Yoshida confirms SCE working on new hardware

[aaronspink]

Sony is screwed. So, we can pretty much assume that even if Sony went with Nvidia for the ps4, we won't see backwards compatibility?

Unless Sony forgoes Fermi and asks Nvidia to enhance the RSX.

BC has and always will be pointless. It really serves no purpose and provides no benefit. You significantly increase the costs of your next gen product so that people who already have your previous gen product won't buy new games for your next gen product.

Even for the end user its pretty pointless, if you want to play your old games, why are you buying a new console? Or better yet, why aren't you playing your old games on your old console which has by definition 100% compatibility?

 

[aaronspink]

We will see in a couple of years or so.

There are over 2000 PS2 games, hardly all of them are worth remastering in HD, but selling them discounted on PSN just like the PS1 games would be like money for nothing.

And that can be achieved just as easily via a re-compile.

 

[aaronspink]

So we have the alternative of homogenous multicore CPUs and heterogenous multicore CPUs. Homogenous multicore offer easier programming while heterogenous multicore like the Cell offer better raw performance/die size and better memory utilisation through asyncronous memroy handling. Which is best for you depends on what your priorites are.

My priorities are good games and lots of them and considering that, CELL is a boat anchor. Ease of programing and ease of performance pretty much always trump peak performance. This has been proven time and again. The harder and more complicated a device is the worst it does across a variety of workloads. For the vast majority of developers, the SPUs are dark silicon. Those that have found use for them have had to contort their programs to get use out of them. Even on workloads that CELL should be easier to program for (scientific applications, etc), it has been proven to be fairly lack luster with the programing and data movement complexities causing it to achieve real performance worse to on par with conventional multi-core designs at the cost of much more developer time. In the world of infinite resources, something like cell or other complex FP array architectures might make sense, but back in the real world we don't have infinite resources.

Going with the mature libraries for the PS3 and all the investments that have gone into that, I don´t really see why they would give up on Cell for the next generation of consoles. But I don´t know when the PS4 is planned either.

Any benefit of the libraries is easily portable to standard cache based multiprocessors.

 

[aaronspink]

Also, I wanted to add that I don't think the PS3 was a failure, Sony sacrificed the PS3 with a new disc format and they won the format war because of this. I think that if the PS3 wasn't around, HD-DVD would've won.

Some wars are worth winning, some aren't. Next gen physical media? eh, not so much. BR will never be as successful as DVD. Sacrificing PS3 for BR will be looked at historically as fighting yesterdays war at the cost of today's war.

 

[willardjuice]

Some wars are worth winning, some aren't. Next gen physical media? eh, not so much. BR will never be as successful as DVD. Sacrificing PS3 for BR will be looked at historically as fighting yesterdays war at the cost of today's war.

But it's not like they are doing horrible in the console arena. And while blu-ray won't be as big as dvd, the combination of their successes in blu-ray and gaming could be worth "winning the war". Or maybe it won't; I think it's just too early to make a definitive statement.

 

[Ninjaprime]

But it's not like they are doing horrible in the console arena. And while blu-ray won't be as big as dvd, the combination of their successes in blu-ray and gaming could be worth "winning the war". Or maybe it won't; I think it's just too early to make a definitive statement.

While thats true, they are 3rd, of 3 consoles, last place isn't that great, despite not being horrible. Personally, I think Cell alone is to blame for them being in last.

 

[patsu]

I don't know if one can blame a technology solely when he/she fails to meet business goals. It'd usually be called an "excuse" during performance review.

For this would mean that Sony marketing and other business people did not make any major mistake. This would also mean that those amazing PS3 developers did not make an impact. Personally, I think the PS3 developers should give themselves for a good hard pat on the back for what they have delivered collectively so far.

A large part of Sony's woes was (is) also organizational issues, and economic realities.

 

[Akumajou]

The reason it was costly this gen was because PS3 launched at $600 and the games didn't look any better at launch. If PS4 games look better and the console costs the same as the next xbox, I think it can launch a year later and still be very competitive.

