servethehome

Author page description

September 1

  • MiTAC G8825Z5 AMD Instinct MI325X 8-GPU Server Review

    ➀ The MiTAC G8825Z5 is an 8-GPU server featuring AMD Instinct MI325X accelerators with 2TB HBM3E memory, powered by dual AMD EPYC Turin CPUs;

    ➀ The 8U chassis employs a modular design with separate GPU and CPU trays, 15 hot-swappable fan modules, and six 3.3kW Titanium PSUs for optimized cooling and redundancy;

    ➂ The server emphasizes serviceability, offering 12 PCIe slots for expansion and a rear-agnostic I/O design preferred in data center environments.

    AMDMiTACHPC

August 29

  • Crucial X6 2TB Portable USB SSD Review

    ➀ The Crucial X6 2TB is a cost-effective, compact USB 3.2 Gen2 (10Gb/s) external SSD, priced lower than competitors like the Crucial X9 Pro and Samsung T7 Shield.

    ➁ It offers a read speed of up to 800MB/s and includes shock resistance, vibration-proofing, and drop resistance up to 6.5 feet, with a lightweight design and bundled Type-C cable.

    ➂ While its sequential write speed is unspecified, it caters to budget-conscious users prioritizing affordability and portability over peak performance.

    CrucialSSDUSB

August 2

  • Deploying AMD Instead of Arm in our Infrastructure 2025 Here is Why

    ➀ STH explains its choice of AMD EPYC over Arm in 2025 infrastructure, citing Arm's challenges in enterprise adoption, including installed base compatibility, hardware limitations, and insufficient software support;

    ➁ Power efficiency gains from Arm are now minimal compared to AMD/Intel x86 CPUs, especially with AI GPU servers dominating power consumption (e.g., AMD Instinct MI325X at 1kW/accelerator);

    ➂ Enterprise Arm adoption faces hardware scarcity and vendor neglect: NVIDIA Grace CPUs are outdated for general workloads, while major vendors like Dell/HPE offer limited Arm options (e.g., HPE ProLiant RL300 Gen11's market failure).

    AMDArmNVIDIA

August 1

  • Seagate Expansion 28TB External Hard Drive and the HAMR HDD Within Mini-Review

    ➀ Seagate's 28TB Expansion external hard drive integrates HAMR (Heat-Assisted Magnetic Recording) technology, offering high-capacity storage in a compact design with USB 3.2 and external power support;

    ➁ Opening the enclosure reveals a Seagate Barracuda 28TB HAMR drive, typically not sold separately and possibly rebadged from enterprise-grade Exos or IronWolf Pro HDDs;

    ➂ Shucking the external drive provides significant cost savings ($240 less than standalone 28TB HDDs), though sacrificing warranty coverage.

    HDD

July 7

  • Kyocera Pegatron and SMART Show CXL Over Optics at Computex 2025

    ➀ Kyocera Pegatron and SMART showcased a prototype demonstrating CXL memory accessed remotely via optical connections at Computex 2025;

    ➁ The setup utilized SMART's CXL memory module, Kyocera's PCIe-to-optical converter, and MTP/MPO fiber cables, enabling dynamic memory allocation in distributed systems;

    ➂ The technology highlights future applications in AI clusters and aligns with upcoming PCIe Gen6 advancements, signaling progress toward optical co-packaging solutions.

    CXL

July 5

  • New Intel E610 NICs Shown for Low Power 10Gbase-T and 2.5GbE

    ➀ Intel showcased its new E610 series NICs at HPE Discover 2025, including a dual-port 10Gbase-T model and a quad-port 2.5GbE variant;

    ➁ The dual 10Gbase-T PCIe Gen4 x4 NIC consumes only 5.1W at full load, supporting 10GbE/5GbE/2.5GbE/1GbE speeds;

    ➂ The quad 2.5GbE OCP NIC 3.0 model offers up to 8.6W power efficiency, positioning it as an upgrade path for legacy Intel i350 users.

