Need to do 560 Teraflops in a Single Rack..?
The D52G-4U also known as D5G is the world's fastest GPU server. Hyperconverged: 8 * NVMe with up to 16 * GP-GPU the D5G does 56 Double Precision Teraflops in 4U. It is ideal for: Big Data, Deep learning, Artificial Intelligence and Inferencing. D52G boasts: NVLink™ support up to 300GB/s GPU to GPU communication | 10x dual-width 300 Watt GPU | 16x single-width 75 Watt GPU support | Diversify GPU topology to Conquer Any Type of Parallel Computing Workload | Up to 4x100Gb/s High Bandwidth RDMA-enabled Network to Scale Out with Efficiency | 8x NVMe Storage to Accelerate Deep Learning.
Deep learning with up to 16 GPU featuring with NvlinkTM
With NVLinkTM to support up to 300GB/s GPU to GPU communication, D52G-4U can shorten model-training time-frame, which is 40x faster than conventional server on processing 408GB Chest X-ray images.
10x dual-width 300 watt GPU or 16x single-width 75 Watt GPU support
As an purpose-built system for Artificial Intelligent (AI) and High-Performance Computing(HPC) workloads, QuantaGrid D52G-4U can deliver up to 896 tensor Tflops to training deep learning model with eight* NVIDIA® Tesla V100 dual-width 10.5 inch or provides up to 293 GOPS/watt of peak INT8 performance to do inferencing with sixteen NVIDIA® Tesla P4 and 2-socket Intel® Xeon® Scalable processor; up to 56 double precision Tflops computing power can accelerate HPC workloads such as Oil & Gas, bioinformatics, Mechanical Engineering. On top of superior computing power, D52G is of 2x100Gb/s high bandwidth low-latency networking to expedite communication among different GPU nodes
Diversify GPU Topology to Conquer Any Type of Parallel Computing Workload
The QuantaGrid D52G provides multiple GPU topology on the same baseboard tray to meeting your different use case.
High-Bandwidth, Low-latency Networking Between GPU Nodes
The QuantaGrid D52G-4U have additional two/four PCIe Gen3x16 LP-MD2 slot to provide optional two/four 100Gb/s low-latency infiniband or Intel®Omni Path options to do GPUDirect in the server or RDMA between different GPU nodes.
NVMe SSD Support to Accelerate Deep Learning
The D52G-4U support max. 8x NVMe SSD, which can accelerate both training and inferencing with Fast I/O data reading as deep learning is a data-driven algorithm.
Intel®Xeon® Processor Scalable Family
|Max. TDP Support||
|Number of Processors||
|W x H x D (inch)||
17.5 x 7x 34.6
|W x H x D (mm)||
444x 177.8x 880
|SKU - #1||
(24) 2.5" hot-plug SATA/SAS SSD
Up to 3TB (128Gx24) of memory for RDIMM/LRDIMM
2666 MHz DDR4 RDIMM
64GB, 32GB, 16GB, 8GB RDIMM
(Refer to CCL)
|Expansion Slot - Motherboard||
(1) PCIe Gen3 x 8 OCP 2.0 mezzanine slot (can switch to U.2)
|Expansion Slot - GPGPU Baseboard||
Dedicated (1) GbE management port
Please refer to our Compatible Component List for more information
Quanta LSI® 3008 12Gb/s SAS mezzanine, RAID 0,1,10
3+1 high efficiency hot-plug 1600W PSU, 80 Plus Platinum (200V-240V)
(2) SATADOM (option)
(2) 2280/2210 SATA/PCIe M.2 (option)
(4) dual rotor fans (7+1 redundant)
Integrated ASPEED AST2500 8MB DDR4 video memory
(2) USB 3.0 ports
Operating temperature*: 5°C to 30°C (41°F to 86°F)
|Weight (Max. Configuration)|
|Weight (Max. Configuration)||61.5 (Kg)|