Vi tager imod kryptovaluta — så den krypto, du ejer, kan købe rigtig hardware.Betal med krypto, brug det på rigtig hardware. Gratis, forsikret levering til hele verden ved køb over 399 $. Neutrale, umærkede papkasser, sendt fra Tyskland.Gratis, forsikret levering over 399 $. Din kurs låses i 30 minutter, så snart kassen åbner.Kurs låst i 30 minutter. Hver vare er forseglet, har kontrolleret serienummer og er dækket af 24 måneders garanti.Forseglet, 24 måneders garanti.

Prisvagt Log ind
Silver 1U rack server front with eight drive bays and network ports

Supermicro

Supermicro ARS-111GL-NHR-OTO-38 1U rack server GH200 72-core 480GB LPDDR5X 3.84TB + 960GB storage 2x10G 2x25G

SKU CH-CTJM-YMSCC MPN ARS-111GL-NHR-OTO-38

1U rack server with NVIDIA GH200 Grace Hopper Superchip, 72-core CPU, 480GB LPDDR5X, 96GB HBM3 GPU memory, 3.84TB and 960GB NVMe storage, dual 10G and dual 25G networking

  • WorkloadAI/ML/HPC Server
  • TypeRack
  • Rack Height1U
  • CPU TypeNVIDIA Grace Hopper
  • CPU SpeedGH200
  • Number of CPU Cores72-Core Processor
  • 72-core Grace CPU powers AI and HPC workloads
  • 480GB LPDDR5X memory feeds the GPU without bottlenecks
  • 84TB plus 960GB NVMe storage keeps datasets local
  • Dual 10G and dual 25G networking moves data fast
  • 1U air-cooled design fits dense rack deployments

GH200 Superchip for AI/HPC workloads

The Supermicro ARS-111GL-NHR-OTO-38 is a 1U rack server built around the NVIDIA GH200 Grace Hopper Superchip. It combines a 72-core Grace CPU with an H100 Tensor Core GPU and 96GB HBM3 video memory. The system targets AI, machine learning and high-performance computing clusters that need unified CPU-GPU memory access.

Unified LPDDR5X memory and E1.S storage limits

The platform provides 480GB of LPDDR5X memory shared between CPU and GPU, removing traditional PCIe bottlenecks for large model workloads. Storage uses 8 front hot-swap E1.S NVMe bays populated with 3.84TB capacity, plus a 960GB M.2 boot drive. Networking is fixed at two 10GbE RJ45 ports and two 25GbE SFP28 ports.

Air-cooled 1U form factor and hot-swap bays

The server fits a standard 1U rack slot and relies on air cooling rated for up to 1000W TDP processors. Eight E1.S bays allow drive replacement without downtime, while two M.2 slots remain available for expansion. Operating temperature range is 10°C to 35°C, so adequate rack airflow must be ensured before deployment.

Fordele

  • 72-core Grace CPU powers AI and HPC workloads
  • 480GB LPDDR5X memory feeds the GPU without bottlenecks
  • 84TB plus 960GB NVMe storage keeps datasets local
  • Dual 10G and dual 25G networking moves data fast
  • 1U air-cooled design fits dense rack deployments

Specifikationer

BrandSUPERMICRO
ModelARS-111GL-NHR-OTO-38
WorkloadAI/ML/HPC Server
TypeRack
Rack Height1U
CPU TypeNVIDIA Grace Hopper
CPU SpeedGH200
Number of CPU Cores72-Core Processor
CPU FeaturesSupports up to 1000W TDP CPUs (Air Cooled)
# of Processors Installed1
ChipsetIntel Cyclone 10 GX FPGA
Installed Memory Size480GB
Memory TypeLPDDR5X
SSD (Installed)3.84TB + 960GB
GPU TypeNVIDIA H100 Tensor Core GPU on GH200 Grace Hopper Superchip (Air-cooled)
Video Memory96GB HBM3
Ethernet Controller1 x Intel X550 2-Port 10 Gigabit (10GbE) PCI-E 3.0 Ethernet Card
1 x Standard NVIDIA 25GbE with 2 SFP28 ports
Expansion Slots2 M.2 NVMe slot(s) (M-key)
Expansion Bays8 front hot-swap E1.S NVMe drive bay(s)
TemperatureOperating Temperature: 10°C ~ 35°C (50°F ~ 95°F)
Non-operating Temperature: -40°C to 60°C (-40°F to 140°F)
HumidityOperating Relative Humidity: 8% to 90% (non-condensing)
Non-operating Relative Humidity: 5% to 95% (non-condensing)
Dimensions1.75" x 17.33" x 37"
Weight48.5 lbs.
Forsendelsesvægt29.71 kg
Pakkestørrelse1219 × 711 × 241 mm

Spørgsmål om denne vare

How do I decide between the 3.84TB E1.S SSD and the 960GB M.2 SSD for my workload?

The 3.84TB E1.S drive occupies one of the eight front hot-swap bays and is suited for large datasets, while the 960GB M.2 drive uses an internal M-key slot for OS or cache.

What does the 480GB LPDDR5X memory mean for AI training jobs?

The 480GB of LPDDR5X is unified memory shared between the 72-core Grace CPU and the H100 GPU, letting models up to that size run without host-device copies.

What real throughput do the 10GbE and 25GbE ports deliver?

The Intel X550 dual-port 10GbE card runs at PCIe 3.0 x4, and the NVIDIA dual-port 25GbE adapter uses SFP28; each port caps at its rated line speed.

Også på denne hylde

Kunderne har også sammenlignet disse

Samme hylde, samme kasse — otte mønter og 30 minutters kurslås.