Backwards Compatibility: poll

[TheChefO]

As ever, you miss completely the point.

Let's see if I did ...

Sure, XB720 could use a tripled up Xenos which would scale pretty efficiently, but that also assumes that such a CPU would be the best performance and value. What if MS can get an x86 or ARM or Cell CPU that would perform better and cost less?

...Why even go With IBM at all? Was ARM (insert favorite alternate architecture here) not accepting new customers at the time?

...Let's see ... how about Itanic? *ahem* I mean Itanium?

That opens the door for ARM or ... ARM? I'm honestly not seeing a ton of other options out there but there seems to be a lot of desire to wipe the slate clean and go with ... something ... but it's never been said what the alternate architecture would/could be that would be worth scrapping Power for.

...MIPS?
Alpha?
AMD is desperate these days, but it's due to the fact that they don't have much to offer that would be suitable from a performance stand point and Intel is king of performance and for that reason their products are in high demand so they don't need to sell their IP to Sony or MS for their new consoles ...

So that leaves ARM ... and Mips or Alpha ... or Sun's Niagra!...

Seems I addressed your point a few times. And yet, I've never seen a rebuttal explaining your position of WHY X architecture would be an advantage over PowerPC. Instead, it seems you leave it open for some possible future magic bullet without anything to even suggest such a thing even exists.

How about even a single benchmark which showcases one alternative ISA that would be worthwhile to break Backwards Compatibility with the Power ISA that both Sony and MS use?

Then the choice of going with Xenos^3 costs the user relative to the option with an x86 processor.

As I have already brought up prior to this thread, and in this thread, x86 isn't going to happen and if it did, it wouldn't be an advantage unless intel was providing the chip which isn't going to happen for various reasons.

Not to mention, you've already highlighted the inefficient nature of legacy transistors inherent in x86 ... yet that is now "costing the user performance" ...

What if Sony go with Cell^2, loads more SPUs, but devs are sick of trying to wring performance from it and the middleware like UE doesn't work well on it?

Well then I suppose Sony might have another alternative such as this:

...a quad core PPE with 8SPEs there for those that know how to utilize them. For those that don't know or care to utilize the SPE's, they could invest in some middleware which puts them to use handling things like Audio, Physics, OS, and post processing effects.

All the while, providing BC, and for a relatively small die cost of around 40mm2 @ 28nm.

Or they could upgrade the SPE's to address their shortcomings and have them more useful for more tasks with improved middleware which takes advantage of them.

*ahem*

Then PS4 looks like poop next to XB3. Whereas if if Sony pick a more developer friendly CPU, the console gives the users a better experience.

I'm quite sure PS4 will be measured significantly more by their GPU choices rather than whatever they decide for their CPU which ties directly back into my whole point for scaled architecture in the first place.

And that ignores the issues of GPU emulation, even within the same family of processors, which you've never recognised because you seem to think that the hardware is accessed through a software layer and so graphics code is portable across GPUs. Emulating GPU adds cost somewhere along the line.

I'm pretty sure Sony has a few smart cookies working for them that can figure out a way to match/translate RSX machine code and OPs to a GCN (or Tesla). Whether this "costs" Sony anything is likely to be tied to whether they decide to go ditch nvidia and are therefore then forced to "license" the tech from nvidia for emulating the gpu much as MS had to pay nvidia for the same purpose.

Of course we've had this debate before, and of course you never appreciated it then, so there's no point repeating this.

Indeed it is difficult getting you to read through all my rubbish to actually address the entire point/post!

 

[TheChefO]

Aw come on now, you do a lot of posts and type a lot of words in each of them. Don't get shy now.

How is the PPE in Cell different to Power4? To keep this somewhat on topic, in what ways would PPE not be compatible with Power4 (or the larger Power ISA) which would break code compatibility?

 

[dobwal]

Did it?

Or was it poor engineering on MS' Part which cause RRoD ($1B)

Bluray along with mandatory HDD, multiple memory traces for XDR and GDDR3, and further increased board complexity with having a ps2 chipset also sharing the motherboard I think caused Sony more Financial trouble than having a 470mm2 budget for Cell and RSX.

Could they be less aggressive this go round? Perhaps, but the decision to go with such a budget wouldn't break the bank. It's what they decide to do outside this silicone that may infringe on the ultimate budget size of the die.

Even without accounting for the $1 B, the 360 cost MS a ton of money in its early life even without a mandatory HDD, split GDDR/XDR memory pool or BluRay. Stripping out the PS3's BC didn't stop Sony from bleeding money. Plus, PS3's cooling wasn't cheap either and you can easily assume that RROD was partially due to the cooling design of the 360 was inadequate for 90nm parts.

