Chinese

ILUVATAR CORE X

ILUVATAR CORE X GPU Tiangai 100

Tiangai general GPU high-end self-developed cloud training chip releases powerful programmability and application versatility through a rich and comprehensive self-developed instruction set, providing industry-leading AI computing power density and energy efficiency ratio. Tiangai series training chips focus on high performance, versatility and flexibility, define and optimize the design from the source, achieve multi-angle technological innovation, and have significant advantages in ecology, computing power, application migration, product reliability, etc., providing Artificial intelligence, general computing and related vertical application industries provide computing power that matches the rapid development of the industry. The performance of this product is predictable. The software and hardware architecture of Tiangai 100 is designed for general computing and artificial intelligence. It is comparable to the software and hardware architecture of mainstream GPU products in the industry. It uses 2.5D COWOS packaging technology and a rich self-developed instruction set to fully support scalars, vector, and tensor operations, providing industry-leading high computing power and high energy efficiency. Development is easy to migrate. Tiangai 100 supports domestic and foreign standardized software and hardware ecology, is compatible with domestic and foreign mainstream frameworks and official operators, commonly used network models and acceleration libraries, and is compatible with mainstream GPU general computing models. Application migration costs are low, time-consuming and does not require Redevelopment. And with wide application coverage, Tiangai 100 focuses on high performance, versatility and flexibility, and supports the industry's cutting-edge algorithms. Currently, more than 200 general computing and artificial intelligence applications have been implemented, and the number continues to increase. It can calmly face future algorithm changes and provide Artificial intelligence, general computing and related vertical application industries provide computing power that matches the rapid development of the industry.

Quick inquiry

Congratulations, the submission was successful. Please be patient and wait for us to contact you.

Back

  • Product quantity

Home > Manufacturer > ILUVATAR CORE X > product details

Merchant qualification:

Shanghai Iluvatar CoreX Semiconductor Co., Ltd. ("Iluvatar CoreX" for short) is China's leading provider of general-purpose GPU high-end chips and super computing systems. Tianshu Zhixin takes "serving the national strategy" as its mission, adheres to the "four orientations", and is committed to developing independently controllable, internationally leading high-performance general-purpose GPU products, exploring the development path of general-purpose GPU catching up and surpassing, accelerating the construction of an independent industrial ecology, and providing Provide high-end computing solutions for the entire industry.

Product Introduction&Application Fields

product details

  • Model: BI-V100

  • Interface: PCleGen0×16 lane, sharing 64GB/s main control bidirectional bandwidth, sharing 64GB/s inter-chip interconnect bandwidth

  • GPU architecture: general purpose GPU

  • Frequency: 1.5GHz

  • Power supply: 8pin power connector

  • Single precision floating point peak computing power (including TCU): 37 TFLOPS

  • Single precision floating point peak computing power (excluding TCU): 15 TFLOPS

  • Semi-precision floating point peak computing power (including TCU): 147 TFLOPS

  • Semi-precise floating point peak computing power (excluding TCU): 37 TFLOPS

  • Integer peak computing power: 295TOPS@INT8 supports INT32 and INT16 calculations

  • Power consumption: Board level power consumption 250W

  • Size: Full length, full height, dual slot PCle card, compatible with mainstream servers

  • Cooling method: passive cooling

  • Package: 2.5DCOWOS package

  • Software ecosystem: Compatible with mainstream general computing software frameworks and supporting mainstream deep learning development frameworks

  • Virtualization: supported

  • Memory: 32GBDRAM (4*8GB) HBM2 

Product specifications

Brand inventory product recommendations

Model
Image
Batch number
Package
Quantity
Warehouse
Description
Specifications
Suppliers
Inquiry

Login

Register

Login
{{codeText}}
Login
{{codeText}}
Submit

Return to login

Close
Subscribe
ITEM
Comparison Clear all