Data Center
Participate in insightful discussions regarding Data Center topics
65 Discussions

Elevating User Experience: The Impact of GPUs on VDI Environments

Glenn_C_Le_Vernois
0 0 861

In a previous blog, I described how Login VSI and Intel worked together to create a new GPU Reference Profile for their Login Enterprise tool to accurately represent today’s workers. These workers regularly utilize Windows applications like Outlook, PowerPoint, etc., but they also need more graphics-heavy apps with high-def video, 3D rendering, and video collaboration.

Using Login Enterprise as the de facto industry standard tool for testing VDI environments and server-based computing, we decided to put the Intel® Data Center Flex Series GPUs to the test to see if and how we could improve the user experience for this new GPU Reference Profile. The cherry on top would be if they did so without compromising on budget.

We are happy to say that we are pleased with the results. Let’s talk about what we did.

Improving User Experience: What Does It Mean?

We classified an improved user experience by higher frame rates and GPU offload while maximizing users. This is demonstrated by the End User Experience (EUX) Score, which represents the performance of any Windows machine (virtual, physical, in the cloud, or on-premises) and ranks it between 0 and 10 as experienced by the virtual user. In the case of VDI systems with shared hardware, the EUX Score will decrease as more users are added.

The second metric we looked at is the Virtual Session Index Maximum (VSIMax), which is the number of users supported.

Maximizing the number of users can hurt system responsiveness. Our testing aimed to maximize users (VSIMax) while maintaining an EUX score of 7.5 or better, which is considered “Very Good.”

What We Tested

Our VMware Horizon tests were run on a 4-node vSAN Cluster with 32 core processors and 2TB system memory per node. Each node was configured with 2 Intel® Data Center GPU Flex 140 to support 1080p (HD) content. Each user was allocated 2 vCPUs and 8 GB of memory. In this configuration, each GPU can support 12 vGPUs per card for a total of 96 vGPUs across the cluster.

Because IT managers configure VMs to support various user requirements, we ran two workloads in parallel. The first workload involved 96 users running a combination of the Knowledge Worker Profile and GPU Reference Workload on GPU-enabled VMs. The second workload involved just the Knowledge Worker profile on CPU-only VMs.

What We Found

Our findings can be summarized by 3 main takeaways, all of which are great news for businesses and users alike!

Improved Frame Rates

While running 96 combined Knowledge Worker and GPU Reference Profile users on GPU-enabled VMs and 860 Knowledge Worker users on CPU-only VMs, we observed that frame rates for a single user increased from 15 frames per second (fps) to 59 fps at peak. That is almost a 400% improvement!

Picture1.jpg

Lower Peak CPU Utilization

Next, we ran two 96-user tests with the combined Knowledge Worker profile and the GPU Reference workload on CPU-only VMs and then again on GPU-enabled VMs. During graphics-intensive operations, we measured a 22% decrease in CPU Utilization at peak. This demonstrates that offloading graphics-intensive activities to a GPU can free up the CPU for other tasks.

Picture2.jpg

Handling Mixed Workloads at Scale

Lastly, we wanted to show how the system handled mixed workloads. To establish a baseline, we ran a test with the Knowledge Worker profile and achieved 950 users on CPU-only VMs. Next, to understand the resources required to run the combined GPU Reference Workload and Knowledge Worker profile, we ran the test again on CPU-only VMs. Our results were only 450 users - less than half the previous result. To demonstrate the value of the Intel Data Center GPU Flex series, we then ran the combined workload with 96 users on GPU-enabled VMs and 860 users running the Knowledge worker profile on CPU-only VMs. We achieved 956 total users in this test, more than our original baseline test.

We created a 3-year TCO model inclusive of hardware, software licensing, and 3 years of support. In our testing, we demonstrated that by incorporating Intel Data Center GPU Flex 140s, we could support more users, some running graphics-intense workloads, for a price-per-user increase of only 1.5%.

Depending on the workload, a GPU can support graphics-intensive users at roughly the same price per user!

Picture3.png

The data doesn’t lie. Graphics-intensive workloads, like the GPU Reference workload, will benefit from using a GPU as it provides a better user experience at roughly the same cost. Intel® Data Center GPU Flex Series supports graphics-intensive workloads and can scale across your existing infrastructure. That is a win-win all around!

If you’d like to learn more about our testing process and results, check out the Lab Insights.

