PlayStation Now - could Sony go after Valve?

[Rurouni]

Cool, article says servers are in southern California, right in my backyard so this should work well for me. It also says running games at 60fps as done on pc to reduce latency can't be done on ps3 but I wonder if that is really true. They could always try faster ps3 hardware and run games at 60fps, but only render every other frame since it's not the visual speed one wants in this case but just the logic and input. That could shave off a bit more latency, and I figure how hard can it be to build a faster cpu side ps3 8 years after the fact.

Since they use cell for compatibility purposes, I doubt that they can really push the architecture faster (clock it higher). Or combine it with an architectural tweak to gain higher IPC? Can the games utilize it?
How about using faster gpu? Can PS3 games take advantage of higher shader count? Can they use faster RAM or better yet, change the memory controller so it can use cheaper/more available DDR3 RAM? Someone definitely need to ask Sony on what hardware tweaks they've done for PS Now.

 

[dagamer]

Since they use cell for compatibility purposes, I doubt that they can really push the architecture faster (clock it higher). Or combine it with an architectural tweak to gain higher IPC? Can the games utilize it?
How about using faster gpu? Can PS3 games take advantage of higher shader count? Can they use faster RAM or better yet, change the memory controller so it can use cheaper/more available DDR3 RAM? Someone definitely need to ask Sony on what hardware tweaks they've done for PS Now.

Messing with either the CPU or GPU is going to introduce some very terrible bugs and since developers aren't interested in patching old games, it's not worth pursuing.

 

[DieH@rd]

Cool, article says servers are in southern California, right in my backyard so this should work well for me. It also says running games at 60fps as done on pc to reduce latency can't be done on ps3 but I wonder if that is really true. They could always try faster ps3 hardware and run games at 60fps, but only render every other frame since it's not the visual speed one wants in this case but just the logic and input. That could shave off a bit more latency, and I figure how hard can it be to build a faster cpu side ps3 8 years after the fact.

It wont happen. Some games are surely based their engine on specific timings. Like latests NFS Rivals, making a game run in 60fps would mess up countless things..

 

[joker454]

It wont happen. Some games are surely based their engine on specific timings. Like latests NFS Rivals, making a game run in 60fps would mess up countless things..

They don't have to do it for every game, just for more twitch heavy games that really need low latency and would benefit from a few more milliseconds saved, and assuming they test ok at a higher frame rate. It's not like they run off retail ps3's, Sony has full control of the rack mounted ps3's and could even upclock/downclock in realtime as needed per game. It's a long shot but I imagine they will need to get very creative to fight latency. Many games should work ok with this as during development it's not uncommon to have vsync off so frame rate will already fluctuate far past 30fps in some cases and the code didn't break as a result. Anyways it's easy to test, run a game off a kit where vsync is stubbed out to do nothing and you will know quick if the game can work or not at higher frame rates.

 

[-tkf-]

How is it with the RSX, could they in theory create a version where it outputs directly to a h.264 hardware encoder instead of a a frame buffer(?)

Or is that likely to be super expensive..

 

[joker454]

How is it with the RSX, could they in theory create a version where it outputs directly to a h.264 hardware encoder instead of a a frame buffer(?)

Or is that likely to be super expensive..

They have to let the games continue to write to a frame buffer because that buffer gets used repeatedly within a given frame in all sorts of ways in game code. Only on the "buffer flip" call would the hardware finally know independent of game code that the buffer can get safely displayed and h264'd. I don't think you can do hardware changes beyond this because for all we know the game may still make use of this last frame in the next frame or two, think of the "you are drunk" visual effect where they use the past few frames for example, so you can't really mess around with how the frame buffers are handled and laid out after the fact. That's why I suggested the overclock because all the hardware is kept identical. The kit could keep rsx and vram the same, and use faster cpu and main ram for games that work ok with that. Playstation Now can determine whether to run the kit stock or in high speed more on the fly, and since it's all rack mounted stuff in a server room they could cool it any way they like. With all the node drops that have happened since 2005 I figure this may be possible. From what I recall the timing issues wouldn't be too bad on the ps3 because the intensive cpu stuff was handled entire on spu in it's work ram, and gpu side it would usually just be doing "jump to self" calls to stall itself waiting until the cpu had removed that stall so it could continue, in other words the timings were not tied to machine ghz.

 

[Brad Grenz]

This post may not be solely directed at PlayStation Now but other [gaming] streaming services as well...

Is streaming-service the future? I simply don't see it. If we take PS-Now for example, which seems to be based off a server-farm of PS3s, you would need at least a console within the farm/cloud for a user at any given time. If you have 10'000 users, you need at least 10'000 virtual PS3s outthere.

Now at which point would you rather have 10'000 subscriptions rather than 10'000 sales of PS3s? If a subscription for a year costs $100 (and the cost for a consumer of a PS3 is what, $149?) I guess it would make sense after 1.5 years, but then at this point, you still haven't factored in that your streaming-service also includes an entire game library, so when going subscription service, you are also losing out on lots of software sales that your consumers would otherwise buy.

