我們接受加密貨幣付款 — 讓您持有的幣種也能買到真正的硬體設備。以加密貨幣付款，直接購入真正的硬體設備。 全球訂單滿 US$399，即享免費保險運送。簡約外箱，不印任何品牌標誌，從德國出貨。訂單滿 US$399，即享免費保險運送。 結帳一開始，您的匯率即鎖定 30 分鐘。匯率鎖定 30 分鐘。 每件商品皆原廠密封、序號經過核對，並享有 24 個月保固。原廠密封，保固 24 個月。

追蹤清單 登入
Silver 5U rackmount barebone server chassis with front drive bays and USB ports

GIGABYTE

GIGABYTE G593-SD0-AAX1 barebone server 5U HGX H100 8-GPU

SKU CH-ZC9P-F5PVR MPN G593-SD0-AAX1

A 5U rackmount barebone server for AI, ML and HPC workloads, supporting dual 4th Gen Intel Xeon processors, 32 DDR5 4800 memory slots and eight NVIDIA HGX H100 SXM5 GPUs with 900GB/s interconnection

  • Form Factor5U Rackmount
  • WorkloadAI/ML/HPC
  • CPU TypeIntel Xeon
  • CPU SocketLGA 4677
  • Installed CPUNo CPU Installed
  • Max Processors2
  • Dual 4th Gen Intel Xeon sockets scale AI training
  • NVIDIA HGX H100 with 8 SXM5 GPUs accelerates inference
  • 900GB/s NVLink interconnect reduces multi-GPU latency
  • 32 DDR5 4800 slots support 3DS RDIMM up to 256GB
  • 8 hot-swap 2.5" Gen5 NVMe/SATA/SAS bays enable fast storage

5U barebone with HGX H100 8-GPU

The GIGABYTE G593-SD0-AAX1 is a 5U rackmount barebone server built around the NVIDIA HGX H100 platform with eight SXM5 GPUs. It supports dual 4th Gen Intel Xeon Scalable processors in LGA 4677 sockets with a TDP ceiling of 350 W each. The Intel C741 north bridge provides eight memory channels per processor across 32 DDR5 4800 DIMM slots.

Suited for AI training and HPC clusters

This platform targets organisations deploying large-scale AI training, inference or high-performance computing workloads that can exploit the 900 GB/s GPU-to-GPU interconnect via NVLink and NVSwitch. It is not suited for buyers who need a general-purpose server with installed CPUs, memory or storage, or for teams that lack the infrastructure to power and cool an 8-GPU HGX system.

Eight SXM5 GPUs with 900 GB/s interconnect

The defining specification is the NVIDIA HGX H100 baseboard with eight SXM5 GPUs linked by NVLink and NVSwitch at up to 900 GB/s. This GPU fabric decides whether the chassis matches the workload; all other components scale around that fixed configuration.

商品特色

  • Dual 4th Gen Intel Xeon sockets scale AI training
  • NVIDIA HGX H100 with 8 SXM5 GPUs accelerates inference
  • 900GB/s NVLink interconnect reduces multi-GPU latency
  • 32 DDR5 4800 slots support 3DS RDIMM up to 256GB
  • 8 hot-swap 2.5" Gen5 NVMe/SATA/SAS bays enable fast storage

商品規格

BrandGIGABYTE
ModelG593-SD0
Form Factor5U Rackmount
WorkloadAI/ML/HPC
CPU TypeIntel Xeon
CPU SocketLGA 4677
Installed CPUNo CPU Installed
Max Processors2
CPU FeaturesDual processor, CPU TDP up to 350W
North BridgeIntel C741
Memory Slots32 x 288Pin
Installed Memory SizeNo Memory Installed
Max Memory SupportedRDIMM modules up to 96GB supported
3DS RDIMM modules up to 256GB supported
Memory Type SupportedDDR5 4800
Channel Supported8 Channel
Installed HDDNo Storage Installed
GPU/VPUNVIDIA HGX H100 with 8 x SXM5 GPUs
PCI Express8 x PCIe x16 (Gen5 x16) low-profile slots, from PEX89104
2 x PCIe x16 (Gen5 x16) low-profile slots, from CPU_0
2 x PCIe x16 (Gen5 x16) low-profile slots, from CPU_1
1 x PCIe x16 (Gen4 x16) low-profile slot, from CPU_0
First LANIntel X710-AT2
Max. LAN Speed10 Gigabit
Hot-Swap Bays8 x 2.5" Gen5 NVMe/SATA/SAS hot-swappable bays, from PEX89104
Front USB2 x USB 3.2 Gen1
RJ452
Other Ports1 x TPM header
1 x VROC connector
1 x Front panel header
Max. Power4+2 3000W (240V) 80 PLUS Titanium redundant power supplies
TemperatureOperating temperature: 10°C to 35°C
Non-operating temperature: -40°C to 60°C
HumidityOperating humidity: 8%-80% (non-condensing)
Non-operating humidity: 20%-95% (non-condensing)
出貨重量30.00 kg
包裝尺寸940 × 447 × 222 mm

商品問答

Does the G593-SD0-AAX1 barebone include CPUs, memory or storage, or must I buy them separately?

The barebone ships with no CPUs, no DDR5 memory and no storage installed. You must supply up to two LGA 4677 Xeon processors, up to 32 DDR5 4800 RDIMM or 3DS RDIMM modules, and drives for the eight 2.5-inch NVMe/SATA/SAS hot-swap bays.

Which DDR5 capacity should I choose — 96 GB RDIMM or 256 GB 3DS RDIMM — for the 32 slots?

Choose 96 GB RDIMM modules for standard density or 256 GB 3DS RDIMM modules for maximum capacity; both run at DDR5 4800 across the 8-channel per processor architecture.

What does the 900 GB/s GPU-to-GPU interconnect mean for training workloads on the eight SXM5 GPUs?

The 900 GB/s NVLink and NVSwitch fabric lets all eight H100 GPUs exchange model data directly at full bandwidth, reducing multi-GPU scaling overhead for large-language-model and HPC training.

同類商品

顧客也比較過

同一分類，同一套結帳流程 — 支援八種加密貨幣，匯率鎖定 30 分鐘。