't

I don't believe price was the only reason (of course there were many) however the gaming media and online blogers and forum boards had way too much opinion going into the generation shift, I went through a couple of old magazines but specifically EGM (highly influencial print magazine back then) on the X360 launch magazine you could find alot of arguments for buying or spending the $399.99 because later you could trade it for something better while in the same magazine they had a cross promotion with Microsoft to give away consoles, later when PS3 launched there was a change in direction to OMG $600 is too much while ignoring that BC and Blu Ray was there and elsewhere on the interwebs certain writters for certain sites arguing how blu ray was not needed or did not add anything to game consoles and focusing too much on "quotes" by Sony reps on PS3 to dedicating an entire magazine arguing somehow that people should not have faith in the new console.

The idea that games didn't look any better or that there were no games is an idiotic type of reasoning or argument because the PS3 was launched withing a week into world wide market unlike PSX, PS2 and other consoles of the past that when they were finally launched outside of Japan, they had a fairly numerous offering of solid games.

This is something that Sony needs to realize, they cannot do another world wide launch within weeks or they are going to once again play into the hands of the naysayer, Microsoft, the anti-Sony media and forum trolls that spread FUD.

They really just need to worry about a Japan launch, then 7 to 9 months later make a US or North American launch and then after a 2 more months go for the rest of the world and they will have a solid number of games so when arguments are made against their next console AND THEY WILL the number of games will be less of an issue.

Sony cannot go around copying microsoft when it comes to launching a console, if MS can do it fine but Sony does not have an endless income from monopolizing the computer OS market.

 

[Akumajou]

While thats true, they are 3rd, of 3 consoles, last place isn't that great, despite not being horrible. Personally, I think Cell alone is to blame for them being in last.

I don't think we can expect a company to maintain being number one and even if they are in 3rd place it does not really make them the one to fail, Sony has put up an amazing fight even with CellBE and the EE before it.

However there is the trend that is very popular in this gen that despite reaching new graphics watermarks like Sony's devs have, the ones countering what they have done have tried to debunk by saying things like oh Killzone 2 had a huge budget, GT5 had a huge budget, RSX is still inferior or crap yet Sony must be using CGi movies when showing Uncharted 2, the excitement of 3d gaming that made it very interesting in the late 90s and early 2000s is gone, replaced with oh I don't think it looks or plays better than CoDMW2 and 32 player matches or 60+ is just too much, I rather stay with 12 player matches /sarcasm.

(its like saying hey devs, don't work hard, its not worth it, just grab a cookie cutter and they will eat it up)

 

[specwarGP2]

But they won't look better, first party games aside !!

If Sony let Microsoft get a one year head start, the next Xbox will lead, not only in numbers, but also be lead platform for many cross platform games the way it was this gen.

Cheers

well if Sony had a bunch of its 1st party studios make launch games that would definitely give the impression that PS4 games look better! :smile:

The numbers thing isnt really that big, its like only a 5m lead ww and 360 was the lead platform because PS3 is much more difficult to develop for. Also if Sony launched a year later it would have more flexibility with the console specs.

Sony got caught off guard with PS3 this gen, and I'm sure they will step their game up (thanks marcus lol) with PS4. Sony does make a bunch of dumb mistakes, but they do recover pretty well from them. I mean seriously, sony launched a $600 console, and they are still competing head to head with MS.

 

[Crossbar]

And that can be achieved just as easily via a re-compile.

Just a re-compile? Really?

BC has and always will be pointless. It really serves no purpose and provides no benefit. You significantly increase the costs of your next gen product so that people who already have your previous gen product won't buy new games for your next gen product.

Even for the end user its pretty pointless, if you want to play your old games, why are you buying a new console? Or better yet, why aren't you playing your old games on your old console which has by definition 100% compatibility?

Maybe BC is pointless to you, but it´s obviously very important to others since every console manufacturer is doing it one way or the other. I think it grows in importance with the size of your previously installed base. I bet that the next-gen consoles will have BC as well, do you bet against that?

Any benefit of the libraries is easily portable to standard cache based multiprocessors.

But that would mean less cores/die space so why should they back-peddle when they already have the libraries in place?

 

[Arwin]

There is nothing that a control store gets you that you can't get in a cache based design with added flexibility on top. Cell is fundamentally an architectural dead end and it was even before it saw first silicon.

