Benchmark Competition - Redway Turbine - Just for fun! :-)

NVIDIA Virtual GPU Forums Virtual GPU Benchmarks
#1

Just for a bit of amusement and interest, thought I’d lay down a challenge to see if anyone has a system that can beat what I’m currently using :-)

I’ve chosen Redway Turbine for the benchmark as there is no configuration other than “Full Screen” or “Windowed” and it’s only a small download and doesn’t take long to complete. If there is preference for another application, then let me know :-)

For consistency, I’m running “Windowed”, as this will run in the same resolution from any system (1024x768) so you get comparable results. I can do a 4K benchmark if anyone would like to up the ante! :-D

Graphics wise, I know this is a GRID forum, but GRID aligns with the Quadro range so if you have a Quadro in your system then that’s cool.

Redway Turbine is available from here: http://www.redway3d.com/home/gallery/demos/

Here are my results, if you think you can beat them, come and have a go! :-D … And even if you can’t, let’s have a look and see if we can help tune your platform :-)
Benchmark.jpg

#2

After a bit of GPU profile tuning, the bar has been raised :-) …
Benchmark v2.jpg

#3

Hello,

Hoping to get some help in optimizing my environment for 2D/3D CAD/Revit workloads.

First my test results:

RES: 1024x768
REALTIME VIEWPORT: 749
HIGH QUALITY REAL-TIME: 926
DYNMAIC OCCLUSION: 1229
HYBRID RAY-TRACE: 829
TOTAL SCORE: 917

ENVIRONMENT:

Horizon View 7.1

Host:
UCSC-C240-M4SX
16 CPUs x 3.19 GHz
260GB RAM
GRID K1 Driver Version 367.92

GUEST VM:
Windows 7 Optimized for view
2vCPU
12GB RAM
grid_K140Q Profile - In-Guest NVIDIA Driver 369.95

#4

Hi

Finally!! Someone has replied!! :-D

That’s not bad, it’s only taken 7 months to get a response despite over 1000 views! haha :-)

So, a fellow Cisco advocate! Great choice, we use Cisco servers too!

Thanks for running that benchmark and generating some results for us. I’m going to be honest, that score is on the lower side of what you should be hoping for, however it is representative of that GPUs performance and use case, which I’m sorry to say is not going to be CAD workloads. You have a good server specification with nice fast CPUs and a nice chunk of RAM, but it is really being handicapped by that GPU. Really, you should be looking at nothing less than Maxwell. The K1 was never intended for anything resource intensive. A K2 would fare a lot better for this task, but it’s older technology now, and they will soon become a bottleneck for your Hypervisor choice, as the last versions to support Kepler are XenServer 7.1 and ESXi 6.5. Drop an M60 in there and watch the performance (and user density when running CAD) change dramatically!

However, let’s see if we can bump up that score a little bit, as the tuning will be the same regardless of GPU … So, starting from the bottom:

  • Ensure your UCS firmware and BIOS are fully up to date. A new version was released last month.
  • Go through your BIOS settings and tune for Maximum Performance. That includes CPU, Memory and Cooling / Fan Speed. Remove all "Balanced" / "Economy" settings.

This will help you with your BIOS settings: http://www.cisco.com/c/en/us/solutions/collateral/data-center-virtualization/unified-computing/whitepaper-c11-737931.html

  • Hypervisor, you don’t mention what you’re using, however as you’re using Horizon, I’m pretty sure you’ll be running ESXi. Make sure this is the latest version of 6.5, nothing older! 6.5 has a lot of performance fixes over 6.0.
  • Set ESXi for "High Performance".
  • Install the latest NVIDIA GRID drivers for the most up to date functionality and performance.

Ok, that’s the infrastructure part covered. Nice and easy, nothing out of the ordinary there.

VM Specifications

  • RAM is fine
  • GPU is a little low, I’d be looking at at least 2GB (K160Q) if not more (K180Q) for CAD. However increasing frame buffer won’t help your benchmark results.
  • CPU is too low. Nothing less than 4 vCPU per VM. Maybe more when you’re running CAD workloads.

Operating System

That should give you a good starting base.

Which model CPUs are you running? v2, v3 or v4?

What storage are you using for your VMs and the data?

Regards

Ben

#5

Better late then never right.

https://www.dropbox.com/s/o6w0fhnzlmv12f6/benchmark.PNG?dl=0

Not so impressed with the score to be honest.

https://www.dropbox.com/s/206ejl66r0ea1yj/specs.PNG?dl=0

Im running:

Dell R730

2x Xeon E5-2667v4 3.2 ghz
512gb ram
1 x P40
1 x M60

Xenserver 7.5
citrix studio 7.1

Guest VM
1 socket 16 vCPU @ 3.2
64gb Ram
P40 12 Q GPU
Win 10 Pro.

Any help would be appreciated.

Thanks!

#6

Wow, that’s some serious thread-resurrection!! I thought this was done! :-D

Have you got an M60 and P40 installed in the same R730? …

Yeah, looking at your specs that score is way too low! … Start from the bottom of the stack and work up …

> Make sure BIOS version / firmware is up to date.
> Optimise BIOS for Performance. These setting vary from vendor to vendor, so research the best way to do it.
> Ideally, make sure all your software is up to date and fully patched (Hypervisor & Operating System)
> Make sure your vGPU Scheduler is set for Best Effort (there’s no GPU cycle limitations)
> Modify your vCPU configuration to 2 Sockets with 8 Cores Per Socket. However, unless your application is multi-threaded, you’re far better off reducing to something like 4 Cores Per Socket (8 vCPU) or even 2 Cores Per Socket (4 vCPU).
> Make sure you’re running the most up to date vGPU drivers, and in the NVIDIA Control Panel, configure for Max Performance
> Make sure you’re running the most up to date VDA
> Run the Citrix Optimiser on your Operating System ( https://support.citrix.com/article/CTX224676 )
> Set Operating System Power settings
> Lastly … When benchmarking, don’t forget to disable the FRL for your VM … ;-)

xe vm-param-set uuid=7a0439c8-bcf0-7e72-abb2-f9f7dd8a8d01 platform:vgpu_extra_args="frame_rate_limiter=0"

Replace that UUID with the one from your VM and run that command on your Xen Host

See how you get on now …

Regards

Ben

#7

ill go through the list.

Just flicked off FRL and it hit 4030.

yeah i have the p40 and the m60 in the same machine…bad idea?

#8

Cool. Obviously don’t disable the FRL in a production environment though …

Yes, although it technically works, having mixed GPUs in the same sever is not supported. Fine for a Lab, but a no no for production running. I personally don’t understand why server vendors won’t support it, I’ve had no issues in my lab and have run mixes of Maxwell, Pascal and Volta without issue. Maybe support for it will come later on, but at the moment, we’re limited to a single type of GPU per server for production environments.

Regards

Ben