NVIDIA GB300 NVL72
Liquid Cooled Rack-Scale Solution with 72 NVIDIA B300 GPUs and 36 Grace CPUs
- GPU: 72 NVIDIA B300 GPUs via NVIDIA Grace Blackwell Superchips
- CPU: 36 NVIDIA Grace CPUs via NVIDIA Grace Blackwell Superchips
- Memory: Up to 21 TB HBM3e (GPU Memory), Up to 17 TB LPDDR5X (System Memory)
- Drives: up to 144 E1.S PCIe 5.0 drive bays
Liquid-Cooled Universal GPU Systems
Direct-to-chip liquid-cooled systems for high-density AI infrastructure at scale.
- GPU: NVIDIA HGX B300/B200/H200
- CPU: Intel® Xeon® or AMD EPYC™
- Memory: Up to 32 DIMMs, 9TB
- Drives: Up to 24 Hot-swap U.2 or 2.5" NVMe/SATA drives
Universal GPU Systems
Modular Building Block Design, Future Proof Open-Standards Based Platform in 4U, 5U, 8U, or 10U for Large Scale AI training and HPC Applications
- GPU: NVIDIA HGX B300/B200/H200, AMD Instinct MI350 Series and MI325X/MI300X/MI250 OAM Accelerator, Intel Data Center GPU Max Series
- CPU: Intel® Xeon® or AMD EPYC™
- Memory: Up to 32 DIMMs, 9TB
- Drives: Up to 24 Hot-swap E1.S, U.2 or 2.5" NVMe/SATA drives
3U/4U/5U GPU Lines with PCIe 5.0
Maximum Acceleration and Flexibility for AI, Deep Learning and HPC Applications
- GPU: Up to 10 NVIDIA H100 PCIe GPUs, or up to 10 double-width PCIe GPUs
- CPU: Intel® Xeon® or AMD EPYC™
- Memory: Up to 32 DIMMs, 9TB or 12TB
- Drives: Up to 24 Hot-swap 2.5" SATA/SAS/NVMe
NVIDIA MGX™ Systems
Modular Building Block Platform Supporting Today's and Future GPUs, CPUs, and DPUs
- GPU: Up to 4 NVIDIA PCIe GPUs including NVIDIA RTX PRO™ 6000 Blackwell Server Edition, H200 NVL, H100 NVL, and L40S
- CPU: NVIDIA GH200 Grace Hopper™ Superchip, Grace™ CPU Superchip, or Intel® Xeon®
- Memory: Up to 960GB ingegrated LPDDR5X memory (Grace Hopper or Grace CPU Superchip) or 16 DIMMs, up to 4TB DRAM (Intel)
- Drives: Up to 8 E1.S + 4 M.2 drives
AMD APU Systems
Multi-processor system combining CPU and GPU, Designed for the Convergence of AI and HPC
- GPU: 4 AMD Instinct MI300A Accelerated Processing Unit (APU)
- CPU: AMD Instinct™ MI300A Accelerated Processing Unit (APU)
- Memory: Up to 512GB integrated HBM3 memory (4x 128GB)
- Drives: Up to 8 2.5" NVMe or Optional 24 2.5" SATA/SAS via storage add-on card + 2 M.2 drives
4U GPU Lines with PCIe 4.0
Flexible Design for AI and Graphically Intensive Workloads, Supporting Up to 10 GPUs
- GPU: NVIDIA HGX A100 8-GPU with NVLink, or up to 10 double-width PCIe GPUs
- CPU: Intel® Xeon® or AMD EPYC™
- Memory: Up to 32 DIMMs, 8TB DRAM or 12TB DRAM + PMem
- Drives: Up to 24 Hot-swap 2.5" SATA/SAS/NVMe
2U 2-Node Multi-GPU with PCIe 4.0
Dense and Resource-saving Multi-GPU Architecture for Cloud-Scale Data Center Applications
- GPU: Up to 3 double-width PCIe GPUs per node
- CPU: Intel® Xeon® or AMD EPYC™
- Memory: Up to 8 DIMMs, 2TB per node
- Drives: Up to 2 front hot-swap 2.5” U.2 per node
GPU Workstation
Flexible Solution for AI/Deep Learning Practitioners and High-end Graphics Professionals
- GPU: Up to 4 double-width PCIe GPUs
- CPU: Intel® Xeon®
- Memory: Up to 16 DIMMs, 6TB
- Drives: Up to 8 hot-swap 2.5” SATA/NVMe
Supermicro JumpStart :與 NVIDIA HGX B200 共度一週
Supermicro JumpStart 在硬體評估方面採取了截然不同的做法。與其在共用實驗室環境中進行短暫且照本宣科的示範JumpStart 符合資格的客戶免費、在限定時間內,直接使用一系列真正的生產級伺服器。
透過Intel® TDX、NVIDIA機密運算,以及搭載NVIDIA HGX™ B200 GPU的Supermicro伺服器,保護AI工作負載:A Foundation for Confidential AI at Scale
閱讀白皮書Supermicro SYS-821GE-TNHR 8x NVIDIA H200 GPU 氣冷式 AI 伺服器
今天我們將繼續探討大型 AI 伺服器,並聚焦於 Supermicro SYS-821GE-TNHR。當人們討論 Supermicro 在 AI 伺服器領域的卓越實力時,這款氣冷式 NVIDIA 伺服器是市場上獨樹一幟的系統之一。
Supermicro AMD EPYC 伺服器提供AI (AS-4125GS-TNRT)
Supermicro GPU 伺服器型號與規格之多,遠非本文所能盡述。今天,我們將聚焦於該公司一款相對較新的 4U 風冷式 GPU 伺服器,該機型支援兩顆AMD EPYC 系列處理器、PCIe 介面,並可選擇搭載八張雙寬度或十二張單寬度的外接 GPU 顯示卡。
深入探討液冷式 Supermicro SYS-821GE-TNHR 8x NVIDIA H100 AI 伺服器
今天我們將探討液冷式 Supermicro SYS-821GE-TNHR 伺服器。這是 Supermicro 獨特的 8x NVIDIA H100 系統,其採用液冷技術,可降低散熱成本及功耗。由於我們已取得相關照片,因此決定撰寫此文。
3000WAMD Epyc 拆解,特邀 Level1Techs 的 Wendell
我們正在檢視一款針對AI 機器學習進行優化的伺服器。Supermicro 大量心力,將盡可能多的元件塞進 2114GT-DNR(2U2N)這款密度優化的伺服器中。這是一款非常酷的設計:這個 2U 機箱內實際上容納了兩個系統。 兩組冗餘電源供應器各具 2,600W 功率,稍後我們將了解為何需要如此大的功率。它搭載了六顆AMD Instinct GPU 以及雙路Epyc 。由此可見Supermicro 在這款伺服器的設計上Supermicro 多少工程心力。
Supermicro TECHTalk:高密度 AI 訓練/深度學習伺服器
混合式 2U2N GPU 工作站-伺服器平台 Supermicro SYS-210GP-DNR 實測
今天我們將結束最新系列,探討 Supermicro SYS-210GP-DNR,這是一款 2U、2 節點、6 GPU 系統,Patrick 最近在 Supermicro 總部進行了實測。
Supermicro SYS-220GQ-TNAR+ 一款 NVIDIA Redstone 2U 伺服器
今天我們將探討 Supermicro SYS-220GQ-TNAR+,Patrick 最近在 Supermicro 總部進行了實測。