Yoshida confirms SCE working on new hardware

[Elan Tedronai]

Sony will make the biggest mistake of their lives if they dump the Cell for next gen. Reasons

Waste a lot of R&D money moving to a new processor. Wasted alot of money developing the cell. Sony still owns the Cell along with toshiba and IBM

completely and utterly fuck up the time and effort in making cell libraries and APIS. So all that beautiful MLAA stuff would be utter wastage too. Imagine if you are PD, ND, GG etc all having to start from scratch again. More money and time wasted again. when you could simply continue where you left off last gen.

No BC. So no GOW 3 or GT5 playable on the PS4

Completely fuck up the firmware APIS as well. The ps3 security relies on one isolated Spu to keep its secrets intact. Completely closed to the rest of the system. Moving to a new CPU means wasting a lot of time and effort for new security architects


Sony does not need to pay for a new processor or a new optical media since Bluray and the Cell are beholden to them. The smart thing to do would be to build upon what they have now. Then again this sony. who knows where they will head off next.

 

[dobwal]

Sony does not need to pay for a new processor or a new optical media since Bluray and the Cell are beholden to them. The smart thing to do would be to build upon what they have now. Then again this sony. who knows where they will head off next.

Agreed.

What Sony needs to do it put the same emphasis in terms of capital investment and engineering on developing a super GPU called "Nucleus" to pair with Cell. At least the first place what that effort will show up is where the playstation is primarily meant for, playing video games.

 

[aaronspink]

How do you recompile a binary?

The same way people have been doing it for almost 50 years: compiler program.lang -o program.binary.

 

[aaronspink]

And yes, I am also strongly convinced that the state of play has changed, not just because programmers have more hours on the Cell under their belt, but because programming PCs and particularly modern GPUs has become more and more like programming the Cell.

No, no it hasn't. There is ample evidence against this and it contradicts everything you are saying. Programming for Cell and other acceleration architectures still takes a disproportionate amount of time to get results that are largely no better and in many case worse than the results on conventional CPUs. Take a look at the work done by prace for example.

 

[aaronspink]

Waste a lot of R&D money moving to a new processor. Wasted alot of money developing the cell. Sony still owns the Cell along with toshiba and IBM

Cell is already dead. All three have stopped further development of the product. IBM has basically pulled out completely, sony and toshiba are supporting it only as a legacy product. Its dead, has been for quite a while.

completely and utterly fuck up the time and effort in making cell libraries and APIS. So all that beautiful MLAA stuff would be utter wastage too. Imagine if you are PD, ND, GG etc all having to start from scratch again. More money and time wasted again. when you could simply continue where you left off last gen.

You mean all those libraries and apis to still be behind what people can currently buy off the shelf for cheaper?

Completely fuck up the firmware APIS as well. The ps3 security relies on one isolated Spu to keep its secrets intact. Completely closed to the rest of the system. Moving to a new CPU means wasting a lot of time and effort for new security architects

The only reason they went with that solution is they were smart and realized the SPUs were going to be so underutilized they might as well make use of them somehow else.

 

[Silent_Buddha]

Sony does not need to pay for a new processor or a new optical media since Bluray and the Cell are beholden to them. The smart thing to do would be to build upon what they have now. Then again this sony. who knows where they will head off next.

That's all going to depend on how much continuing R&D is being put into the Cell Architechture. If rumors are true there may be only one company (or possibly none depending on rumor) continuing active developement of Cell where there were originally three.

If R&D funds is lacking then it's entirely possible that Cell's developement won't be able to keep up with the developement of more traditional CPUs in which case it could still end up being a bad choice even with the ability to re-use most of the existing libraries and SDK's.

With how little is known with regards to continuing R&D on Cell, it's a hard call to say whether it's going to be the obvious "good" choice or not.

Regards,
SB

 

[patsu]

Cell is already dead. All three have stopped further development of the product. IBM has basically pulled out completely, sony and toshiba are supporting it only as a legacy product. Its dead, has been for quite a while.

