Nvidia Shield: 4K Streaming And Gaming For The Living Room

[Mitrovah]

Intel's Compute Stick, now Shield. The next step build a 42 tv with all the components in back, maybe powerful AMD APU, with a SSD Optical drive = Heaven.

 

[InvalidError]

This is very possible with Gigabit internet. Go go Google fiber!

I'm pretty much certain the amount of traffic between the CPU and GPU easily exceeds 1Gbps, especially during bursts. The control flow and serialization delay between the two would likely not fare too well with over 10ms of round-trip latency either instead of the microsecond scale GPUs normally deal with over PCIe to the CPU.

For comparison's sake PCIE 3.0 x16 is 128Gbps while 2.0x8 is 40Gbps. 1GbE is nowhere near fast enough to remotely drive a GPU if the goal is high responsiveness.

 

[Alec Mowat]

This is very possible with Gigabit internet. Go go Google fiber!

I'm pretty much certain the amount of traffic between the CPU and GPU easily exceeds 1Gbps, especially during bursts. The control flow and serialization delay between the two would likely not fare too well with over 10ms of round-trip latency either instead of the microsecond scale GPUs normally deal with over PCIe to the CPU.

For comparison's sake PCIE 3.0 x16 is 128Gbps while 2.0x8 is 40Gbps. 1GbE is nowhere near fast enough to remotely drive a GPU if the goal is high responsiveness.

That's assuming that a modern GPU ever uses the full performance of the 128Gbps. It's also 4 GB per lane with 16 lanes, that means there's no single connection over 4 GB. A pair of multi-port 10 Gbps NIC in a PCI-E 3.0 slot could provide similar performance.
http://www.tomshardware.com/reviews/pci-express-scaling-p67-chipset-gaming-performance,2887-9.html
http://www.newegg.ca/Product/Product.aspx?Item=9SIA5SC2GK4549

It's most likely bottlenecked by anything else in use as well. I don't really think it's the amount of data that really makes the difference, it's the speed it travels. There's no latency inside the case.

However, a PCI-E x16 VRAM cache card with a 10Gbps NIC built in could store data while it streams.

I think a more realistic option would be a virtual world that is rendered completely offsite, but your local character and actions render locally. You could have multiple people walking around a pre-rendered city that streams into your computer.

We are stuck with Cloud Computing, whether we like it or. It's trying to find the best solution to get local game performance feel from offsite hardware.

 

[InvalidError]

That's assuming that a modern GPU ever uses the full performance of the 128Gbps. It's also 4 GB per lane with 16 lanes, that means there's no single connection over 4 GB.

PCIe 3.0 is 8Gbps per lane in each direction and data across the PCIe interface gets spread across all available lanes unless they have been put in power-down/standby state. Burst transfers across the interface do occur at 128Gbps when all lanes are active and such bursts in the 10s of MBs are happening all the time when the CPU updates scene geometry before the GPU can do its part of the rendering job.

Transferring even a meager 10MB (80Mbits) at 1Gbps already introduces 80ms worth of latency due to (de-)serialization. That's why GPUs need PCIe 2.0x8 (40Gbps) or faster to achieve their peak performance - can't render faster than scene geometry can be uploaded and in modern games, there can be a lot more than 10MB worth of scene data to update between frames.

 

[santiago7245]

Meh...I hate playing with a controller so I will pass on this.

 

[Alec Mowat]

That's assuming that a modern GPU ever uses the full performance of the 128Gbps. It's also 4 GB per lane with 16 lanes, that means there's no single connection over 4 GB.

PCIe 3.0 is 8Gbps per lane in each direction and data across the PCIe interface gets spread across all available lanes unless they have been put in power-down/standby state. Burst transfers across the interface do occur at 128Gbps when all lanes are active and such bursts in the 10s of MBs are happening all the time when the CPU updates scene geometry before the GPU can do its part of the rendering job.

