vritualization Nvidia A6000 GPU aggregation using Nvidia vGPU?

[xx_glue_consumer_69]

So I was scouring eBay for some graphics cards when I came across an AMD v340 engineering sample for like 1500 bucks.  What was super cool about this card is that it supported GPU virtualization.  Unfortunately its way underpowered for what I need so I looked to see if Nvidia had anything available to pros or enterprise and I found the A6000 that supported their vGPU tech.  They claim that with aggregation, the performance increase is close to linear, but I'm not sure if that would apply to something like a game engine that uses gaming drivers.  The goal here is to have a dynamic server that can allocate GPU resources on the fly for several tasks, including a kick-ass gaming rig with that aggregated gpu.  In principle if it scaled the way that I hope, It would make a single virtualized GPU that is 2x more powerful than a 3090 (the A6000 is like 10% faster I'm pretty sure).  Why do I need such power for games?  I work in the VR industry and I've been wanting to work with an ultra high fidelity VR game to demo a piece of hardware.  Does anyone have experience with vGPUs or testing this kind of aggregation?

[ki8aras]

i find their claims of 2x the performance of a 3090 kinda hard to believe, yes it has 48 gb of memory, which is twice the gpu memory of the 3090 but that doesn't mean in any way that the performance is doubled

[xx_glue_consumer_69]

1 minute ago, ki8aras said:

i find their claims of 2x the performance of a 3090 kinda hard to believe

Thats not what they claim, I said it was only 10% faster than a 3090 and likely only in compute applications.  The doubling of performance would come from aggregating 2 gpus using gpu virtualization.  Sorry if I was a bit unclear on that.

[igormp]

3 hours ago, RaylaVR said:

 They claim that with aggregation, the performance increase is close to linear, but I'm not sure if that would apply to something like a game engine that uses gaming drivers.  The goal here is to have a dynamic server that can allocate GPU resources on the fly for several tasks, including a kick-ass gaming rig with that aggregated gpu.

That's not how it works. What they mean is that you can have both a single GPU being used by many instances at once, or assign multiple GPUs onto a single instance, which will present themselves as different GPU inside the VM.

 

The gains are for software than can make use of multi gpu anyway, not something like games, that's why it's usually linear.