AI, \u0111\u1ed3ng th\u1eddi gi\u1ea3m r\u1ee7i ro \u0111\u1ee9t g\u00e3y chu\u1ed7i cung \u1ee9ng khi nhu c\u1ea7u chip t\u0103ng \u0111\u1ed9t bi\u1ebfn.<\/p>\n\n\n\n

C\u00f4ng ngh\u1ec7 HBM4 (High Bandwidth Memory 4) c\u0169ng \u0111\u00f3ng vai tr\u00f2 then ch\u1ed1t trong k\u1ebf ho\u1ea1ch n\u00e0y. So v\u1edbi HBM3E hi\u1ec7n nay, HBM4 mang l\u1ea1i b\u0103ng th\u00f4ng cao h\u01a1n \u0111\u00e1ng k\u1ec3, ti\u00eau th\u1ee5 \u0111i\u1ec7n n\u0103ng th\u1ea5p h\u01a1n v\u00e0 kh\u1ea3 n\u0103ng m\u1edf r\u1ed9ng dung l\u01b0\u1ee3ng l\u1edbn h\u01a1n nh\u1eefng y\u1ebfu t\u1ed1 t\u1ed1i quan tr\u1ecdng \u0111\u1ed1i v\u1edbi c\u00e1c m\u00f4 h\u00ecnh AI<\/a> kh\u1ed5ng l\u1ed3 nh\u01b0 GPT-6 hay Gemini Ultra. Vi\u1ec7c Nvidia nh\u1eadn m\u1eabu t\u1eeb to\u00e0n b\u1ed9 c\u00e1c nh\u00e0 s\u1ea3n xu\u1ea5t cho th\u1ea5y h\u00e3ng \u0111ang ch\u1ee7 \u0111\u1ed9ng \u0111\u1ea3m b\u1ea3o ngu\u1ed3n cung \u1ed5n \u0111\u1ecbnh cho giai \u0111o\u1ea1n t\u0103ng t\u1ed1c c\u1ee7a m\u1ea3ng AI trong 2026\u20132027.<\/p>\n\n\n\n

Theo d\u1ef1 ki\u1ebfn, d\u00f2ng GPU Rubin s\u1ebd b\u1eaft \u0111\u1ea7u s\u1ea3n xu\u1ea5t h\u00e0ng lo\u1ea1t v\u00e0o qu\u00fd III\/2026, tr\u01b0\u1edbc khi xu\u1ea5t hi\u1ec7n tr\u00ean c\u00e1c n\u1ec1n t\u1ea3ng si\u00eau m\u00e1y ch\u1ee7 v\u00e0 trung t\u00e2m d\u1eef li\u1ec7u l\u1edbn. M\u1edbi \u0111\u00e2y, Nvidia c\u00f4ng b\u1ed1 v\u1ed1n h\u00f3a Nvidia s\u1eafp c\u00e1n m\u1ed1c 5.000 t\u1ef7 USD<\/a>. \u0110\u00e2y \u0111\u01b0\u1ee3c xem l\u00e0 b\u01b0\u1edbc \u0111\u1ec7m gi\u00fap Nvidia c\u1ee7ng c\u1ed1 v\u1ecb th\u1ebf d\u1eabn \u0111\u1ea7u trong l\u0129nh v\u1ef1c x\u1eed l\u00fd AI, khi c\u00e1c \u0111\u1ed1i th\u1ee7 nh\u01b0 AMD v\u00e0 Intel c\u0169ng \u0111ang r\u00e1o ri\u1ebft tung ra c\u00e1c gi\u1ea3i ph\u00e1p c\u1ea1nh tranh. <\/p>\n\n\n\n

Ngu\u1ed3n: wccftech<\/a><\/em><\/p>\n\n\n\n

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