Or will PS-Now cost at a 'per play/game rate'? E.g. similar to rending a movie-stream, you're effectively paying $20 for a game for a week. Once the week expires, you have to rent it again?

I just can't for the life of me come up with numbers where running a server-farm full of PS3s to offer a streaming service actually outweighs the benefit of just selling consoles. And after all this, you still are basically living with a second-rate experience short of the real thing because you are playing a highly compressed with added latency stream.

Streaming services might work with movies, TV shows etc where you don't need dedicated 3d processing hardware to "stream" and the cost is probably a minor fraction than what you would need if you wanted to create a gaming streaming network...

Anyone care to enlighten what I'm missing?

Do you think your local gym buys a treadmill for every single member? That 10,000 PS3 servers could probably service 100,000 subscribers. And you're making the mistake of assuming the potential lost sales are greater than the potential subscription revenue.

 

[London Geezer]

Do you think your local gym buys a treadmill for every single member? That 10,000 PS3 servers could probably service 100,000 subscribers. And you're making the mistake of assuming the potential lost sales are greater than the potential subscription revenue.

Very true. Not that I know anything about the business, but logically I would think that using some statistics they will base their need for hardware over how many people in each territory play at any given time, between 4pm and 6pm on a saturday. Versus the number of people wanting to use the service at other times.
Also users (think globally) need to sleep and work or go to school, all activities that take up a large part of the day (say 8 hours each approx) - so a huge chunk of them won't be using it at any given time. Let's not start on people's need for real social interaction and - shock horror! - sex! Which might take 5 minutes for some but a lot, lot longer for others!
There's so many variables, and it might be a little complicated to know the exact number but they definitely will never need a PS3 for every registered user. Nowhere near.

 

[Phil]

Do you think your local gym buys a treadmill for every single member? That 10,000 PS3 servers could probably service 100,000 subscribers. And you're making the mistake of assuming the potential lost sales are greater than the potential subscription revenue.

That is of course true. Though there's a point that the more successful your service becomes, the less the gap will become of required PS3 servers needed to service active users.

At some point though, you are weighing up selling subscriptions vs. selling consoles (if we assume that at some point, you are able to use the service on non console devices that is). The subscription surely will have a higher initial cost, given you have to service the 'computing farm' and somehow figure out how you sell your software. If you offer your software at a flatrate included into the subscription, how will publishers get their royalty? Or will we be looking at a subscription fee and a pay-per-play-session micro-cost?

 

[DieH@rd]

Steam has ~70+ million of active subscribers, but concurrent numbers of users who are playing the games are 10 times lower.

If Sony can create tall server cabinet that houses 50 PSNow 1U racks, they would need only 50 of those cabinets to serve 20 thousand playing users. Smack that kind of setup into every US city and EU/Asia region, and after 50 locations, they will have ability to serve 1 mil active players. But, they will most probably go with smaller numbers. I don't expect them to attract multi-million subscriber base easily. Pricing, catalogues and individual user experiences will determine much of its success.

 

[Phil]

Steam has ~70+ million of active subscribers, but concurrent numbers of users who are playing the games are 10 times lower.

Correct me if I'm wrong, but Steam isn't exactly streaming your gaming feed. From what I understand, it's a different business model; you sign up to their service and get an automated service that mirrors a more plug & play experience in that the service downloads the game for you and simplifies the installation and gaming process etc, but games are run and executed locally. To run this kind of service, you don't have to invest millions into dedicated processing hardware.

A gaming-streaming-service is quite a different beast to run given the difference in costs and expenses. In a streaming service, there's a limit on how many concurrent players you can have (limited by the size of your farm). Because you also have to fight latency, you can't afford to build up a huge farm in the NA and service gamers on the other side of the globe either - you would have to carefully invest and build up data-centers in strategic places to have a service that *works* (works for the most part meaning that latency is acceptible, which I am still not convinced of - given I already find a local remote-play lag to be bordering on unacceptable [judging PSP-PS3, not Vita-PS4 mind you, which I know is better]).

 

[DieH@rd]

Yes, Steam needs much less processing. But my point was more aimed at "subscriber level vs concurrent payers" problem. They wont need to create batshit insane server facilities to serve 1 mil subs.

 

[Phil]

Yeah, I get that and I'm not doubting that it would be a different ratio in the case of PlayStation Now.

My point is rather:
In order to keep latency low, you can't build up one huge (efficient) server-side farm to service all your customers world wide. In order to get a satisfactory (sellable and marketable) service, they would need to have multiple farms worldwide to reduce latency. for simplicity sake, you could say one in NA, one in Europe and one in Pacific/Asia. Or for every continent one. So in the end, the costs are higher and the ratio lower - where as with Steam, because you are only offering what's effectively a "download service", it could very well be serviced from one location and if needs be, because costs to offer additional download servers, adding a few locations won't necessarely increase costs substantially.

Because costs are higher and the ratio is larger for active users vs. server nodes, the whole initial cost is higher. And I'm simply wondering if in the end, all these resources in creating this service, if it will be wasted because in the end, it won't work properly (due to latency) for a majority of customers and the whole thing will be a loss-making business. It also raises the question, what kind of numbers they would need to actually brake-even.