Workloads and Configurations

Config H14 – 4th Gen Intel® Xeon® Platinum Processor (32c), 950 users

4-node, Each node, Intel Software Development Platform, 2x Intel® Xeon® Platinum 8462Y+ processor (32C, 2.8GHz, 300W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 2 TB (16x128GB DDR5 1DPC 4800 MHz), BIOS: 3A11.uh, ucode 0x2b000111, Intel E810-CQDA2 100G, 2 Groups: Per node cache tier: 2x 1.6TB P5800X Gen 4 SD 5800X, Per node capacity tier: 6x 3.84TB D7-P5510 Series, Gen4, ESXi, 8.0.2, 22380479 vCenter 22385739, Horizon 8.11.0 2309 Build 22629722, Login Enterprise 5.7.2. Knowledge worker profile, 2vCPU/8GB. EUX 7.5. Testing by The Futurum Group Dec 2023 through Feb 2024

Config H15 – 4th Gen Intel® Xeon® Platinum Processor (32c), 450 users.

4-node, Each node, Intel Software Development Platform, 2x Intel® Xeon® Platinum 8462Y+ processor (32C, 2.8GHz, 300W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 2 TB (16x128GB DDR5 1DPC 4800 MHz), BIOS: 3A11.uh, ucode 0x2b000111, Intel E810-CQDA2 100G, 2 Groups: Per node cache tier: 2x 1.6TB P5800X Gen 4 SD 5800X, Per node capacity tier: 6x 3.84TB D7-P5510 Series, Gen4, ESXi, 8.0.2, 22380479 vCenter 22385739, Horizon 8.11.0 2309 Build 22629722, Login Enterprise 5.7.2. Knowledge worker profile + GPU Reference Workload + Zoom client on CPU-only VMs,2vCPU/8GB. EUX 7.5. Testing by The Futurum Group Dec 2023 through Feb 2024.

Config H18 – 4th Gen Intel® Xeon® Platinum Processor (32c), 956 total users (96 GPU-enabled users, 860 CPU-only users)

4-node, Each node, Intel Software Development Platform, 2x Intel® Xeon® Platinum 8462Y+ processor (32C, 2.8GHz, 300W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 2 TB (16x128GB DDR5 1DPC 4800 MHz), BIOS: 3A11.uh, ucode 0x2b000111, Intel E810-CQDA2 100G, 2 Groups: Per node cache tier: 2x 1.6TB P5800X Gen 4 SD 5800X, Per node capacity tier: 6x 3.84TB D7-P5510 Series, Gen4, 2xIntel Data Center GPU Flex Series 140, ESXi, 8.0.2, 22380479 vCenter 22385739, Horizon 8.11.0 2309 Build 22629722, Login Enterprise 5.7.2. Appliance 1: 96 Knowledge worker profile + GPU Reference Workload (1080p) + Zoom client on GPU-enabled VMs, 2vCPU/8GB. EUX 7.5. Appliance 2: 860 Knowledge Worker users on CPU-only VMs, 2vCPU/8GB, EUX 7.5. Testing by The Futurum Group Dec 2023 through Feb 2024

Config H20 – 5th Gen Intel® Xeon® Platinum Processor (32c), 956 total users (96 GPU-enabled users, 860 CPU-only users).

4-node, Each node, Intel Software Development Platform, 2x Intel® Xeon® Platinum 8562Y+ processor (32C, 2.8GHz, 300W TDP), HT On, Turbo ON, SNC OFF, Total Memory: 2 TB (16x128GB DDR5 1DPC 4800 MHz), BIOS: 3B05, ucode 0x2b000111, Intel E810-CQDA2 100G, 2 Groups: Per node cache tier: 2x 1.6TB P5800X Gen 4 SD 5800X, Per node capacity tier: 6x 3.84TB D7-P5510 Series, Gen4, 2xIntel Data Center GPU Flex Series 140, ESXi, 8.0.2, 22380479 vCenter 22385739, Horizon 8.11.0 2309 Build 22629722, Login Enterprise 5.7.2. Appliance 1: 96 Knowledge worker profile + GPU Reference Workload (1080p) + Zoom client on GPU-enabled VMs, 2vCPU/8GB. EUX 7.6. Appliance 2: 860 Knowledge Worker users on CPU-only VMs, 2vCPU/8GB, EUX 7.7. Testing by The Futurum Group Dec 2023 through Feb 2024

Notices and Disclaimers

Performance varies by use, configuration and other factors. Learn more at the Performance Index Site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See backup for configuration details.
No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.

 

 

About the Author
Glenn Le Vernois is the Sales Enablement Manager in Intel’s Data Center Segment Marketing organization. Over the past 5 years, he has led the VDI solution development efforts, working across the industry to enable better user outcomes with adoption of GPUs and new technologies as organizations have transitioned to support a hybrid work model. After joining Intel in 2005, Mr Le Vernois has held positions in the Sales & Marketing Group, Data Center and A&I Group, Client Computing Group, Software and Services Group, among others. A 35-year veteran of the computer industry, Mr. Le Vernois has held prior Senior Management positions with Sequent Computer Systems, Xerox, and IBM. Mr. Le Vernois holds a B.S. in Computer Science from Pacific Lutheran University, and an MBA from the University of Portland.