Artificial Intelligence Computing Leadership from NVIDIA
Consolidation\Density for vSGA on M10
Hi All, Just wanted to clarify the math on the M10 card specs. I see that if I'm running vDGA utilizing a profile, say the M10-2q, that I can run 4x VMs per pGPU (8GB per pGPU / 2GB per VM = 4 VMs). Depending on the profile selected, an M10 can run anywhere from 64 to 4 VMs in this manner. How does this apply to vSGA? The VMware literature appears to indicate that maximum consolidation limits are applied by total pGPU VRAM divided by VM-allocated VRAM. So, assuming my VMs have 96MB of VRAM allocated by Horizon, does this mean I can run ~80 VMs on a single M10 pGPU? I really don't see any mention of consolidation ratios on NVIDIA's literature outside of the vDGA scenarios which are utilizing the GRID 'profiles' -- however in vSGA-land such profiles are not used, so just curious if my assumption about calculating density is correct and if there's any official literature on that from NVIDIA's side.
Hi All,

Just wanted to clarify the math on the M10 card specs. I see that if I'm running vDGA utilizing a profile, say the M10-2q, that I can run 4x VMs per pGPU (8GB per pGPU / 2GB per VM = 4 VMs). Depending on the profile selected, an M10 can run anywhere from 64 to 4 VMs in this manner.

How does this apply to vSGA? The VMware literature appears to indicate that maximum consolidation limits are applied by total pGPU VRAM divided by VM-allocated VRAM. So, assuming my VMs have 96MB of VRAM allocated by Horizon, does this mean I can run ~80 VMs on a single M10 pGPU?

I really don't see any mention of consolidation ratios on NVIDIA's literature outside of the vDGA scenarios which are utilizing the GRID 'profiles' -- however in vSGA-land such profiles are not used, so just curious if my assumption about calculating density is correct and if there's any official literature on that from NVIDIA's side.

#1
Posted 04/01/2019 10:03 PM   
You cannot compare vSGA with vGPU. Totally different technology. Only vGPU runs with Nvidia driver in the guest and is not limited in terms of API intercepting.
You cannot compare vSGA with vGPU. Totally different technology. Only vGPU runs with Nvidia driver in the guest and is not limited in terms of API intercepting.

#2
Posted 04/04/2019 06:05 PM   
[quote="sschaber"]You cannot compare vSGA with vGPU. Totally different technology. Only vGPU runs with Nvidia driver in the guest and is not limited in terms of API intercepting.[/quote] ^^ Somebody don't read too good. I'm not comparing the technology at all, and in fact have a great understanding of the differences between the two. I am asking [u]very specifically[/u] about [b]consolidation ratios[/b] -- e.g: what is the formula to determine [b][u]HOW MANY[/u][/b] vSGA users I can fit onto a single M10 card?
sschaber said:You cannot compare vSGA with vGPU. Totally different technology. Only vGPU runs with Nvidia driver in the guest and is not limited in terms of API intercepting.


^^ Somebody don't read too good.

I'm not comparing the technology at all, and in fact have a great understanding of the differences between the two.

I am asking very specifically about consolidation ratios -- e.g: what is the formula to determine HOW MANY vSGA users I can fit onto a single M10 card?

#3
Posted 04/08/2019 03:59 PM   
Well, give it try and let us know. As I said it is a dying technology we don't have focus on. Didn't have a single customer in the last 2 years using vSGA. BTW: I read pretty well but without having real field numbers it makes no sense to recommend any number at all.
Well, give it try and let us know. As I said it is a dying technology we don't have focus on. Didn't have a single customer in the last 2 years using vSGA.
BTW: I read pretty well but without having real field numbers it makes no sense to recommend any number at all.

#4
Posted 04/08/2019 04:24 PM   
Scroll To Top

Add Reply