NVIDIA
Performance Difference Between K1 and K2
Hello All! I've recently been fortunate enough to be able to work on deploying a Dell 720 with two K2 cards, VMware 6.0 and XenDesktop/XenApp 7.6. I must admit, I am impressed with the performance I have received with this combination. It won't beat out a custom quad SLI rig, but for an enterprise/business environment, its great. My question is, does anyone have performance specs ("real world") between the two different cards? Or even a gut feel? We're in a position to get another server, and wanted to look at the possibility of K1 cards for a higher user density. Just looking at the specs of the boards tells me it will be slower, but its hard to know "how much slower" without kicking the tires. Any information would be greatly appreciated! I'll post any information I find as I go along as I'm just beginning my search on this information.
Hello All!

I've recently been fortunate enough to be able to work on deploying a Dell 720 with two K2 cards, VMware 6.0 and XenDesktop/XenApp 7.6. I must admit, I am impressed with the performance I have received with this combination. It won't beat out a custom quad SLI rig, but for an enterprise/business environment, its great.

My question is, does anyone have performance specs ("real world") between the two different cards? Or even a gut feel? We're in a position to get another server, and wanted to look at the possibility of K1 cards for a higher user density.

Just looking at the specs of the boards tells me it will be slower, but its hard to know "how much slower" without kicking the tires.

Any information would be greatly appreciated! I'll post any information I find as I go along as I'm just beginning my search on this information.

#1
Posted 08/14/2015 03:37 PM   
I'll go ahead and throw this in here as it is relevant and good information: http://www.nvidia.com/content/grid/resources/AutoCAD_GRID_vGPU_Scalability_Solutions_Guide.pdf
I'll go ahead and throw this in here as it is relevant and good information:

http://www.nvidia.com/content/grid/resources/AutoCAD_GRID_vGPU_Scalability_Solutions_Guide.pdf

#2
Posted 08/14/2015 03:41 PM   
Thank you for the nudge! Given subsequent updates from both VMware and Citrix we need to refresh that document. Our primary problem in testing, with most ISV's and certainly with AutoCAD, is a lack of agreed upon user acceptability thresholds. All previous testing has been to establish the max performance on that custom quad SLI rig you referenced, which is great except now you need to know what the least performance can be before end users begin to complain. In reference to the document you linked to, if we use CADALYST as our benchmark (And is this an ideal test? Does it test AutoCAD as a user would? Etc.) then what is the index value where users stop being happy? Next issue, how similar are your models to the tests used in that benchmark? We continue to test and work on these issues but also need your own results to add real field data to our findings. Please message me here if you would be willing to share such data.
Thank you for the nudge! Given subsequent updates from both VMware and Citrix we need to refresh that document. Our primary problem in testing, with most ISV's and certainly with AutoCAD, is a lack of agreed upon user acceptability thresholds. All previous testing has been to establish the max performance on that custom quad SLI rig you referenced, which is great except now you need to know what the least performance can be before end users begin to complain. In reference to the document you linked to, if we use CADALYST as our benchmark (And is this an ideal test? Does it test AutoCAD as a user would? Etc.) then what is the index value where users stop being happy? Next issue, how similar are your models to the tests used in that benchmark? We continue to test and work on these issues but also need your own results to add real field data to our findings. Please message me here if you would be willing to share such data.

Regards,

Luke Wignall
Performance Engineering Manager
NVIDIA | Worldwide Sales ­ GRID Computing
http://www.linkedin.com/in/lukewignall/
https://twitter.com/lwignall

#3
Posted 08/17/2015 10:59 PM   
Luke, Sorry for my late reply, but thank you so much for getting back to me. I have some benchmarks from Uniengine, and Sky Diver 1.0. I have a few others that were not as detailed as those. Typically I ran those against my Surface Pro 3 and a Dell Precision Workstation with a K4000 card in it as well. Let me compile some of that and also look into running the CADALYST benchmark and see what results I can pull. I'll shoot you a PM for contact information. Thanks!
Luke,

Sorry for my late reply, but thank you so much for getting back to me. I have some benchmarks from Uniengine, and Sky Diver 1.0. I have a few others that were not as detailed as those. Typically I ran those against my Surface Pro 3 and a Dell Precision Workstation with a K4000 card in it as well.

Let me compile some of that and also look into running the CADALYST benchmark and see what results I can pull. I'll shoot you a PM for contact information.

Thanks!

#4
Posted 08/25/2015 02:22 PM   
Scroll To Top

Add Reply