Nevermind, the fact that both of these consoles still have ASPs in the mid/high $200s, which I doubt either manufacturer planned for 7 or 8 years ago. The only thing that saved both of these manufacturers from even greater losses is the fact that gamers were willing to consume these consoles at far higher prices for a far longer period of time than anyone could have predicted and that their direct competitor also had to deal with less than ideal circumstances surrounding the manufacturing cost of their console. Can MS or Sony depend on the same circumstances next gen?

Neither the 360 or the PS3 are great examples of how to properly and cost effectively design and manufacture a console.

And neither of these consoles were greatly helped by BC so actually devoting silicon to BC in their next gen successors is a waste of resources and capital.

 

[TheChefO]

Neither the 360 or the PS3 are great examples of how to properly and cost effectively design and manufacture a console.

I think you might have an audience with Sony on that one, but the design decisions along with a few other factors lead MS to profitability this gen ... along with beating their competition in sales, instilling a foothold in the livingroom more than twice the size of the previous console (and growing) and all of this, despite losing money upfront.

I'm not saying there aren't lessons to be learned and applied, but gimping the die budget I don't think is one of them.

And neither of these consoles were greatly helped by BC so actually devoting silicon to BC in their next gen successors is a waste of resources and capital.

Actually, the xb360 WAS helped by BC. Having BC enabled MS to carry Live over to xb360 and bring the old users along to the new without disruption to the most popular online game of the time: Halo.

Times change and the online component of these consoles is more important than ever. The ability to play online with your friends on the most popular games is a crucial aspect of the gaming scene.

To sever this connection would be detrimental to the original intent of the online networks Sony and MS have been working so hard to create in the firstplace.

Not to mention all of the DLC.


Now, as for the best allocation of silicon, I agree that having dedicated silicon to legacy hardware seems rather wasteful and it would be better to find another way to achieve BC.

Hence my argument for Scaled/Derivative hardware.

 

[function]

How is the PPE in Cell different to Power4?

Looks like it's my job then!

To keep this somewhat on topic, in what ways would PPE not be compatible with Power4 (or the larger Power ISA) which would break code compatibility?

That's a completely different question, and whatever the answer it can't support your matter-of-fact claim that the PPE is a "slightly modified" Power 4.

Power 4:
Out-of-Order
Single thread per core
64kb L1 instruction cache and 32kb L1 data cache
8 execution units

PPE:
In-Order
SMT (2 threads per core)
32 + 32 L1 cache
6 execution units (probably, says Ars)

Right or wrong, Anandtech had this to say in 2005: "The PPE is a new core unlike any other PowerPC core made by IBM*" . Which just made me laugh a bit, under the circumstances.

So ... how do you figure that the PPE is a slightly modified Power 4?

*http://www.anandtech.com/show/1647/3

 

[novcze]

If CPU in PS4 will be some quad core PPE with 6 SPU's for BC, can this solve PS3 emulation?
I would like to see every PS3 game to run on PS4 in 1080p with stable framerate and more AA. Is that possible?

 

[TheChefO]

If CPU in PS4 will be some quad core PPE with 6 SPU's for BC, can this solve PS3 emulation?

Indeed it would.

That would be scaling the existing architecture to fulfill future software needs while staying committed to the codebases which were already developed.

Option 3

 

[novcze]

Indeed it would.

That would be scaling the existing architecture to fulfill future software needs while staying committed to the codebases which were already developed.

Option 3

Yep, I already voted for this option. RSX part of emulation isn't problem I suppose, so increasing resolution and AA is possible, right? Dunno in case when post-processing is done on SPU's, higher resolution can be problem there.

 

[NavNucST3]

I have never cared about BC until this generation. If the literal hundreds of digital downloads I have purchased can't move over then I'm not either. I don't make a habit of keeping previous gen systems if I have moved on to the next generation. I have about 25+ games from Games on Demand and 2 or 300 XBLA titles so I will either be sticking with the current platform for quite some time or until such time that enough software/services have interested me enough to want to jump to the new generation of hardware. Now if the new generation has a new box and then some portable hdd/apple tv sized dongle or whatever that gives me full BC then I'll make the move.

 

[TheChefO]

Looks like it's my job then!



That's a completely different question, and whatever the answer it can't support your matter-of-fact claim that the PPE is a "slightly modified" Power 4.

Power 4:
Out-of-Order
Single thread per core
64kb L1 instruction cache and 32kb L1 data cache
8 execution units

PPE:
In-Order
SMT (2 threads per core)
32 + 32 L1 cache
6 execution units (probably, says Ars)

Right or wrong, Anandtech had this to say in 2005: "The PPE is a new core unlike any other PowerPC core made by IBM*" . Which just made me laugh a bit, under the circumstances.

