Vừa nhận được máy chủ Pegatron Nvidia GH200 từ Đài Loan: 144 GB HBM3e, 624 GB RAM, 2U, cực yên tĩnh, có thể đặt dưới bàn làm việc. #Nvidia #GH200 #Pegatron #Server #AI #CôngNghệ #Hardware #AI #Vietnam

https://www.reddit.com/r/LocalLLaMA/comments/1qr0uxi/just_arrived_from_taiwan_pegatron_nvidia_gh200/

Why it matters: The #GH200’s unified memory lets both #CPU and #GPU access data in place giving smoother #OBS workflows, and a future-ready infrastructure for #Armv9. #Opensource innovation powered by #hardware! https://news.opensuse.org/2025/06/20/grace-hopper-to-boost-tw-armv9-builds/
Grace Hopper to Boost Tumbleweed Armv9 Builds

Grace Hopper to Boost Tumbleweed Armv9 Builds The openSUSE Project is preparing to expand its hardware capabilities with a high-performance system designed t...

openSUSE News
#Europe’s #Jupiter #supercomputer hits #exascale threshold at inauguration
The long-awaited Jupiter system was officially inaugurated on Friday at the Jülich Supercomputing Center near Köln (Cologne) in #Germany, and has surpassed the exascale threshold of one quintillion (10¹⁸) operations per second. It comprises roughly 6000 compute nodes, each featuring four of #Nvidia's #GH200 #GraceHopper superchips, built by #Eviden the compute division of #Atos Group.
https://www.theregister.com/2025/09/05/jupiter_supercompute_exascale_threshold/
Europe’s Jupiter supercomputer hits exascale threshold at inauguration

video: It won't be fully functional for a while, though

The Register
In partnership with #SUSE & #NVIDIA, #openSUSE gains top-tier hardware to accelerate #Armv9-based build pipelines. Expect boosted performance, reduced latency, and leaner energy use in #Tumbleweed’s rolling development! #OBS #Superchip #GH200 #GraceHopper https://news.opensuse.org/2025/06/20/grace-hopper-to-boost-tw-armv9-builds/
Grace Hopper to Boost Tumbleweed Armv9 Builds

Grace Hopper to Boost Tumbleweed Armv9 Builds The openSUSE Project is preparing to expand its hardware capabilities with a high-performance system designed t...

openSUSE News
#UK's #IsambardAI, an #HPE #Cray #EX4000 with 5,448 NVIDIA #GH200 #GraceHopper Superchips, powers up as government goes #AI crazy
Brace yourselves #Britain, PM #KeirStarmer's challenged his teams: "set the challenge to all of my teams: show me how they can use AI - not just in the output of government, not just in partnership with yourselves and others in the delivery of services - but also in the very way we do government. How can we transform what we do?"
https://www.theregister.com/2025/06/09/uks_isambardai_super_powers_up/
UK's Isambard-AI super powers up as government goes AI crazy

: Brace yourselves Britain, PM Keir Starmer's challenged his teams: 'show me how they can use AI'

The Register
Korea Institute of Science and Technology Information (KISTI) selects HPE to build South Korea’s largest #supercomputer KISTI-6: Cray EX4000 with two partitions, one featuring NVIDIA #GH200 and another featuring 5th Gen AMD EPYC 100% fan-less DLC hpe.com/us/en/newsro... #HPC #AI

hpe.com/us/en/newsroom...
Bluesky

Bluesky Social
OpenAI 整合超級電腦進入核武領域
OpenAI 宣佈與美國洛斯阿拉莫斯國家實驗室(Los Alamos National Laboratory) […]
The post OpenAI 整合超級電腦進入核武領域 appeared first on 香港 unwire.hk 玩生活.樂科技.
#科技新聞 #gh200 #NVIDIA #venado
https://unwire.hk/2025/01/31/openai-venado-nuclear-weapon/fun-tech/?utm_source=rss&utm_medium=rss&utm_campaign=openai-venado-nuclear-weapon
OpenAI 整合超級電腦進入核武領域