This may very well be true. Whatever Sony decides on, they may have to forge new relationships.

You mean all those libraries and apis to still be behind what people can currently buy off the shelf for cheaper?

Err... they are either open source and free (like the Bullet library, PhyreEngine), or proprietary to Sony and partners (like the MLAA library, Edge tools).

Third party commercial libraries (e.g., physics, gestures, animation) are available on Cell and other platforms too.

The only reason they went with that solution is they were smart and realized the SPUs were going to be so underutilized they might as well make use of them somehow else.

It's designed to be used that way. Kutaragi talked about it before Cell was shown to the public. It's a pretty obvious Cell advantage actually, if you're not blind to its strengths.

EDIT:

No, no it hasn't. There is ample evidence against this and it contradicts everything you are saying. Programming for Cell and other acceleration architectures still takes a disproportionate amount of time to get results that are largely no better and in many case worse than the results on conventional CPUs. Take a look at the work done by prace for example.

They focus on petaflop problems, use up tons of electricity. Their work may not scale down to consoles. In parallel, there are of course GPU-based acceleration work that some labs are looking into.

 

[JPT]

The same way people have been doing it for almost 50 years: compiler program.lang -o program.binary.

I am not a hot C programmer, but are you not a tad optimistic about being able to just recompile something that is very closely bound to the architecture it was written for?
What about all those SPU routines and and moving of data to SPU's for processing when there is no SPU 's to do it? Will the compiler automagically create something that will take care of it or modify the code to not use SPU's?
Since you are not a proponent of BC, I assume there will be no emulator that you can target with you SPU code anyway.

And I assume that long before we actually get to the recompile stage, we would get into the problem of having the source available. Ie for instance Lair, who got the sources to be able to recompile it? The company went bust, if I remember correctly, how many other companies have or will by the time PS4 arrives?

 

[Prophecy2k]

Going back to the original post:

“When Ken Kutaragi moved on and Kaz Harai became the president of SCE, the first thing Kaz said was, ‘get World Wide Studios in on hardware development’,” Yoshida said.

“So he wanted developers in meetings at the very beginning of concepting new hardware, and he demanded SCE people talk to us [developers].”

I think it's pretty obvious here that he's referring to SCE devs and not third parties.

Considering his comment also regarding MOVE being an example of this new platform development paradigm, i'm sure he's really refering to the way in which Sony will employ the use of their own in-house software dev guys to ensure that all new HW platforms are launched with sufficient software; support, dev tools etc.

I'm personally of the opinion (based a little off what i've heard from the dev community) is that Sony's biggest problem with the PS3 at launch, and its issue of not being "easy to program for", was the tools and software libraries released at the time. Compared to what MS put out with the xbox360 from launch, PS3 devs would pretty much have months of work to do to even get to a comparable point in their game's developement than they would have if they'd made their game on xbox360... Consider also the full year headstart 360 devs would have had with 360 devkits, and it was inevitable that devs would rage when they realised they'd have to pretty spend much double the time working on PS3.

Sony really cannot afford to make the same mistake with PS4 or even PSP2 (if there'll even be one), and so employing the services of their in-house coding ninjas and programming gurus makes sense in serving to fashion not only a competitive HW platform but also a competitively-featured software plaform also.

 

[tuna]

The same way people have been doing it for almost 50 years: compiler program.lang -o program.binary.

Wouldn't you need the source code in a buildable state then?

 

[corduroygt]

An algorithm designed to access a very large data set randomly doesn't sound like it'd run very well on any architecture, but yeah, I guess it'd run worse on CELL. Do you have an example, though?

