INTEL H3C XG310 Server GPU Professional Computing Solution

Product status: Official | Last Update: 2020-11-11 | Report Error
Overview
Manufacturer
INTEL
Original Series
Xe Pro
Announcement Date
November 11th, 2020
Release Date
November 11th, 2020
Graphics Processing Unit
GPU Model
4× DG1
Architecture
Intel Xe-LP
Fabrication Process
10 nm Intel SuperFin
Die Size
4× 72 mm2
Shading Units
4× 768 (3072)
Execution Units
4× 96 (384)
Subslices
4× 12 (48)
TMUs
4× 48 (192)
ROPs
4× 24 (96)
Clocks
Base Clock
900 MHz
Boost Clock
1100 MHz
Memory Clock
2133 MHz
Effective Memory Clock
4266 Mbps
Memory Configuration
Memory Size
4× 8192 (32768) MB
Memory Type
LPDDR4
Memory Bus Width
4× 128 (512)-bit
Memory Bandwidth
4× 68.3 (273.2) GB/s
Physical
Interface
PCI-Express 3.0 x16
Height
1-slot
Power Connectors
1× 8-pin
TDP/TBP
150 W
Recommended PSU
450 W
API Support
DirectX
-
Vulkan
1.2
OpenGL
4.6
OpenCL
2.1

Performance
Pixel Fillrate
105.6 GPixels/s
Texture Fillrate
211.2 GTexel/s
FP32 Performance
6.8 TFLOPS
FP32 Perf. per Watt
45.1 GFLOPS/W
FP32 Perf. per mm2
23.5 GFLOPS/mm2




 ModelCoresBoost ClockMemory ClockMemory Config.
Thumbnail
INTEL H3C XG310 Server GPU
 
3072
 
1100 MHz
 
4.3 Gbps
 
128 GB LPD4 128b
Intel Arctic Sound G4 TBC TBC TBC TBC
Thumbnail
Intel Xe Ponte Vecchio TBC TBC TBC TBC
Intel Arctic Sound G2 TBC TBC TBC TBC
Intel Arctic Sound G1 TBC TBC TBC TBC
 ModelCoresBoost ClockMemory ClockMemory Config.
Thumbnail
INTEL H3C XG310 Server GPU
 
3072
 
1100 MHz
 
4.3 GB/s
 
128 GB LPD4 128b
Thumbnail
Intel Iris Xe MAX
 
768
 
1650 MHz
 
4.3 GB/s
 
4 GB LPD4X 128b

Cloud service providers (CSPs) and communication service providers (CoSPs) are looking for ways to capture business benefits from the explosive growth of high-intensity media and gaming. They are striving to build infrastructure that increases the density of streams supported per server, while ensuring high quality and availability.

The Intel® Server GPU, based on Intel’s Xe architecture, offers these benefits to service providers, while keeping total cost of ownership (TCO) low. This discrete graphics solution is purpose-built for the needs of differentiating services in the visual cloud era and can be paired with Intel Xeon® Scalable processors to help providers rapidly scale out new offerings to meet customer demands.

The GPU can help reduce CapEx in terms of equipment and facilities costs, and it also includes architectural efficiencies that reduce power consumption to help lower OpEx. Additionally, by enabling a standards-based hardware and software development environment, it can provide further advantages by unifying the graphics landscape across usages and performance levels.

Building on years of experience with data center graphics—including the Intel Xeon processor E3 v5 with Intel Iris™ Pro Graphics and Intel Visual Compute Accelerator 2—Intel delivers a new generation of architecture benefits to service providers. The new Intel Server GPU can support around 20 Android games per GPU, depending on the specific game title and server configuration.1 With 96 independent execution units per GPU, large-scale memory and I/O resources, and a highly efficient pipeline, this product is purpose-built to accelerate rendering and media processing for a variety of visual cloud workloads.

Based on a low-power system-onchip (SoC) design, the Intel Server GPU offers flexibility and scalability to address varied system requirements. The H3C XG310 graphics card is one configuration that can help providers increase density. This card is a single-slot, full height, 3/4 length, PCIe Gen3 x16 form factor and is powered by four Intel Server GPUs. It includes 32 GB of low power DDR4 memory at a TDP of 150 W and is passively cooled, allowing airflow in both directions for flexible server chassis design. With support for up to four XG310 PCIe cards (sixteen GPUs) per system, the Intel Server GPU platform is a truly scalable solution offering density, scalability, and efficiency that help address data center needs. Cloud & Communication Service Providers Graphics

System Spec
dimension¾ longth, 28mm X 34mm)
Package TypeFCBGA
TDP23W
OpenGL4.1
Internal Thermal SensorYes
L3 cache size16MB
TDP150W
GPU chip4
PCIe adapterGen3 x16
thermalPassive
GPU memory128-bit width,8GB per GPU chip,LPDDR4,68.25GB/s

2133MHz/ 4267MT/s(max)

Operating Frequency4267MT/s
GPU memory capacity8GB per GPU, total 32GB
OSCentOS,Debian OS
SWHost BIOS –Default Production BIOS will need Upgrading

SG1-M IFWI

Telemetry – MCU on card