    2.5GbEIntel

July 4

  • Dell and CoreWeave Show Off First NVIDIA GB300 NVL72 Rack

    ➀ Dell and CoreWeave unveiled the first NVIDIA GB300 NVL72 rack, utilizing Dell PowerEdge XE9712 servers and Vertiv cooling systems with a capacity of up to 2MW per rack;

    ➁ The GB300 NVL72 integrates NVIDIA's Grace Blackwell Ultra architecture, delivering 21TB of memory and leveraging InfiniBand Quantum-X networking for high-speed AI workloads;

    ➂ This ultra-high-density rack represents a significant advancement over previous GB200 systems, featuring enhanced liquid cooling and accelerators, with costs potentially exceeding $3.7 million per rack.

    DellHPCNVIDIA

July 3

  • HPE Juniper Deal Closes Ending a Saga

    ➀ HPE finalized its acquisition of Juniper Networks after 18 months, completing the $14 billion deal following a DOJ settlement requiring divestiture of its Aruba Instant On business and AI Ops licensing;

    ➁ The merger strengthens HPE's networking portfolio, positioning it to compete with Cisco and NVIDIA in enterprise/service provider markets and AI cluster networking;

    ➂ Key challenges include integrating product roadmaps and challenging NVIDIA's dominance in AI infrastructure East-West networking with Juniper's UltraEthernet technology.

    HPENetworking

July 2

  • Crucial T710 2TB PCIe Gen5 NVMe SSD Review

    ➀ The Crucial T710 2TB PCIe Gen5 NVMe SSD claims top-tier performance with up to 14,500 MB/s sequential reads and 13,800 MB/s writes, surpassing competitors like the T705 and Samsung 9100 PRO;

    ➁ Featuring a Silicon Motion SM2508 controller and Micron TLC NAND with 2GB DRAM, it emphasizes improved power efficiency and endurance (1,200TBW for 2TB model);

    ➂ Real-world testing confirms its leadership in Gen5 performance benchmarks, though practical user gains over prior generations may vary depending on workloads.

    CrucialPCIe Gen5SSD

July 1

  • The 2025 PCIe GPU in Server Guide

    ➀ The article analyzes PCIe GPU configurations in servers for 2025, focusing on Supermicro and NVIDIA solutions tailored for enterprise AI factories, edge computing, and hybrid workloads.

    ➁ Key GPU models include NVIDIA H200 NVL for AI inference, RTX PRO 6000 Blackwell for graphics-AI hybrid tasks, and L40S for cost efficiency, with systems supporting up to 8x GPUs via PCIe switches.

    ➂ Supermicro introduces a new NVIDIA MGX platform with ConnectX-8 SuperNICs, replacing traditional PCIe switches to enhance GPU networking efficiency and reduce system complexity.

    NVIDIAPCIeSupermicro

June 30

  • ASRock Rack 2U12L2S-SIENA Review 2U AMD EPYC 8004 Siena Server

    ➀ The ASRock Rack 2U12L2S-SIENA is a 2U server featuring AMD's EPYC 8004 'Siena' processors, designed for users seeking balanced performance and flexibility with 12 front-accessible 3.5" drive bays;

    ➁ It includes redundant 800W 80Plus Platinum PSUs, dual 1GbE ports, and tool-less drive installation options for HDDs/SSDs;

    ➂ Compact at 680mm depth, the server targets environments prioritizing storage density and mid-range compute power without extreme core counts.

    AMDASRock RackEPYC 8004

June 29

  • STH Q2 2025 Letter from the Editor Travel Time

    ➀ STH expands its digital footprint, with six properties exceeding 100,000 monthly views, including the main site, YouTube channels, Substack, and newsletter;

    ➁ Extensive travel for content production (e.g., Dell factory tours, Computex coverage) impacts content pipeline, requiring meticulous post-production efforts like barcode blurring;

    ➂ Team scaling in Scottsdale continues, with new hires and plans to address production challenges, while the Axautik Group's Substack grows rapidly as a key revenue stream.