So ... how do you figure that the PPE is a slightly modified Power 4?

*http://www.anandtech.com/show/1647/3

Indeed, you are correct!

I remembered reading an article at about the time the new consoles were being introduced which suggested the Cell/Xenon were derivative of Power4, but that doesn't seem to be the case. Nor can I find any other implementation which mirrors Xenos/PPE prior to 2005. The closest I found is Power6 which came after 2005, but did implement in-order execution instead of OOOE as previous processors in the power line along with 2-way SMT. (though it was said to be faster than Power5)

Thank you for correcting me on this!

Having said that, the code-base would still be compatible as long as it's PowerISA, 64bit, =>cache, includes at least the VMX128 registers (probably add a few as well) and runs/executes at least as fast as Xenos/PPE.

 

[ERP]

FWIW BC is never actually "free".
Assume fixed expenditure anything you spend on it doesn't get spent elsewhere, at some level that doesn't matter if it's hardware/software or just test.
There are the obvious technical decisions, like being forced to scale an architecture vs picking best available, but it's much more than that.
If a manufacturer wants to claim backwards compatibility they have to test all of the existing titles, and continue to do so with every revision of the hardware. This is especially expensive before final hardware is readily available.

The questions are how much performance penalty are people willing to pay for BC or how much money are they willing to pay for it, because the consumer will pick up any additional cost.

 

[TheChefO]

Yep, I already voted for this option. RSX part of emulation isn't problem I suppose, so increasing resolution and AA is possible, right? Dunno in case when post-processing is done on SPU's, higher resolution can be problem there.

RSX emulation shouldn't be a problem on the monster GCN/Tesla GPUs I'm expecting we will see nextgen. Increased Res would throw a monkey wrench into things if it isn't done carefully. Same for AA, unless they go with an MLAA hardware solution which may happen.

I wouldn't expect increased Resolution, but perhaps better AA.

 

[TheChefO]

FWIW BC is never actually "free".
Assume fixed expenditure anything you spend on it doesn't get spent elsewhere, at some level that doesn't matter if it's hardware/software or just test.
There are the obvious technical decisions, like being forced to scale an architecture vs picking best available, but it's much more than that.
If a manufacturer wants to claim backwards compatibility they have to test all of the existing titles, and continue to do so with every revision of the hardware. This is especially expensive before final hardware is readily available.

The questions are how much performance penalty are people willing to pay for BC or how much money are they willing to pay for it, because the consumer will pick up any additional cost.

First of all, thanks ERP for joining in the discussion, I appreciate your insight.

_________________________

Two different things here.

1) chose hardware

2) support a marketing checkbox via testing


1) If they chose to scale the hardware, yes that means other hardware is not an option which in theory could be detrimental to performance if the alternate hardware architecture is superior.

But as we see time and time again, nothing is in a vacuum. Gains here equals losses there (assuming equal die/process budget). SPE's may be limited in local store which therefore limits what code they can run without significant time being spent to optimize code for the SPE, BUT ... when optimizations are made, the SPE's are very fast, and the Cell design overall solves many problems of future hardware scaling bottlenecks.

Perhaps a tweaking of the design of the SPE/Cell to accommodate the desires of developers while still providing high performance/mm and backwards compatibility.

Would this be suitable?

Are there other architectures out there which would be head and shoulders above the Power ISA family per mm?

2) MS/Sony can either pay for testing and have the checkbox or don't pay and don't have the checkbox. Perhaps a large enough public beta after they DO test for the most popular online games. This testing would have to take place no matter what method of BC they choose. Scale hardware, partial or full redundant HW .. all would require testing. However, that's not to say that some software may not work. Heck even the PS2 had a few ps2 games that didn't work after certain revisions!

Either way Sony/MS decide to go on the subject, I do fully expect that the most popular online games at the time WILL have BC with the new consoles when they launch. Not following suit with this would be tantamount to starting over from scratch as it would not invite nor lure in peoples social network to upgrade to the new console when everyone is still playing the favs on their old boxes.

 

[AlphaWolf]

if you launch with a version of COD you dont need to have last years version running on the box

 

[TheChefO]

if you launch with a version of COD you dont need to have last years version running on the box

Your friends won't have the new box version or the new box.

 

[Shifty Geezer]

Let's see if I did ...

You've come up with reasoning why you consider there no other options. In forming that reasoning you have zero access to the financial implications (how much does one IP cost versus another, or using an off-the-shelf component versus a customised one), wider system strategy (what if you want your console to be hardware compatible with your tablets and TVs, and are willing to take a performance hit for wider system synergy), software strategy (do you want to run legacy code or have you access to a new load of code on a new architecture), and whatever else a console designer will need to consider.