Are you John Hable from ND?

http://filmicgames.com/archives/401

The core problem with CPUs is that if you have a program that takes 2 gigs of memory, you can read and write to any byte anywhere in memory at any time. It’s simply not possible to create a processor that handles this memory access pattern well. For better or for worse, most programs are written with no regard for cache coherency. How many Java programmers that you know think about cache-efficient layouts? AMD and Intel try to make these operations as quick as they can through various caching strategies. This leads to lots of transistors taken up by larger and larger caches, but still results in lots of stalls due to waiting for memory.

 

[boommoob1]

I knew about this 3 years ago. I can remember similiar comments from Kaz Hirai after Kuturagi was ousted. The biggest change with the PS4 is it'l certainly have generic innards. I wouldnt expect dev tools etc to be a problem at all from here on out. Kuturagi's no longer the head of SCEI...

 

[obonicus]

Are you John Hable from ND?

http://filmicgames.com/archives/401

Hah, yes I am, Rings of Power 2 will be amazing

Spoiler

no I'm not

. More like the specific situation mentioned sounds like a textbook worst-case scenario. To make it worse, joker would have to say the problem being solved is NP-hard.

 

[Elan Tedronai]

Cell is already dead. All three have stopped further development of the product. IBM has basically pulled out completely, sony and toshiba are supporting it only as a legacy product. Its dead, has been for quite a while.

The cell is not dead. It's still in use currently. It's a scalable processor. There's nothing stopping sony from proceding to a 32 spu cell for the PS4 in the future if they wanted to. IBM would happily go back and develop one for Sony if they wanted provided they cough up the funds for it. Only this time it will be a lot cheaper than what sony shelled out back in 2005


You mean all those libraries and apis to still be behind what people can currently buy off the shelf for cheaper?

Name me another processor other than the cell where MLAA is possible? The cell can do things other processors cant. Left behind? you're having a laugh. The PS3's weakness is not the cell in case you never heard. it's the GPU which suprise suprise gets a lot of help from the Cell. The problem here was Sony's lack of help and support to developers when the Ps3 launched. Now after 4 years and a mature tech, you would like sony to dump all the hard work, the libraries, APIs etc made by first party and third party devs so that they could 'buy off the shelf stuff for cheaper'? Yes dump the God of war engine for unreal engine. Not sure if it makes sense


The only reason they went with that solution is they were smart and realized the SPUs were going to be so underutilized they might as well make use of them somehow else.

Security was a cornerstone when the Cell was first designed. Please read the IBM papers before making such assumptions. To this day the ps3 remains unhacked because of the Cell. I am sure Sony is quite pleased with it seeing how the psp is sufferring from piracy.

comments in black

 

[Silent_Buddha]

Umm...

The cell basically is dead if rumors of all 3 partners have stopped active R&D on Cell are true. Without continued active funding of R&D it's basically a dead end.

And as to MLAA, any CPU can do it, there isn't anything particularly special about Cell that makes it impossible to do on other architechtures. You may have a valid point that at that clock speed Cell may currently be the fastest at MLAA, but considering other architechtures scale much higher in clock speed and up to 6 (soon to be 8) "real" cores, I'm not sure that's going to be a very good argument for the next generation of consoles.

However, the security system of Cell is probably the best I've seen. Hopefully it makes it into other CPU designs as I'd hate to see it die with Cell.

Regards,
SB

 

[aaronspink]

However, the security system of Cell is probably the best I've seen. Hopefully it makes it into other CPU designs as I'd hate to see it die with Cell.

Why? Its just your standard root of trust model with hashes, signatures, and encryption. Odd are you've been using a cpu for years with the same capabilities.

 

[flynn]

Umm...

The cell basically is dead if rumors of all 3 partners have stopped active R&D on Cell are true. Without continued active funding of R&D it's basically a dead end.

That depends on whether the next gen will be simply evolutionary or not. To get 1080/60p all you need is more memory and a beefier GPU, so a souped up Cell could work very well. You need very little, if any, investment to get that working.