Ehm, I am going to respectfully disagree here. Performance-per-watt and data streaming performance are superb on the Cell. It's very efficient to have such a simplified per-core 'cache' and designing many-core CPUs with a better system may be possible, but as far as I know we're still waiting to see it?

 

[Squilliam]

't

I don't believe price was the only reason (conspiracy against Sony) .

I don't think that you can explain the performance of the PS3 as some huge conspiracy or justify a lack of performance against other consoles which managed to release far more compelling content on day one. Sony screwed up and then people turned against them. The same thing they took advantage of against the N64, its just history turning full circle again nothing more.

 

[aaronspink]

The idea that games didn't look any better or that there were no games is an idiotic type of reasoning or argument because the PS3 was launched withing a week into world wide market unlike PSX, PS2 and other consoles of the past that when they were finally launched outside of Japan, they had a fairly numerous offering of solid games.

They really didn't have any games and what they had really didn't look any better.

This is something that Sony needs to realize, they cannot do another world wide launch within weeks or they are going to once again play into the hands of the naysayer, Microsoft, the anti-Sony media and forum trolls that spread FUD.

Or maybe they need to make a console that doesn't suck to develop for. You know, so they can launch with more games and those games will actually look and play decent.

They really just need to worry about a Japan launch, then 7 to 9 months later make a US or North American launch and then after a 2 more months go for the rest of the world and they will have a solid number of games so when arguments are made against their next console AND THEY WILL the number of games will be less of an issue.

And then you'll complain that they are failing because it took them 18 months to actually launch the console after everyone else. The failings of PS3 can be put down to two things: CELL and BR.

 

[aaronspink]

Just a re-compile? Really?

Its a whole lot easier than you think.

Maybe BC is pointless to you, but it´s obviously very important to others since every console manufacturer is doing it one way or the other. I think it grows in importance with the size of your previously installed base. I bet that the next-gen consoles will have BC as well, do you bet against that?

BC is pointless, not only to me but to the vast majority of the market. You either already have the previous gen in which case it is pointless, or you don't so you don't have the game to use it with in which case it is pointless.

Lets put it this way, are you willing to spend the cost of both a current gen and a next gen console to get backwards compatibility? If yes then you don't need it, just buy both, if no, then there is no way to justify it because it basically adds that much additional cost.

BC hasn't seemed to have hurt 360 nor PS3.

But that would mean less cores/die space so why should they back-peddle when they already have the libraries in place?

because they will get better performance on the vast majority of workloads with less resources. Peak != delivered.

Or in other words, a whole lot more things can run on cpus than on SPUs and CPUs can run everything that an SPU can. As I've said in other forums, I can deliver today and infinite flops processor for limited workloads, do you think it would be a good idea to incorporate that into the next gen? It would allow infinitely higher performance in select workloads!

 

[aaronspink]

Ehm, I am going to respectfully disagree here. Performance-per-watt and data streaming performance are superb on the Cell. It's very efficient to have such a simplified per-core 'cache' and designing many-core CPUs with a better system may be possible, but as far as I know we're still waiting to see it?

Perf/watt effectively the same if not better for a normal cache. And streaming performance is effectively the same IF NOT BETTER. Oh and guess what, more games will actually use it and more games will get better performance because of it.

 

[Silent_Buddha]

well if Sony had a bunch of its 1st party studios make launch games that would definitely give the impression that PS4 games look better! :smile:

The problem there is that all the really good looking 1st party titles take quite a few years + a lot of cash investment to develope. Unless, of course, you're already leveraging work done previously (KZ3 leveraging work done on KZ2 for example).

All of which means that your launch titles are almost always going to be the least impressive of the titles for that console generation. It's pretty unavoidable unless the next generation isn't a big leap over the previous.

Applied to the PS3, imagine how much worse it'd have fared if Sony waited until KZ2 was ready before launching? GT5? GoW 3? Those titles take lots of time and money which is just compounded when you have a completely unorthodox system you basically have to learn from scratch. Where virtually everything you want to do has to be redone from scratch.

