-1

How does NVIDIA GeForce NOW manage such high and massive data transmission between PC and servers?

PS: Please attach relevant articles or journal papers to support your claim.

3
  • Could you qualify the massive data transmission? Per shield.nvidia.com/support/geforce-now/system-requirements/2, their top bandwidth requirement is 50Mb/s (i.e. about 6.25MB/s), and they have a requirement for a round trip to their nearest data center to be below 60ms. A quick check for Netflix (help.netflix.com/en/node/306) suggests Nvidia has far higher bandwidth requirements for a given picture quality. – iwaseatenbyagrue Mar 3 '17 at 9:34
  • The data of the movement of the Mouse and Keyboard has to be sent with almost zero latency and even processed and received back almost instantly. Can the servers be placed half world apart to perform this? – Muddu Mar 3 '17 at 10:16
  • I think the 60ms RTT sort of rules that out. Speed of light in a vaccuum (en.wikipedia.org/wiki/Speed_of_light) is 299792458 meters/second. One second is 1000ms, so roughly speaking, 60ms RTT is 299792458 * 0.03 = 8993773.74 meters (0.03 because a 60ms RTT is the same as a 30ms one-way trip). So even in ideal conditions, you would need to be within 9000 kilometers of one of their DCs. The real world is not ideal (and involves more than fiber as a medium), so the real figure is probably a fair bit lower. – iwaseatenbyagrue Mar 3 '17 at 10:29
2

Figured it might be better to work my comment into an answer:

Nvidia requires a fair amount of available bandwidth to be available, and also has a requirement for users to be within a 60ms RTT of one of their (6) datacenters.

Their maximum bandwidth requirement is 50Mb/s (6.25MB/s) for 1080p quality. This is substantially more than Netflix requires for an equivalent quality stream (1080p is an HD feed, Netflix wants 10 times less (5Mb/s) for it).

If we assumed the entire network to be one continuous strand of fiber AND that the light was travelling in a perfect vacuum (neither of these assumptions is actually realistic), then you would need to be within a 9000 kilometer radius of one of their DCs:

299792458 * 0.03 = 8993773.74 meters

Or, in plain english: speed of light in meters/second times 30 milliseconds (60ms RTT means a 30ms one way trip) gives you about 9000km.

The real world is not generally ideal, so I suspect the real distance you need to be from their DCs is going to be quite a bit lower (speed of light in fiber is 1.5 times the speed of light in a vacuum, so on a pure fiber network with no switching, max distance is already something like 6000km, not 9000km).

In other words: the great performance is totally dependent on the quality of content delivered, and on the distance between you and an Nvidia DC.

What exactly is transferred to/from their servers is also an unknown to me, and it may be mouse/keyboard is processed locally, and the resulting coordinates/events are transferred up (this is a total guess).

So, on the face of it, the answer is that they have pretty high network requirements, which allows them to move a fair bit of data round when they need to.

Refs:

3
  • Can you use any other mode of the data channel to achieve it? i.e. something other than an optic fibre? – Muddu Mar 3 '17 at 11:26
  • Yes, any transport/medium can be used - the calculations above are more or less an 'ideal' scenario. In the real world, there is some overhead for switching/routing, and generally, a heterogenous network (some fiber, some copper ethernet, some ATM/DSL/cable copper, etc). You just adjust the calculations for whatever the medium is (all of this is kind of a rule of thumb anyway, because networks congest, packets get lost, etc) – iwaseatenbyagrue Mar 3 '17 at 11:30
  • The point was more that Nvidia (possibly) has good performance when there is a good network for it to use, but I am not sure there is anything remarkable (in terms of network performance) about the product of theirs you mention. – iwaseatenbyagrue Mar 3 '17 at 11:36

Not the answer you're looking for? Browse other questions tagged or ask your own question.