NVIDIA GRID – AppGuide (AutoCAD)
I want to share this with you all, Luke Wignall and his team from NVIDIA have created some great AppGuides, that helps with understanding how many users can you put on a NVIDIA GRID system with either K1/K2 in a VMware environment. These guides are made together with vendors such as Autodesk. The guides focus on following apps Autodesk AutoCAD.
In this blogpost my goal is to highlight the great work NVIDIA have done creating the scalability app guides and these guides helps you if you want to virtualize Autodesk AutoCAD 2015 with NVIDIA GRID and VMware Horizon. The guides are great – cause they give an idea what you would require in a virtualized environment and these can be reused for other virtualized platforms such as Citrix and Microsoft – keep in mind that results might be different. If you would like to get more informations about how the setup is configured and which methodology i recommend you read the AppGuide, you can download it in under source in the end of this article.
The appguides gives a great idea to understand the impact of CPU and how the GPU are giving value.
About AutoCAD 2015
Autodesk AutoCAD is one of the most globally used software platforms for CAD design and documentation. AutoCAD leverages both CPU and GPU to deliver a high quality user experience, and as a result, there are several things that need to be considered in architecting your GRID vGPU solution: the size of your map data, the concurrency of your users, and the level of interaction with primary 2D and some 3D data.
Results NVIDIA Appguide for Autodesk AutoCAD 2015
The following are results of our testing, looking for the greatest scalability while still within performance expectations. It’s important to note that your users, your data, and your hardware will impact these results, and you may decide a different level of performance or scalability is required to meet your individual business needs. As the Catalyst benchmark does not push AutoCAD’s GPU capabilities, and was built to push the limits of dedicated hardware versus the shared resources of VDI, the decision was made to stop testing once the CPU was approaching 100% utilized and test times had climbed past twice what we were finding on the a single physical workstation with dedicated resources. We then met with the Autodesk AutoCAD team, discussed the results, reviewed the tests in action, and physically verified that this was still within what a typical user would deem acceptable and usable. When running AutoCAD, anything higher than 20 FPS is awesome, but users generally don’t notice the difference once you exceed 30 FPS. However, once you drop below 10 FPS, the software is going to feel very sluggish and become unusable by the time you hit 5 FPS. • 20 fps above is good • Below 10fps – sluggish • 5 fps – unusable
- 32VM : K220Q perform better than K120Q but K120Q is still maintain avg 31FPS.
- 40VM : Due to vGPU configuration, need K120Q, avg FPS is 24 FPS
- 48VM : one you hit 40 VM running, avg FPS is dropping below 10 FPS.
AutoCAD requires significant CPU resources, so investing in higher CPU speeds and more cores will pay off on performance and scalability. For medium to large models, K240Q performance might be better for a real use scenario. Since your own situation is different, you must test with your own models to ensure the most accurate results. The chart below shows a comparison between two different sets of VDI guests, one using the K2 and its K220Q profile, the other a K1 and its similarly sized frame buffer profile, the K120Q:
Notice that as scale, the number of concurrent VM’s increases, there is an obvious drop in performance, and logs will show this is the CPU becoming 100% utilized. This is a “peak workload” test as all sessions are started in unison, we would expect humans to be more staggered in their workflow.
The next chart shows results when we add that synthetic human behavior, we delay the launching of each session by 5 seconds, this offsets the tests and prevents all tests hitting the same function and impacting the CPU in unison.
Additional findings from this testing:
• vCPU – 2 vCPU are sufficient for this benchmark, however increasing model size and complexity will change this.
• Virtual System Memory – 2 GB is sufficient for this benchmark, however increasing model size and complexity will change this.
• K220Q/K120Q : both show capped FPS caused by frame rate limiting (FRL)
• vGPU has 45 fps frame limiter for performance balancing across multiple VM
AutoCAD 2015 Users per server
Based on the NVIDIA GRID™ Performance Engineering Lab (GRID P.E.L.) findings, NVIDIA GRID provides the following performance and scalability metrics for Autodesk AutoCAD 2015. These metrics are based on tests with the lab equipment shown in the graphic below, using the Cadalyst benchmark, and in working with Autodesk and their emphasis on end user usability. Of course, your usage will depend on your models and equipment, so this Application Guide is intended to be used as a starting point for your implementation.