It's no wonder good looking titles appeared on the X360 first. But not a surprise that better ones eventually appeared on the PS3. You want better, spend more money and take more time.

Regards,
SB

 

[patsu]

because they will get better performance on the vast majority of workloads with less resources. Peak != delivered.

Or in other words, a whole lot more things can run on cpus than on SPUs and CPUs can run everything that an SPU can. As I've said in other forums, I can deliver today and infinite flops processor for limited workloads, do you think it would be a good idea to incorporate that into the next gen? It would allow infinitely higher performance in select workloads!

I think the devil is in the details. Peak vs delivered depends on the application and programming effort. The problem with Cell is that it requires significantly more effort to develop for, but it is possible to get better utilization (than the generic cases) because the developer has more control over the architecture. I remember reading one of the earlier papers that mentioned this.

What other CPU can deliver at the computational power and heat level of Cell in 2005 ? In year 201*, a Cell-successor would have a higher technology base and ideas to work with. Some of its earlier shortcomings should be addressable.

 

[aaronspink]

I think the devil is in the details. Peak vs delivered depends on the application and programming effort. The problem with Cell is that it requires significantly more effort to develop for, but it is possible to get better utilization (than the generic cases) because the developer has more control over the architecture. I remember reading one of the earlier papers that mentioned this.

Look a whole slew of FP arrays over the years have had niches here and there where they could get good utilization and therefore performance. But the niches were few and far between and the programming effort incredibly large such that like CELL after them, they are in the dustbin of history as the failures they were and are. It isn't good enough to be able to not suck on one or two things, you have to not suck on most things, and micromanaged architectures just don't do that.

What other CPU can deliver at the computational power and heat level of Cell in 2005 ? In year 201*, a Cell-successor would have a higher technology base and ideas to work with. Some of its earlier shortcomings should be addressable.

The fundamental shortcoming of cell is the memory model and programming model. Those aren't things that will be overcome with technology. Comp and heat level in 2005? There were plenty of other failed architectures from then that did better when they talked about peak. As far as delivered? Cell is orders of magnitude WORSE than any duel core x86 from that era on a variety of workloads. Add in the increase in cores, AVX, etc. There is a reason CELL is dead, kaput, canceled, all development stopped. 5 years later there are still a lot of workloads where CELL is far behind x360 in cpu performance even with all that time to work on programming tools, etc.

 

[patsu]

Look a whole slew of FP arrays over the years have had niches here and there where they could get good utilization and therefore performance. But the niches were few and far between and the programming effort incredibly large such that like CELL after them, they are in the dustbin of history as the failures they were and are. It isn't good enough to be able to not suck on one or two things, you have to not suck on most things, and micromanaged architectures just don't do that.

Since 2005, we have seen Cell Fold@Home, decode & encode H.264 video in real-time, track motion for gaming, recognize speech, perform GPU functions (culling, lighting, postprocessing, AA), tackle scientific computing applications on RoadRunner, execute Air Force R&D apps, power the entire Blu-ray 2.0 stack (Java VM + 1080p main profile AVC video + TCP/IP with capacity to spare), perform financial modelling, hunt for oil fields, crack security keys including the root key for PKI infrastructure, bring up secure kernel for DRM, run assorted game logic including physics simulation, enable DVR + Cell HDTV @ 480Hz refresh rate, support Linux and even a web browser + Flash. It may not be able to do everything under the sun well, but for a media box, it's not bad.

Which CPU in 2005 can do the above at the same performance level, heat envelope and cost as Cell ?

The fundamental shortcoming of cell is the memory model and programming model. Those aren't things that will be overcome with technology. Comp and heat level in 2005? There were plenty of other failed architectures from then that did better when they talked about peak. As far as delivered? Cell is orders of magnitude WORSE than any duel core x86 from that era on a variety of workloads. Add in the increase in cores, AVX, etc. There is a reason CELL is dead, kaput, canceled, all development stopped.

Sure, but who says Cell 2 can't evolve to incorporate more modern programming model when it needs to ?

5 years later there are still a lot of workloads where CELL is far behind x360 in cpu performance even with all that time to work on programming tools, etc.

And there's a lot of workload where 360 and x86 falls way behind Cell.