why don't Microsoft make the cloud to change Xbox one into Xbox two

20 ms is already above the 16.666ms needed to compute a frame at 60fps. And that's only the latency. Even in that case it won't work.

Let's be serious, we argue about the GDDR5's 176GB/s vs eSRAM's 204GB/s bandwidths and then some expect computations sent through a channel  with a 500KB/s-100MB/s badwidth and a latency of 10-300ms to improve the frame rate? The demo MS showed was made on a PC connected to a computer farm via either Ethernet or fiber optics, where the latency is negligible for that purpose and the bandwidth is at least 1Gbps. It's an interesting use of an old technology and it isn't much different from the Uncharted 2 cutscenes rendered on a farm of PS3s.

You're comparing two completely different types of computational processes. Memory calculations happen extremely quickly on local hardware. That information would NOT be sent across the wire (in the same way today, that information is NOT sent over a HDMI cable attached to your television). Local computations are run on the hardware, once processed, they are passed to the output medium. Example:

Code (written by developers) > Calculations (depending on variables such as player movement, calculations are required) > Processor (the calcuations are processed) > Output (the result of these calcuations is passed to your screen).

Neither the code, nor the calcuations ever reach the output. They are processed by the hardware and forwarded to the output device.

It would work exactly the same with a server. The server processes the calculations required for physics/light/AI etc and then the output is passed to the end user via the network (internet). The only difference between the technology used by game streaming and Crackdown's tech is that instead of passing the raw output (a video stream), the servers are passing a partially computed result to the end user device for futher processing.

With regards to 'can this tech improve frame rate?', of course it can. It cannot be used to improve the frame rate of existing games because they simply aren't designed with this technology in mind. New games however can take advantage of this technology. Crackdown is the perfect example. If it wasn't for the power of the servers running the physics calcuations the frame rate on the Xbox would grind to a hault. Instead, it is able to run more fluidly as some of the calcuations are offloaded to a seperate device (servers). Phyics is just one area this technology can be used. As other companies have demonstrated (including NVIDIA) it can be used to calcuate some lighting informaiton, physics, AI etc.

The real question is not CAN it be done, but WILL it be done. Most developers simply will not entertain the idea of using cloud computing for a long time due to:

• Development budget - it's new tech, it will cost more to learn and develop for.
• Audience - not everyone has a good enough internet connection and currently, Microsoft is the only platform holder with this technology so what 3rd party in their right mind would limit themselves?
• Risk - It's new, the tech is uncertain, it's capabilities are uncertain, publishers are never keen to take risks when they already have a working formula.

I think it may have been mentioned before, but even if latency wasn't an issue the costs of this tech for widespread applications across a bunch of games are not going to be viable as a business, unless you can get each player to pay a monthly fee for it or a big one off yearly one, extra alongside their XBL Gold Membership or some all inclusive service that includes both XBLG and Cloud Compute.

MS are not just going to give this away for free, for a game that won't make them any money and even if said game does net GTAV like profits it's still not cost effective to not have some supplemented payment by customers.

Think about if each user requires 20X the power of XB1, then they have to pay for the implementation and maintenance of that piece of hardware, alongside Microsoft's bandwidth costs.
Even buying bulk amounts of processors/RAM and reducing costs down per unit for all of that tech it's still going to cost Microsoft at least a few times the price of an XB1 at retail.
Being conservative say $900 per user for the hardware demands of that one person, over the life of their usage.

Now scale that up to all of the players that could play Crackdown 3 online, even if only 100K people play at any one time $90,000,000 for the hardware alone for that one game, 100K is consistent, average usage, not peak or lowest and that's only for those 100K average users playing Crackdown 3.

Then add other games, say over the rest of XB1's life Microsoft adds a dozen or so other 1st party games that use the tech, all placing similar demands on the tech, that increases the usage 12X+ higher $1,080,000,000.

Any 3rd party games that use it will have similar demands or perhaps even great if more people use it or the features required increase per game session.
Microsoft would have to claw back at least a few billion dollars just to break even.

Say we're conservative and we say it's $2B, divided over a few million gamers, they'd have to pay back about a thousand dollars over the life of their console, say the rest of the generation lasts 5 years, that's $200 per year, per user.
Most people aren't happy with the costs of XBL Gold or similar services and this is without Microsoft adding anything to generate a profit off of it.

You can't even say "well the game's going to make a profit from copies sold", because that goes to the developers and Microsoft for the costs of developing the standard game.
Cloud usage is a separate venture.