    AsusDell

June 28

  • Supermicro Hyper SuperServer SYS-222HA-TN Review

    ➀ The Supermicro Hyper SuperServer SYS-222HA-TN is a 2U high-performance server featuring dual Intel Xeon 6900 series CPUs and support for MCRDIMMs, designed for demanding workloads;

    ➁ It includes eight PCIe Gen5 NVMe bays, supporting storage capacities up to 1PB with high-density SSDs like the Solidigm D5-P5336 122.88TB;

    ➂ The system offers extensive expansion options, including support for four double-width GPUs, redundant power supplies, and OCP NIC 3.0 slots for flexible networking configurations.

    HPCPCIeSSD

June 27

  • Solidigm D5-P5336 122.88TB NVMe SSD Review

    ➀ The Solidigm D5-P5336 122.88TB NVMe SSD offers extreme capacity in a 2.5" U.2 form factor, leveraging 192-layer QLC NAND and PCIe Gen4 technology;

    ➀ The drive prioritizes capacity over speed, delivering up to 7300 MB/s sequential read speeds and 25K random write IOPS (32K block size) with a 5-year endurance rating of 0.6 DWPD (73TB/day);

    ➂ Designed for enterprise storage needs, the SSD eliminates traditional endurance concerns through massive overprovisioning, supporting 134.3PBW total writes over its lifespan.

    NVMeSSD

June 26

  • New HPE ProLiant DL325 Gen12 and DL345 Gen12 at HPE Discover with AMD EPYC

    ➀ HPE introduced new ProLiant DL325 Gen12 and DL345 Gen12 servers at HPE Discover 2025, featuring AMD EPYC 9005 'Turin' processors;

    ➁ Both models support up to 24 DIMMs per socket for expanded memory capacity and include the iLO 7 management system;

    ➂ The 2U DL345 Gen12 offers superior storage flexibility and I/O expansion, targeting high-core-count workloads and modernization from legacy platforms.

    AMDHPCHPE

June 23

  • SK Hynix Beetle X31 USB SSD Review

    ➀ The SK Hynix Beetle X31 is a compact 2TB external USB SSD offering read/write speeds up to 1050/1000 MB/s via USB 3.2 Gen2;

    ➁ Positioned in the $140-$155 range, it competes with SanDisk Extreme, Crucial X9 Pro, and Samsung T7 series;

    ➂ Features aluminum and silicone casing for durability, includes dual USB cables, but has mixed aesthetic appeal compared to rivals.

    SK HynixSSDUSB

June 20

  • Kioxia CD9P Released for a 61.44TB PCIe Gen5 NVMe SSD Era

    ➀ Kioxia launched the CD9P series PCIe Gen5 NVMe SSDs, offering capacities up to 61.44TB in 2.5-inch and E3.S form factors, targeting high-performance and high-capacity storage needs in data centers;

    ➁ The series includes CD9P-V (3 DWPD, mixed-use) with up to 12.8TB and CD9P-R (1 DWPD, read-optimized) up to 61.44TB, prioritizing performance scalability for large-scale storage;

    ➂ Positioned as a single-port solution for scale-out architectures, the CD9P complements Kioxia's dual-port CM9 Gen5 SSDs, emphasizing capacity expansion and enterprise storage evolution.

    KioxiaPCIe Gen5SSD

June 19

  • Xsight Labs E1 DPU Offers Up to 64 Arm Neoverse N2 Cores and 2x 400Gbps Networking

    ➀ Xsight Labs introduced its E1 DPU featuring up to 64 Arm Neoverse N2 cores, dual 400Gbps networking interfaces, and 32 PCIe Gen5 lanes;

    ➁ The DPU integrates DDR5 memory, hardware accelerators for crypto and NVMe-oF, and supports flexible deployment models like NVMe SSD aggregation and storage offloading;

    ➂ It competes with NVIDIA BlueField-3 DPU by offering Arm SystemReady-certified Linux compatibility and scalability for cloud data centers.