OpenAI 宣佈與美國洛斯阿拉莫斯國家實驗室(Los Alamos National Laboratory)深化合作,運用 AI 模型協助應對核武及相關問題。這項合作旨在減少核戰風險,並提升全球核材料和武器的安全性。合作計劃包括將 OpenAI 最新的 o-series 模型安裝到洛斯阿拉莫斯的 Venado 超級計算機中,利用其強大運算能力處理多項全球性挑戰,例如: 開發新方法來預防和治療疾病 改善對自然及人為威脅的檢測能力 推動美國能源領導地位,優化基礎設施 深化對宇宙基本法則的理解 加強美國電網的網絡安全 加速基礎科學研究,維持科技領先地位Venado 超級計算機由 NVIDIA 和 Hewlett-Packard Enterprises 合作建造,採用最先進的 NVIDIA GH200 Grace Hopper 超級芯片。洛斯阿拉莫斯國家實驗室主任 Thom Mason 表示,隨着國家面臨越來越複雜的威脅,AI 技術能更高效地應對挑戰,並推動科學使命的實現。未來將被移至一個安全的保密網絡,供洛斯阿拉莫斯、勞倫斯利弗莫爾(Lawrence Livermore)和桑迪亞(Sandia)國家實驗室的研究人員共享使用。這項合作標誌著 OpenAI 與政府及科研機構關係的進一步加深。去年夏天,兩者已針對生物科學領域進行試點合作,探索 AI 如何助力科學研究,同時評估其潛在風險。儘管洛斯阿拉莫斯對 AI 的態度保持謹慎,但此次合作顯示他們對 OpenAI 模型的信心大增,準備應對更具挑戰性的全球議題。位於新墨西哥州聖塔菲的洛斯阿拉莫斯國家實驗室(LANL),於1940年代成立,作為曼哈頓計劃開發核武器的重要據點。時至今日,該實驗室已轉型為高科技研究中心,成為多台超級計算機的所在地。

香港 unwire.hk 玩生活.樂科技
#NVIDIA #GH200 Grace #CPU vs. #AMD #EPYC 9005 #Turin CPU Performance Review
The article is focusing just on the CPU performance across these servers. Of course, the main selling point of the GH200 Superchip is when leveraging the Hopper GPU in tandem. But for those curious how well Grace / Neoverse-V2 compares to the new #EPYC9005 series in single socket configurations, hopefully today's numbers provided some insight.
https://www.phoronix.com/review/nvidia-grace-epyc-turin
NVIDIA GH200 Grace CPU vs. AMD EPYC 9005 Turin CPU Performance Review

The converged memory architecture of #GH200 can boost TTFT (time to first token) in multiturn user interactions by up to 2x on Llama 3 70B, compared to x86-based #H100 servers

https://developer.nvidia.com/blog/nvidia-gh200-superchip-accelerates-inference-by-2x-in-multiturn-interactions-with-llama-models/

#AI #Inference

NVIDIA GH200 Superchip Accelerates Inference by 2x in Multiturn Interactions with Llama Models | NVIDIA Technical Blog

Deploying large language models (LLMs) in production environments often requires making hard trade-offs between enhancing user interactivity and increasing system throughput.

NVIDIA Technical Blog
A Quick Introduction to the #NVIDIA #GH200 aka #GraceHopper
If you want Arm CPU with LPDDR5X memory & high-speed interconnect to a Hopper GPU today, you are looking at NVIDIA GH200. NVIDIA has two key innovations here that really set it apart from what you can get elsewhere. First, LPDDR5X memory is soldered onto the package. Second, the NVLink-C2C is a high-bandwidth interface between the CPU and the GPU. The 72 Arm v9 cores are Arm Neoverse V2 from 2022 so not exciting.
https://www.servethehome.com/a-quick-introduction-to-the-nvidia-gh200-aka-grace-hopper-arm/
A Quick Introduction to the NVIDIA GH200 aka Grace Hopper

The NVIDIA GH200 or "Grace Hopper" is far from a single product. We have a quick guide so when someone says "GH200" you know what to look for

ServeTheHome