Transferring even a meager 10MB (80Mbits) at 1Gbps already introduces 80ms worth of latency due to (de-)serialization. That's why GPUs need PCIe 2.0x8 (40Gbps) or faster to achieve their peak performance - can't render faster than scene geometry can be uploaded and in modern games, there can be a lot more than 10MB worth of scene data to update between frames.

According to Nvidia, the encoding/decoding performance of GRID is almost double the performance of a GPU in console, which allows for double the latency elsewhere to get equal performance.

We have Fiber speed records near 32 TBps.

http://www.extremetech.com/extreme/192929-255tbps-worlds-fastest-network-could-carry-all-the-internet-traffic-single-fiber

"This wasn’t just a short-range laboratory demo, either: The multi-mode fiber link was one kilometer (0.62 miles) long."

The speed of light is the speed of light. If we start developing proper fiber infrastructures, instead of the old copper cable networks, we'd overcome the latency really quickly. IPV6 would reduce NAT and switching latency.
The only thing holding us back from cloud computing on a massive scale is the ISP's inability to provide a massive infrastructure overhaul.

 

[photonboy]

Awesome for the price asked.

$200 for a rig that can play a recoded Crysis 3 (1080p @ 30FPS) is great. A steadily increasing collection of Android games and the ability (where possible) to use NVidia GRID.

They have to start somewhere and $200 is a very good price point. Obviously won't a appeal to everybody but what does?

One problem is the lack of HDD starts eroding the price advantage. We've seen XBOX ONE for $350USD and they'll get cloud-based gaming too, but offer a 500GB HDD, BluRay player, and have a more powerful machine.

NVidia needs to get traction quickly. I don't think STEAMOS is much competition really since that's a different price point. For $200 it's going to appeal to a lot of people as 2nd machines.

 

[photonboy]

Update:
I meant "lack of HDD" as in you'd have to buy one if downloading many games (16GB local storage). Puts cost at $250+ if buying a $50 or more USB hard drive.

It was the best call though since not all people need an HDD.

 

[InvalidError]

The speed of light is the speed of light. If we start developing proper fiber infrastructures, instead of the old copper cable networks, we'd overcome the latency really quickly.

In real life, your GRiD servers are not going to be merely 1km away and the cost of implementing optical links increases drastically beyond 1Gbps. The switching backplanes in some of the fastest carrier-grade switches and routers operate at less than 1Tbps per slot, so pushing over 1Tbps down a single fiber requires extra optical multiplexing equipment.

Yes, you can push 32Tbps down a single fiber. But you need millions of dollar worth of equipment and a few kW of power to do it. That sort of speed is not going to be economically viable anywhere near the first mile within the next 30+ years.

 

[Alec Mowat]

The speed of light is the speed of light. If we start developing proper fiber infrastructures, instead of the old copper cable networks, we'd overcome the latency really quickly.

In real life, your GRiD servers are not going to be merely 1km away and the cost of implementing optical links increases drastically beyond 1Gbps. The switching backplanes in some of the fastest carrier-grade switches and routers operate at less than 1Tbps per slot, so pushing over 1Tbps down a single fiber requires extra optical multiplexing equipment.

Yes, you can push 32Tbps down a single fiber. But you need millions of dollar worth of equipment and a few kW of power to do it. That sort of speed is not going to be economically viable anywhere near the first mile within the next 30+ years.

Sure, but in real life, we play games online and our computers are not 1K away. Grid makes online gaming more like remoting into a lan. You can get good onsite performance at the cost of poor online performance, or good online performance at the cost of poor on site.

And you don't need a massive infrastructure change. People have been able to push 1 TB over the current fiber infrastructure. It's not the cost, necessarily. It's the desire for change when ISP's are making a fortune nickle and diming.
http://gizmodo.com/the-fastest-real-world-internet-is-1000x-quicker-than-g-1506564949

 

[InvalidError]

And you don't need a massive infrastructure change. People have been able to push 1 TB over the current fiber infrastructure. It's not the cost, necessarily. It's the desire for change when ISP's are making a fortune nickle and diming.

