Last year, the Intel® Data Center GPU Flex Series introduced customers to a flexible, general-purpose graphics processing unit (GPU) for the data center and the intelligent visual cloud. Since its introduction, Intel has expanded the Intel Flex Series GPU’s production-level software capabilities, including new support for Windows cloud gaming, AI inference and digital content creation.
As Flex Series GPU adoption grows in the market, customers, solution providers and developers are utilizing the hardware’s capabilities across many real-world scenarios:
- Cloud service providers are deploying instances of Windows cloud gaming alongside streaming and media acceleration.
- Media studios are testing and deploying Flex Series GPUs for high-density streaming and transcoding.
- Digital content creators are using the platform for fast, real-time rendering using ray tracing hardware acceleration.
In particular, the addition of Windows cloud gaming allows the Flex Series GPU to address the growing gaming market. The rapid growth of the global cloud gaming market is estimated to reach a targeted market value of about $13.3 billion by 2028, with a projected compound annual growth rate (CAGR) of approximately 42.5% from 2022 to 2028.1 Game service providers must continually innovate to deliver first-rate playing experiences to their subscribers while operating the most efficient infrastructure possible.
Intel’s new software capabilities for the Intel Flex Series GPU enable customers to realize new capabilities and gains across real-world workloads coupled with 3rd Gen and 4th Gen Intel® Xeon® Scalable processors.
Windows Cloud Gaming: Intel provides a cloud gaming reference stack for Windows to show how to unleash Intel Flex Series for remote gaming, including how to enumerate and retarget multiple game titles to run concurrently on each of the multiple Flex Series adapters and two types of virtual displays. Intel Flex Series is optimized for DirectX 9, 11 and 12 render target capture and desktop capture.
Gamestream, a cloud game streaming solution, allows users to stream and play games on devices with high-end graphics across the entertainment, hospitality, media and telecom markets.
Olivier Lebigot, CTO at Gamestream, said, “One of our key challenges in cloud gaming is to find the right GPU to increase the overall concurrent users (CCU) number per server, while delivering the best end-user experience. We have been pleased to find that Intel Data Center GPU Flex Series 170 offers a high-density, low-power solution. During our tests, our reference CCU was improved by nearly 19% compared to our current hardware solution. This is a real asset for massive scaling.2”
Artificial Intelligence Workloads: Intel has expanded its Flex Series capabilities for AI, including workloads for smart city, library indexing and compliance, AI-guided video enhancement, intelligent traffic management, smart buildings and factories, and retail applications. Intel provides guidance for developers on how to get started on its Flex Series GPU software page and in its developer catalog.
- AI Inference: Through the Intel® AI Analytics Toolkit, Intel Flex Series GPU supports most common AI frameworks, including TensorFlow and PyTorch, among others. Intel has validated more than 100 AI inference models.
- AI Visual Inference: The Intel software stack for visual inference now includes the open source GStreamer multimedia framework, the Intel® OpenVINO™ toolkit inference engine and open source sample implementation of media analytics framework for creating complex visual inference and analytics pipelines.
Autonomous Driving: Autonomous driving (AD) systems are a focal point of development for numerous companies, but training autonomous vehicles requires comprehensive testing for complex, challenging situations on roads, in addition to significant costs. Autonomous driving testing via simulation helps fill this gap.
A prominent open source solution for supporting the advancement, training and validation of AD systems is CARLA, which utilizes the Unreal Engine 4. In the CARLA benchmark performance – particularly in single- and multiple-sensor scenarios – the Intel Flex Series GPU showcases better capabilities compared to data center competitors.
- In a single 1080p sensor scenario, a single Intel® Data Center Flex 170 GPU achieves 56 frames per second (fps), which is 16% faster than the Nvidia A10G (48 fps), 40 percent faster than the Nvidia GPU Tesla T4 (40 fps) and 60 percent faster than the AMD Radeon Pro V520 MxGPU (35 fps).3
- The performance of the Intel Data Center Flex 170 GPU scales accordingly for four 1080p sensors, reaching 23 fps while the Nvidia A10G achieves 19 fps.3
Digital Content Creation: Real-time rendering is delivered via ray tracing hardware acceleration on the Intel Flex Series through Intel® Embree. AI-based denoising can be completed in milliseconds using the latest Intel® Open Image Denoise. Further productivity can be gained through the use of a single SYCL codebase for CPU/GPU rendering through oneAPI.
Flex Series offers customers a comprehensive graphics solution, an open and full software stack, no licensing fees, and a unified programming model for CPUs and GPUs for performance and productivity via oneAPI. View the software stack and development tools supporting these workloads at Software for Intel Data Center GPU Flex Series.
System Availability: Original equipment manufacturers (OEMs) are bringing more than 70 designs to market with Flex Series GPUs, with more than 40 designs expected by mid-year. OEMs that are delivering enabled designs with Intel Flex Series GPUs include Cisco, Dell, H3C, HPE, IEIT, Lenovo, Nettrix, PowerLeader, Supermicro and xFusion. For the latest updates on system designs, availability and product updates, visit the Intel Data Center GPU Flex Series page on Intel.com.
1 Facts & Factors, October 6, 2022. “Demand for Global Cloud Gaming Market Size to Surpass USD 13,333.41 Million by 2028, Exhibit a CAGR of 42.50% | Cloud Gaming Industry Trends, Share, Value, Analysis & Forecast Report by Facts & Factors.” https://www.globenewswire.com/en/news-release/2022/10/06/2529877/0/en/Demand-for-Global-Cloud-Gaming-Market-Size-to-Surpass-USD-13-333-41-Million-by-2028-Exhibit-a-CAGR-of-42-50-Cloud-Gaming-Industry-Trends-Share-Value-Analysis-Forecast-Report-by-Fac.html.
2 Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
3 Intel Setup:
Hardware Config: 32x Intel(R) Xeon(R) Gold 6338N CPU 2.2Ghz, HT On, Turbo On, ucode 0xd00033, 512 GB DDR4-3200, 2x Intel Data Center GPU Flex 170 (utilized only one for the comparison), GPU Driver IFWI_22WW05 version 30.0.101.2106
System SW Config: Microsoft Windows Server 2019 Standard, Version 10.0.17763 Build 17763, BIOS American Megatrends Inc. 06.00.04, 4/12/2022
AWS Nvidia Setup (g5.8xlarge instance type):
Hardware Config: 16x AMD EPYC 7R32 CPU 2.8 Ghz, HT On, Turbo On, 128 GB DDR4-3200, Nvidia A10G, GPU Driver 531.41 (CUDA version 12.1) - 31.0.15.3141
System SW Config: Microsoft Windows Server 2019 Datacenter, Version 10.0.17763 Build 17763, BIOS Version Amazon EC2 1.0, 10/16/2017
Identical Application Setup: CARLA-0.9.14 (commit id f14acb257ebf44c302b225b02080ac5f0eedcf7f) with Unreal Engine 4.26 (CARLA's fork commit id: 1f66bc130e52b50067448f220b1dbea9d9b2d14a), Visual Studio 2019 Community Edition (x64 Visual C++ Toolset, .Net framework 4.6.2, Windows 8.1 SDK), Python 3.8.10 x64, WDDM 2.5, DirectX 12
Results may vary.