NVIDIAがVera Rubinの帯域幅を10ヶ月で3回書き換えました
13TB/s → 20.5TB/s → 22TB/s、70%の上方修正
計画的ロードマップではなく「追われる者の軌跡」に見えます
AMDのInstinct MI455Xが19.6TB/sで迫り、OpenAIとの6GW契約が締結された今、王座は揺れています
競争がもたらした恩恵と、市場構造の変化を解説しました
note.com/joho_no_toda...
#NVIDIA #AMD #AI半導体
NVIDIA rewrites specs 3 times
NVIDIAがVera Rubinの帯域幅を10ヶ月で3回書き換えました
13TB/s → 20.5TB/s → 22TB/s、70%の上方修正
計画的ロードマップではなく「追われる者の軌跡」に見えます
AMDのInstinct MI455Xが19.6TB/sで迫り、OpenAIとの6GW契約が締結された今、王座は揺れています
競争がもたらした恩恵と、市場構造の変化を解説しました
note.com/joho_no_toda...
#NVIDIA #AMD #AI半導体
NVIDIA rewrites specs 3 times
wccftech.com/nvidia-is-fe...
wccftech.com/nvidia-is-fe...
GPU: HBM + (LP) DDR DRAM
NVIDIA x.com/ogawa_tter/s...
Patent, Oct 2023, (Apr 2021)
ASSYST 2023, with GaTech
H100 HBM + Tensor Offloading-Memory
AMD
"Integrated Memory Controller Circuitry", (P: Sep 2023) patents.google.com/patent/US202...
MI455X, CES 2026 x.com/SemiAnalysis...
HBM4 + LPDDR5
GPU: HBM + (LP) DDR DRAM
NVIDIA x.com/ogawa_tter/s...
Patent, Oct 2023, (Apr 2021)
ASSYST 2023, with GaTech
H100 HBM + Tensor Offloading-Memory
AMD
"Integrated Memory Controller Circuitry", (P: Sep 2023) patents.google.com/patent/US202...
MI455X, CES 2026 x.com/SemiAnalysis...
HBM4 + LPDDR5
https://diit.cz/clanek/nvidia-znovu-odklada-rubin-masova-vyroba-se-posouva-na-brezen
https://diit.cz/clanek/nvidia-znovu-odklada-rubin-masova-vyroba-se-posouva-na-brezen
Pro akceleraci úloh umělé inteligence má AMD nový rack Helios. Ten obsahuje první 2nm čipy společnosti, a to jak CPU EPYC, tak i GPU Instinct MI455X....
Pro akceleraci úloh umělé inteligence má AMD nový rack Helios. Ten obsahuje první 2nm čipy společnosti, a to jak CPU EPYC, tak i GPU Instinct MI455X....
https://diit.cz/clanek/amd-ukazala-kremik-instinct-mi455x-s-320-miliardami-tranzistoru-432-gb-hbm4
https://diit.cz/clanek/amd-ukazala-kremik-instinct-mi455x-s-320-miliardami-tranzistoru-432-gb-hbm4
📊 After a big rally last year, some traders are wary amid lofty expectations and concerns about an AI spending bubble.
Source: Yahoo Finance
📊 After a big rally last year, some traders are wary amid lofty expectations and concerns about an AI spending bubble.
Source: Yahoo Finance
https://www.heise.de/en/news/AMD-shows-huge-chip-constructs-Epyc-Venice-and-Instinct-MI455X-11133190.html?utm_source=flipboard&utm_medium=activitypub
Posted into heise online in English @heise-online-in-english-heiseonline
https://www.heise.de/en/news/AMD-shows-huge-chip-constructs-Epyc-Venice-and-Instinct-MI455X-11133190.html?utm_source=flipboard&utm_medium=activitypub
Posted into heise online in English @heise-online-in-english-heiseonline
Erstmals zeigt AMD-Chefin Lisa Su nackte Versionen der nächsten Serverprozessoren und KI-Beschleuniger. Beide sind riesig […]
Erstmals zeigt AMD-Chefin Lisa Su nackte Versionen der nächsten Serverprozessoren und KI-Beschleuniger. Beide sind riesig […]
https://www.heise.de/news/AMD-zeigt-riesige-Chipkonstrukte-Epyc-Venice-und-Instinct-MI455X-11133033.html?utm_source=flipboard&utm_medium=activitypub
Gepostet in Künstliche Intelligenz @k-nstliche-intelligenz-heiseonline
https://www.heise.de/news/AMD-zeigt-riesige-Chipkonstrukte-Epyc-Venice-und-Instinct-MI455X-11133033.html?utm_source=flipboard&utm_medium=activitypub
Gepostet in Künstliche Intelligenz @k-nstliche-intelligenz-heiseonline
quasarzone.com/bbs/qn_repor...
- AMD가 MI455X 탑재 AI 인프라 헬리오스, 라이젠 AI 400 시리즈, AI Max+ 모델, ROCm 7.2 업데이트, 라이젠 7 9850X3D CPU(9800X3D보다 400MHz 빠름)를 발표
- 모두 2026년 1~2분기 출시 예정
quasarzone.com/bbs/qn_repor...
- AMD가 MI455X 탑재 AI 인프라 헬리오스, 라이젠 AI 400 시리즈, AI Max+ 모델, ROCm 7.2 업데이트, 라이젠 7 9850X3D CPU(9800X3D보다 400MHz 빠름)를 발표
- 모두 2026년 1~2분기 출시 예정
#AMD
open.substack.com/pub/chippub/...
#AMD
open.substack.com/pub/chippub/...
Ryan Shrout / @ryanshrout:
@AMD The total compute available in each rack is really...astounding. To put this performance in context, @LisaSu says that MI455X will offer up to 10x better perf than the previous generation. [image]
Ryan Shrout / @ryanshrout:
@AMD The total compute available in each rack is really...astounding. To put this performance in context, @LisaSu says that MI455X will offer up to 10x better perf than the previous generation. [image]
Patrick Moorhead / @patrickmoorhead:
"Helios, the world's best AI rack"... oooh this is 🌶️ MI455X is a giant chip. 4 per compute tray. Venice CPU with 256 cores. "Data at full speed". Plus Pensando Volcano and Salina. $AMD [image]
Patrick Moorhead / @patrickmoorhead:
"Helios, the world's best AI rack"... oooh this is 🌶️ MI455X is a giant chip. 4 per compute tray. Venice CPU with 256 cores. "Data at full speed". Plus Pensando Volcano and Salina. $AMD [image]
@economyapp:
$AMD AMD Data Center at CES: • Introduces "Helios" AI Rack. • MI455X 10x Performance vs. MI355X. • MI500 coming in 2027 with a massive leap. [image]
@economyapp:
$AMD AMD Data Center at CES: • Introduces "Helios" AI Rack. • MI455X 10x Performance vs. MI355X. • MI500 coming in 2027 with a massive leap. [image]