NVIDIA Unveils Cloud GPU Technologies

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
NVIDIA today kicked off the GPU Technology Conference (GTC) by unveiling its third industry-shaping innovation since its founding 19 years ago -- technologies that accelerate cloud computing using the enormous computing capabilities of the GPU. Five years in the making, NVIDIA's cloud GPU technologies are based on the company's new Kepler GPU architecture, designed for use in large-scale data centers. Its virtualization capabilities allow GPUs to be simultaneously shared by multiple users. Its ultra-fast streaming display capability eliminates lag, making a remote data center feel like it's just next door. And its extreme energy efficiency and processing density lowers data center costs.
 
I'm excited about this announcement. I'm an administrator for a 200 server Citrix farm that serves up CAD applications to developers all acorss the world. Windows 2008 R2 allowed us to share the GPU and provide a better performance to my users. If this hardware gets us another performance boost, it will be a good day!
 
idc what the industry guys say, 60fps gaming just isn't realistic 'in the cloud'. to get 60fps you need <14ms latency. even 30ms while realistic isn't feasible for a cloud provider as they will need to have the ability to originate services basically everywhere. in theory you could have a POP in all the major markets and cover much of the market but you're going to lose market share, especially in the US where because of our size broadband penetration just isn't that great.

IMO there is nothing wrong with having 'a box' which runs 'games' that you play locally. requires much less in terms of network and avoids things like over provision, scalability, and things like per seat licensing of individual VDI sessions.

now if someone created a stripped down linux VM thus avoiding the microsoft tax and based the hypervisor on XCP bypassing citrix's licensing then you can avoid some of the licensing costs but you still have the problem of over provisioning. games require more horsepower than any enterprise app to run. you can't slam 20 'cloud gaming' VDI sessions onto a single socket and expect to have an acceptable level of performance, idc how great nvidia's gpu sharing is that just won't work. this leads to significant compute layer problems where the margins get real thin because you can't drive the density.
 
Back
Top