    ArmDPU

June 18

  • New AMD Ryzen Threadripper 9000 Series and AMD Radeon AI Pro R9700 Series Details

    ① AMD unveiled the Ryzen Threadripper 9000 series, featuring Zen 5 architecture, up to 96 cores, 8-channel DDR5-6400 support, and backward compatibility with existing sTR5 platforms, delivering 15-25% performance gains.

    ② The Radeon AI Pro R9700 GPU utilizes RDNA 4 architecture with 32GB GDDR6, 71 TOPS AI performance, and FP8 support, targeting AI-accelerated workstation workflows.

    ③ AMD reinforces its workstation leadership against Intel’s aging Xeon W lineup, while the R9700 competes with both NVIDIA GPUs and AMD’s own Ryzen AI Max mobile solutions.

    AIAMD

June 16

  • PCI-SIG Releases PCIe 7.0 Specifications and Optical PCIe

    ➀ PCI-SIG released PCIe 7.0 specifications, delivering 128 GT/s raw bit rate and up to 512GB/s bidirectional bandwidth in x16 slots using PAM4 signaling and flit-based encoding;

    ➁ A new optical PCIe interconnect solution (PCIe 6.4/7.0) was introduced, enabling extended reach and multiplexing for applications like NVMe storage pools and CXL memory shelves in hyperscale AI data centers;

    ➂ PCIe Gen8 development is already underway, with adoption expected to accelerate in high-performance computing and AI-driven infrastructure.

    AI ChipHPCPCIe

June 15

  • This is the AMD Instinct MI350

    ➀ AMD announced the Instinct MI350 series, currently shipping to customers, featuring HBM3E memory and OAM packaging;

    ➁ The MI350 utilizes eight HBM3E stacks and two I/O dies, compatible with the OCP UBB platform for scalable 8-GPU configurations;

    ➂ Both air-cooled (MI350X) and liquid-cooled (MI355X) variants are available, promising higher performance and memory density compared to the previous MI300X generation.

    AMDGPUHBM

June 12

  • Google has New NVIDIA RTX Pro 6000 8-GPU Instances with AMD EPYC 9005 CPUs

    ➀ Google Cloud announces G4 VMs with dual AMD EPYC 9005 Turin CPUs and eight NVIDIA RTX Pro 6000 GPUs, delivering 768GB GDDR7 memory and 384 vCPUs;

    ➁ The instances feature 400Gbps network bandwidth, NVIDIA MIG partitioning for up to 32 GPU partitions, and 1.4TB system memory;

    ➂ Targets cost-sensitive workloads with GDDR7 instead of HBM3E, offering flexibility for AI, rendering, and VDI applications.

    AMDGoogleNVIDIA

June 11

  • Micron Begins Shipping HBM4 Memory for Next-Gen AI

    ➀ Micron has commenced shipments of HBM4 memory, delivering 2.0TB/s per stack with a 2048-bit interface, a 60% performance boost over HBM3E;

    ➁ Initial 36GB stacks target next-gen AI accelerators, built on Micron's 1-beta process with advanced memory testing (MBIST) for reliability;

    ➂ Full production ramp is planned for 2026, aligning with next-gen AI hardware releases, while future designs may combine HBM with LPDDR for expanded memory capacity.

    AIHBMMicron

June 8

  • SanDisk Extreme PRO Portable SSD 4TB External USB SSD Review

    ➀ The SanDisk Extreme PRO Portable SSD 4TB is a USB 3.2 Gen 2 x2 external drive offering high-speed storage, competing with models like the Samsung T9 and Crucial X10 Pro;

    ➁ The drive features a rugged design with 3-meter drop protection and IP65 water/dust resistance, along with a large 4TB capacity clearly labeled for easy identification;

    ➂ It includes both USB-C to USB-C and USB-C to USB-A cables, supporting 20Gbps transfer speeds for real-world performance advantages.

