Refurbished Supermicro GPU Servers
A Supermicro GPU server on the NVIDIA HGX 8-GPU platform carries eight SXM GPUs on one baseboard, linked by NVLink and NVSwitch, in an 8U chassis that Supermicro lists for AI and deep learning training, conversational AI and HPC. Supermicro builds this design with HGX H100 or HGX H200 boards and with two Intel Xeon or two AMD EPYC processors. Enterasource tests each refurbished Supermicro system in our lab before dispatch and backs it with a 2-year warranty, with freight delivery to the lower 48 states at no charge. Request a quote for any system shown without a price.
-
Supermicro SYS-821GE-TNHR 2x Intel 8470 8x NVIDIA H100 80GB 1TB DDR5 8x 3.84TB NVMeAvailability: Available by QuoteManufacturer: SuperMicro Condition: Refurbished 2 Year WarrantyRequest Pricing - Supermicro SYS-821GE-TNHR 2x Intel 8462Y+ 8x H100 80GB 1TB DDR5 3x NVMeAvailability: Available by QuoteManufacturer: SuperMicro Condition: Refurbished 2 Year WarrantyRequest Pricing
- Supermicro AS-8125GS-TNHR 2x AMD 9534 8x H100 80GB 1.5TB DDR5 2x 3.8TB NVMeAvailability: Available by QuoteManufacturer: SuperMicro Condition: Refurbished 2 Year WarrantyRequest Pricing
- Supermicro SYS-821GE-TNHR 2x Intel 8468 8x H100 80GB 2TB DDR5 4x 7.68TBAvailability: Available by QuoteManufacturer: SuperMicro Condition: Refurbished 2 Year WarrantyRequest Pricing
The Supermicro HGX 8-GPU platform
Supermicro's SYS-821GE-TNHR shows what the 8U HGX design holds. Its product page lists a pair of 4th or 5th Gen Xeon Scalable CPUs, an HGX H100 8-GPU board with 80 GB per GPU or an HGX H200 8-GPU board with 141 GB, 32 DIMM slots taking up to 8 TB of DDR5, and 12 front 2.5-inch NVMe bays plus 3 SATA bays. It has eight low-profile and two full-height PCIe 5.0 x16 slots, and its power comes from six 3,000 W Titanium supplies, four active and two redundant, with up to ten heavy-duty fans. Supermicro's AMD version pairs the same 8U HGX layout with two EPYC 9004 processors and 24 DIMM slots. NVMe bay layouts across brands are compared on our NVMe servers page.
HGX H100 or HGX H200
The two HGX boards differ in GPU memory and memory bandwidth, while NVLink bandwidth and maximum power match. NVIDIA's figures per GPU:
| NVIDIA SXM GPU | H100 SXM | H200 SXM |
|---|---|---|
| GPU memory | 80 GB | 141 GB |
| Memory bandwidth | 3.35 TB/s | 4.8 TB/s |
| NVLink bandwidth | 900 GB/s | 900 GB/s |
| Max power | Up to 700 W, configurable | Up to 700 W, configurable |
| GPU memory on an 8-GPU board | 640 GB | 1,128 GB |
Supermicro HGX or a four-GPU PCIe server
Choose by GPU count and interconnect. An 8U Supermicro HGX system joins eight SXM GPUs through NVSwitch, each drawing up to 700 W, while Dell's 2U PowerEdge R760xa holds at most four double-width PCIe GPUs of 400 W, and PCIe cards that support NVLink connect through bridges rather than a switch. For one to four PCIe GPUs in less rack space, compare Dell and HPE PCIe systems on our GPU servers page; for eight GPUs on one NVLink fabric, an HGX system is the class to buy.
Supermicro GPU server questions
- What GPUs are supported in Supermicro servers?
- It depends on the system's GPU design. HGX systems such as Supermicro's SYS-821GE-TNHR take eight NVIDIA SXM GPUs on one baseboard, either an HGX H100 8-GPU board with 80 GB per GPU or an HGX H200 8-GPU board with 141 GB. In PCIe GPU servers the GPUs are cards, which NVIDIA rates at 70 W for the low-profile T4, 165 W for the dual-slot A30 and 350 W for the dual-slot L40S, and the chassis power and airflow decide how many fit.
- What warranty comes with a refurbished Supermicro server?
- Every refurbished Supermicro server from Enterasource includes a 2-year Enterasource warranty at no extra charge, the same term that covers every server we sell. Each system passes our lab testing before dispatch, and we ship it free to the lower 48 states, freight included. The full terms are on the Enterasource warranty and support page.
Read the full warranty and support terms, compare all refurbished servers or GPU-capable Dell PowerEdge systems, or buy single cards under server GPUs.
Choosing an AI system? Our AI GPU server buying guide walks through VRAM sizing and multi-GPU interconnects, and sizing on-prem LLM hardware covers GPU memory, platform and power for local models.