And as to MLAA, any CPU can do it, there isn't anything particularly special about Cell that makes it impossible to do on other architechtures. You may have a valid point that at that clock speed Cell may currently be the fastest at MLAA, but considering other architechtures scale much higher in clock speed and up to 6 (soon to be 8) "real" cores, I'm not sure that's going to be a very good argument for the next generation of consoles.

What's the TDP and cost per unit of those 6-8 core CPUs? Can Sony own the IP and work on the design at will?

However, the security system of Cell is probably the best I've seen. Hopefully it makes it into other CPU designs as I'd hate to see it die with Cell.

The Xenon has some pretty robust security built-in. I'm sure we'll see an evolution (+ lessons learned from 360) in the next xbox. Nintendo probably also learned some tricks from the PS3 security architecture. If they take their time and don't rush products out of the door without proper QA and testing they can build pretty secure systems as well.

 

[Silent_Buddha]

That depends on whether the next gen will be simply evolutionary or not. To get 1080/60p all you need is more memory and a beefier GPU, so a souped up Cell could work very well. You need very little, if any, investment to get that working.

Quite possible, but a stagnant system is still going to end up falling behind those that are in active developement.

What's the TDP and cost per unit of those 6-8 core CPUs? Can Sony own the IP and work on the design at will?

Sure right now it may not be feasible to take an off the shelf 6 core CPU (~200 USD) and throw it in a console, but by the time they are due to come out? It didn't take long for AMD for example to drop the price of 4 core desktop CPUs from 200+ usd to ~100 USD. Granted there were other market forces at play (Intel) that forced the issue, but that isn't to say something simular couldn't be contracted in a similar way to how MS contracted IBM to make the CPU for X360 using existing building blocks from existing CPUs and crafting a somewhat custom CPU for their use.

The Xenon has some pretty robust security built-in. I'm sure we'll see an evolution (+ lessons learned from 360) in the next xbox. Nintendo probably also learned some tricks from the PS3 security architecture. If they take their time and don't rush products out of the door without proper QA and testing they can build pretty secure systems as well.

Yes being able to blow "fuses" in the CPU when the JTAG exploit was discovered was able to prevent mass hacking on the order of the original Xbox1. I wasn't implying that Xenon was particularly weak, just that Cell appears particularly strong. I could, of course, be completely wrong about this as aaronspink implies, as I'm only observing what's been done with regards to consoles.

Regards,
SB

 

[aaronspink]

Yes being able to blow "fuses" in the CPU when the JTAG exploit was discovered was able to prevent mass hacking on the order of the original Xbox1. I wasn't implying that Xenon was particularly weak, just that Cell appears particularly strong. I could, of course, be completely wrong about this as aaronspink implies, as I'm only observing what's been done with regards to consoles.

Well, the JTAG exploit was pretty much a sign that they goofed pretty hard. CPUs have had proprietary lockouts on any advanced JTAG/debug functionality for quite some time.

The root of trust model is effectively the only model that works. It relies on cryptographic signatures and hardened cryptographic resources to measure the authenticity of any software that is loaded and provide a method for software to determine if any tampering has occurred on the system before its was loaded. It involves things like sony et al signing various pieces of software that are then checked by other pieces of software that are verified via hardware only/hardware restricted mechanisms. They aren't full proof however and given adequate access to hardware and advanced debug hardware it is possible to crack them, just very hard and very costly.

 

[patsu]

And as to MLAA, any CPU can do it, there isn't anything particularly special about Cell that makes it impossible to do on other architechtures. You may have a valid point that at that clock speed Cell may currently be the fastest at MLAA, but considering other architechtures scale much higher in clock speed and up to 6 (soon to be 8) "real" cores, I'm not sure that's going to be a very good argument for the next generation of consoles.

Cell is a 2005 (or earlier) design. If they were to do an "equivalent" CPU today, they will have more options. However, to do what Cell did/does, they will still need to come up with L1 level memory access/window and fast shared memory access with the GPU.

As for number of cores, they can put in even more SPUs (aka real cores) these days if they want, but it may be more fruitful to improve the PPU also.