If you read your own link, they aggregated multiple 200Gbps channels to reach that 1Tbps. 100Gbps hardware is not exactly cheap and speeds beyond that are not quite standard yet. Those are not going to become commodity speeds any time soon due to hugely expensive hardware - if you fit a Brocade, Juniper, Cisco or other router with enough cards and optical modules to handle 1Tbps, you are looking at a bill over 1M$ at each end for the routers alone and you still need to add all the optical gear to merge those 10/40/100Gbps streams into a single fiber.

At the more realistic end of the range, 10Gbps hardware is starting to approach commodity level.

Just because it can be done with current hardware does not make it economically viable.

 

[Alec Mowat]

And you don't need a massive infrastructure change. People have been able to push 1 TB over the current fiber infrastructure. It's not the cost, necessarily. It's the desire for change when ISP's are making a fortune nickle and diming.

If you read your own link, they aggregated multiple 200Gbps channels to reach that 1Tbps. 100Gbps hardware is not exactly cheap and speeds beyond that are not quite standard yet. Those are not going to become commodity speeds any time soon due to hugely expensive hardware - if you fit a Brocade, Juniper, Cisco or other router with enough cards and optical modules to handle 1Tbps, you are looking at a bill over 1M$ at each end for the routers alone and you still need to add all the optical gear to merge those 10/40/100Gbps streams into a single fiber.

At the more realistic end of the range, 10Gbps hardware is starting to approach commodity level.

Just because it can be done with current hardware does not make it economically viable.

$1000 consume GTX Titan cards are not exactly economically viable either. It only takes 1,000 sales to hit 1M$.

It seems like Nvidia is investing more in their Grid at the moment. Things are moving into the cloud.

 

[dlf420]

Who the hell has 50 mbs download? That's not available anywhere around here.

 

[TechyInAZ]

Who the hell has 50 mbs download? That's not available anywhere around here.

Yeah i know nobody who has that kind of speed, the fastest speed i've ever seen somebody have is 20mbps down.

However, with google fiber and all the other fiber internet connections coming, 50mbps will be the standard soon.

 

[InvalidError]

However, with google fiber and all the other fiber internet connections coming, 50mbps will be the standard soon.

No need for fiber to get 50Mbps. I can get 50Mbps on VDSL2 (rumors say 75Mbps might be coming soon) or 120Mbps on cable if I was willing to pay $90/month.

For now though, my $30/month 7Mbps ADSL2+ is good enough for 99% of what I do online, so I'm in no hurry to pay more for something I might find genuinely useful less than 1% of the time.

 

[grizzbair]

I read all of the comments, I am up in Canada and we have the infrastructure to run these specs in most of our major cities so I find it hard to believe the USA can't. Yes if you live in rural areas you are SOL for now, but those speeds are getting better soon.

As for the streaming, Yes you can go Steam Box, I almost did. My current PC meets the minimum specs required to be a Steam Box so all I would have to do is load the Steam OS. But it is also my PC so I can't stream it to where it's not anyway, (I don't want to build or buy a separate Steam Box). I currently have it on wheels so I can wheel it to the living room for when my son and I want t have a gaming night, him on his console or PC and myself on mine, but it's a pain. I rather just stream it. Yes you will lose some quality but for me it is an occasional option, not my norm. This may make the Nvidia device may just be the most economical easy solution. All others either cost more o are a pain in the butt. I hate wires! Lol.

My worry is the last time I checked Google an Android TV were very limited in Canada. Has that ben fixed yet? Anyone? Also, yes they need to get some more big titles on board.!!! If they get enough I may be able to get my son to finally give up his console, which aren't very economical anymore. I didn't mind when console were all less $, but now that they cost $500 I find them a waste of $ when other solutions do more for allot less. I have got hi back into PC gaming somewhat, and he seems interested in either the Steam or Shield route, but he I keeping a wait and see attitude so far.

On a separate note, rumors a plenty that Windows 10 on your PC and Xbox One will allow you to stream your Windows game to Xbox One. If so this could be the push back to both Steam and Shield. Just saying.