    SSDUSB

June 6

  • Broadcom Tomahawk 6 102.4T 64-port 1.6TbE Switches at Computex 2025

    ➀ Broadcom unveiled its Tomahawk 6 series switches at Computex 2025, featuring 102.4T bandwidth and 64-port 1.6TbE configurations.

    ➁ Delta and Wistron/Wiwynn showcased prototype switches, with Delta's 3U design focusing on airflow cooling and Wistron/Wiwynn's 4U model emphasizing enhanced thermal management.

    ➂ Limited photography was permitted due to embargo constraints, but more Tomahawk 6-based hardware is expected to emerge following the official announcement.

    BroadcomNetworking

June 5

  • MLPerf Training v5.0 is Out

    ➀ MLPerf Training v5.0 results highlight NVIDIA's dominance with Blackwell and Hopper architectures, supported by submissions from Dell and Supermicro;

    ➁ AMD showcases improved performance of Instinct MI325X, claiming 30% generational gains over MI300X and competitive results against NVIDIA H200 in Llama2 70B training;

    ➂ Limited participation from other vendors noted, with Google's Trillium system appearing and a focus on accelerated hardware-software optimization trends.

    AMDDellNVIDIA

June 4

  • Broadcom Tomahawk 6 Launched for 1.6TbE Generation

    ➀ Broadcom launched the Tomahawk 6 series switch ASICs, supporting 102.4Tbps bandwidth and up to 64 ports of 1.6TbE, marking the transition to terabit Ethernet;

    ➁ The chips enable both scale-up configurations (e.g., connecting HBM4 memory across 512 XPUs) and scale-out architectures for AI clusters, reducing network tiers and latency;

    ➂ Features like Global Load Balancing 2.0, Cognitive Routing, and enhanced telemetry optimize data flow and cost efficiency, positioning Broadcom as a key player amid competition with NVIDIA in AI infrastructure.

    AIBroadcomHPC
  • Cornelis CN5000 400Gbps Omni-Path Launched with UltraEthernet Roadmap

    ➀ Cornelis Networks launched the CN5000 series, a 400Gbps Omni-Path solution targeting AI and HPC markets, offering liquid/air-cooled director-class switches and low-latency SuperNICs;

    ➁ The new architecture emphasizes zero-congestion networking to optimize GPU utilization in AI clusters, with production availability starting Q3 2025;

    ➂ The company revealed a roadmap integrating UltraEthernet by 2026-2027, positioning itself as an alternative to NVIDIA's InfiniBand and challenging Broadcom/Marvell in Ethernet switching.

    HPCNetworking

June 2

  • ASRock Rack EPYC4000D4U AMD EPYC 4005 Motherboard Review

    ➀ The ASRock Rack EPYC4000D4U is an mATX motherboard designed for AMD's socket AM5 CPUs, bridging server-class features to compact builds with support for up to 192GB DDR5 ECC memory and PCIe Gen5 slots.

    ➀ It features an ASPEED AST2600 BMC for IPMI management, dual Intel i210-at NICs, and eliminates traditional chipsets by using ASMedia controllers for SATA ports, reducing power and cost and aligning with modern server designs.

    ➁ The motherboard prioritizes EPYC CPUs for server workloads (e.g., 170W liquid-cooled EPYC 4565P) but supports Ryzen CPUs, though with reduced PCIe Gen5 speed on M.2 slots for Ryzen 8000 series processors.

    AMDASRock RackEPYC

June 1

  • GIGABYTE G383-R80-AAP1 AMD Instinct MI300A Server Review

    ➀ The GIGABYTE G383-R80-AAP1 is a 3U server featuring AMD's Instinct MI300A APUs, combining CPU and GPU cores with HBM3 memory for high-performance AI and HPC workloads;

    ➀ The server houses four MI300A APUs, eight 2.5" NVMe drives, and eight PCIe Gen5 slots, along with four 3kW PSUs for efficient power delivery;

    ➂ Designed for supercomputing tasks, it emphasizes cooling efficiency and scalability, targeting applications like the AMD-powered El Capitan supercomputer.

    AMDGigabyteHPC