Any additional users would be required to pay for their own cost of usage.
Even if Microsoft only operated on thin margins of say 10%, it's only going to net them $200M for a massive gamble.
If MS adds a small margin like that then it adds to the negative appeal of this whole thing.

Would you be willing to pay $220 a year for 5 years, with no price reductions to use this?
Maybe MS can sell it as a Monthly Subscription service, charge $20 a month per person and say, "for $20 a month you get 20X the power of XBox One at your finger tips", which would be massive marketing spin, because it's not guaranteed to work 100% perfect 100% of the time.
MS could of course add in other features like additional free games, movie and music streaming, data storage, it's going to have to be sold as a thing for more than just cloud processing assist tech for use in games.

Although any additions eat into profit.
TBH cloud assist makes more sense if it adds a fraction of the boost we're talking about it here, because maybe Microsoft can get away with charging less, they could add say a few dollars per month to XBL memberships and then give users one or 2 XB1's in the cloud, more users would share the costs, which brings pricing down for each individual.
It's still going to be a huge gamble from a business perspective, but it's a smaller one.

People assuming that every XB1 user can have access to 20X the power of XB1 for free whenever they want are going to be sorely disappointed and if you think every game can use that at once then well you're living in a dream world.
For that you'd have to take the costs a multiply it a whole bunch.

Not even remotely likely to be feasible, even for Microsoft and their deep pockets.

Crackdown is indeed the perfect example for the relatively small requirements in bandwidth to update geometry and collisions of angular building pieces.

To help the GPU you need a lot more data. Cloudlight photons already proved to be up in the 40 mbps range after using lossy compression. Light and shadowmaps are huge, especially if you want to see a significant improvement at 1080p resolution. Other things are very latency sensitive like reflections. Using cloud computing for ray tracing is very possible, yet only if you send the completed image to the client. There's no point in sending intermediary results as it's just too much data.

Physics has it limits too. It would be cool to have realistic collision physics in car games. Yet considering car models are already over 100k polygons, a multiple car collision that needs to update all the cars easily comes in to 12 mbps per car and that's a very conservative estimate. (eg a quarter of the car is affected in the collision, 25k polygons, assume you can compress it to 1 byte per polygon to describe the displacements, 25kb per frame x 60 = 12mbps) A big pile up would stump the best of internet connections. The same goes for fluid dynamics, way too much data. I don't expect a larger version of From dust to be able to run through cloud servers. Too much data to transfer.

What it is suited for is AI, which is as old as the first mmorpg, and a few things that rely on small yet hard to calculate data sets like Crackdown 3. And simply for sharing data like Drivatars and community created content. Now combine it all together to make a persistant slowly evolving world. RPGs with actual seasons and weather affecting and wearing away the terrain. You'll have a relatively big download when you log back in, yet while its running the changes should be gradual enough for the bandwidth to keep up. Godus is such a game, too bad that's about the only thing it has going for it. Eco will be such a game, and hopefully future rpgs will go this route too. I had expected mmorpgs to be there already, yet WoW's popularity has held innovation back, plus the enormous costs involved in setting up such a risky project.

So I fully agree with you final points.
- Expensive to develop. And also to rent or maintain all that extra server time compared to a static game. (which already requires massive resources)
- The world average internet speed is 3.9 mbps. It's higher in developed countries, yet also going through wifi shared with a whole bunch of devices.
- And indeed risk. THQ went bankrupt after Red faction Armageddon. Total destruction didn't help to popularize the game. Godus is in shambles, Eco sim kickstarter is only at 19k of 100k (408 backers...)

Again I don't think people are questioning the future potential of this technology but trying to market it now as a serious enhancement to consoles especially the xbox one seems premature. Happy to see it done though as an xbox one owner.

Dare I say some of the pro comments seem to be those who are keen on xbox'es and not so keen on playstations and who seem to be pushing this as a solution to the xbone's weaker spec. I feel the ps4's dominance in console spec will remain for sometime and this is just a marketing exercise for now.

Actually people did not have a big issue with the console online all the time, they had a problem with the 24 check in.  It was the 24 check in that people cried about the most and MS did not explain properly why this was needed.

Really? That's a demo limited to one building. Am I goint to need 1000x pc's to do an open world city, if so that's really bad. Really really bad.

The 2015 footage uses bullets...and is a game.

So is the demo in the video you and Snoopy posted several times.

I posted a video demonstrating how cloud could increase the framerate.

PC still has a long way to go before we're there.