I'm actually interested in throwing around a few hypothetical numbers to see after which points such a service could "work" and pave the way for a future where perhaps consoles are not required anymore and most people play on streaming servers.

From my point; i think it will take *a lot* of subscribers to actually prove to be more profitable than a more historical model of selling expensive consoles and per-game-sales with a required online subscription to play games... but as I said, I may be lacking some foresight or some calculations...

 

[Phil]

EDIT:

Just to add some clarification to my post:

The difference between Steam and a service like Now simplified:

- Steam has relatively fixed costs. Even if they had way more subscribers, their fixed cost would still be more or less the same, give or take some additional download-servers to satisfy demand.

- The more subscribers PS Now has, the higher the costs to maintain a run of the service. More subscribers = more active users = more nodes required to satisfy demand.

I think that's quite a substantial difference. And the other point about having multiple farm locations:

Lets say you start with a farm capable of servicing 1 million subscribers. If we take a ration of 1/10 as with steam, you would need around 100'000 nodes (1 node = active user). Now unless all those 1 million subscribers are in one location, you would need to split those nods into the 5 continents to make the service acceptable due to lag. So in effect, depending on size of the region, you would split them up. For simplicity sake, you would have 20'000 nodes on each continent that could potentially service 200'000 users. But that might be cutting it, so in practical sense, you would maybe have more nodes but a lower ratio of active users = meaning proportionally higher costs per subscriber.

This raises some interesting questions:

In order to keep latency "acceptible" - how many farms are we loooking at? One in each continent might be pushing it in regards to playability, but then, I might be a bit to skeptical in regards to what is considered "playable" and what isn't.

What is the cost of such a farm?

What is the business model? Pay subscription and be able to play everything? Or pay for subscription and pay-to-play / or pay-for-game? What would the price have to be to achieve a break-even scenario?

 

[DrJay24]

But with Steam you pay $5.00 and play a game for months. With Now you play a game and pay by the month ($15/m likely), so the revenue is going to be higher.

 

[Nesh]

But with Steam you pay $5.00 and play a game for months. With Now you play a game and pay by the month ($15/m likely), so the revenue is going to be higher.

I doubt its going to be that much

 

[DieH@rd]

From the moment Onlive and Gaikai started trading blows, Onlive was focused on very large server farms [2 for US coasts, 1 for UK, maybe another one launched later in central EU], and Gaikai on dispersion of smaller servers.

EU userbase heatmap - http://www.edge-online.com/features/david-perry-interview/3/
US userbase heatmap with 12 server locations marked - http://www.engadget.com/2010/12/02/gaikai-enters-closed-beta-we-get-an-exclusive-first-look/

As of Dec 2010, Gaikai had 23 servers worldwide [probably only US + EU]

 

[Phil]

But with Steam you pay $5.00 and play a game for months. With Now you play a game and pay by the month ($15/m likely), so the revenue is going to be higher.

Is that subscription fee? With or without game flatrate?

If this assumes a games flatrate, I don't see how $15/m would cover even a micro fraction of the costs, even if they had 1 million subscribers (100'000 active users at any given time).

1 million * 15 * 12 = 180 million dollars a year
costs of 100'000 ps3s (including all the costs for the farm, including bandwidth, power, infrastructure etc) = ???
what about the royalties for the games? I'm sure game publishers would want a slice of the pie...


As I said above; Steam isn't a streaming service. The costs are relatively fixed. That's why they can get away with $5.00 or whatever they are charging. Even if PS3 hardware is now relatively cheap, especially if it's 8 on a board, it will still be substantial since you also have to account for the server-farm infrastructure, people to run it, power, bandwidth etc...

And if we are to compare a streaming model vs. a traditional console model for the future, the revenue would need to a lot higher to sustain the same amount of profits that they have now.

In a practical sense, I agree that $15 a month is likely, but I'm willing to bet that at this price, the service will be operating at a loss for a long time... and this is assuming the service runs well enough that it even gains some traction....

 

[Arwin]

While we're on the topic of Steam, can't Valve do something about those stupid The HomeGuard group invites? Grr.

 

[joker454]

Steam has ~70+ million of active subscribers, but concurrent numbers of users who are playing the games are 10 times lower.

If Sony can create tall server cabinet that houses 50 PSNow 1U racks, they would need only 50 of those cabinets to serve 20 thousand playing users. Smack that kind of setup into every US city and EU/Asia region, and after 50 locations, they will have ability to serve 1 mil active players. But, they will most probably go with smaller numbers. I don't expect them to attract multi-million subscriber base easily. Pricing, catalogues and individual user experiences will determine much of its success.

Thing is Playstation now is supposed to work on computers, tablets, phones, tv's, etc so it's active user base can potentially be orders of magnitude more than any existing gaming device or service like Steam. The other added complication as I've learned with my websites is people always seem to be playing around with their ip's to make themselves appear to be in cities or countries that they are not actually in. Not a huge deal for typical online play but that may cause some havoc when Playstation Now tries to determine which data center should be used for a given user.