Seems I addressed your point a few times. And yet, I've never seen a rebuttal explaining your position of WHY X architecture would be an advantage over PowerPC. Instead, it seems you leave it open for some possible future magic bullet without anything to even suggest such a thing even exists.

How about even a single benchmark which showcases one alternative ISA that would be worthwhile to break Backwards Compatibility with the Power ISA that both Sony and MS use?

That's not a discussion I'd enter into because I don't have that much knowledge. I can go by the performance specs that we all have access, but that's far from the whole picture. I have no idea whatsoever what developing an advanced Cell 2 that's BC would cost, versus using the existing one, versus creating a better one that's not BC, versus giving devs a more forgiving architecture completely, and I have no idea what sorts of contracts could be negotiated to get whatever processors at whatever price. And that's just the CPU! I have no idea what would be needed to get games that hit RSX at the bare-bones metal level to work on some new unified shader GPU, let alone what would be required to switch GPU vendor entirely. Nor do I know how much using eDRAM on a new XB will cost versus not using eDRAM and going with a different design; nor how well XB360's graphics can be run on an off-the-shelf AMD GPU; nor what adding extended 256 register VMX to a new PPC would cost versus going with a different PPC architecture that's more focussed on multithreading; nor what the performance benefits and costs of the two architectures would provide and the yieldability of the performance such that devs can harness the CPU to best effect, ergo having the option to choose the less powerful processor for its better results in most cases. None of us have that info.

My point isn't to challenge your choice of hardware, but your belief that BC can be free. It isn't*. Every console dev who replies to you on this matter tells you it isn't free. As such, your third option is unrealistic, which is what I was contributing. The cost could come in performance, in system price, in system flexibility, in work the devs have to do to enable compatibility in old games that they could be spending making new games, in dividing the larger entertainment division's efforts between supporting two hardware platforms or just one more flexible one (eg. MS supporting XB hardware and ARM hardware and x86, where they could just support ARM and x86).

(*unless you're very lucky and it just so happens that having considered all the many permutations of hardware available, a scaled up 360 or PS3 happens to be the best fit for the console company's aims)

 

[TheChefO]

You've come up with reasoning why you consider there no other options. In forming that reasoning you have zero access to the financial implications (how much does one IP cost versus another, or using an off-the-shelf component versus a customised one), wider system strategy (what if you want your console to be hardware compatible with your tablets and TVs, and are willing to take a performance hit for wider system synergy), software strategy (do you want to run legacy code or have you access to a new load of code on a new architecture), and whatever else a console designer will need to consider.

IP costs are part of the game, but I'm not going to assume that this small portion of the budget will ultimately dictate the design choices.

Platform-wide-compatibility I addressed in my suggestion of adding a small ARM core to the console.

Future proofing a console is never possible. Switching to the hot commodity of the day with the thought that in the future it will become THE defacto standard leaving all else behind can lead to disastrous results. Itanium developers can speak to that.

My whole point WRT scaling the existing Power Architecture is that the grass isn't always greener just because it's on the other side of the fence.

Funny thing... If Sony does decide to scale the Cell and stick with it, The ps2 SOC is small enough at 28nm to throw that back in the box and get back to their original intent of Playstation as a platform, giving ps4 full compatibility with ps1, 2, 3 and ps4.

That's not a discussion I'd enter into ....

Ok ...

My point isn't to challenge your choice of hardware, but your belief that BC can be free. It isn't*.

While this may be true, until you have alternate hardware to compare it to, we have no idea exactly what it does or doesn't cost because it is being compared to fairy dust!

The idea of BC testing and validation costs is valid and a real cost, but it exists no matter what BC HW is chosen as long as it is a marketing bulletpoint which is NOT necessary in order to actually fullfill BC. Much like what xb360 did. This can be done after the fact, and the fixes for broken scenarios can also come afterwards. In fact, they could open a public "beta" program for consumers to test and report on BC games.

Free? No. But scaling IS a free HARDWARE implementation.

(*unless you're very lucky and it just so happens that having considered all the many permutations of hardware available, a scaled up 360 or PS3 happens to be the best fit for the console company's aims)

Indeed.

The only difficult decision I could see would be how Sony intends to handle Cell, but anyway they decide to incorporate Cell tech in ps4, the silicon will not go to waste if they are smart about it.

MS should have no problem sticking with PowerISA and scaling beyond Xenos while maintaining BC.

 

[Shifty Geezer]

My point isn't to challenge your choice of hardware, but your belief that BC can be free. It isn't*.

While this may be true, until you have alternate hardware to compare it to, we have no idea exactly what it does or doesn't cost

So agree BC isn't free? Great. Change your third option to remove the zero cost aspect.

Here's another take:

MS should scale XB360, and Sony should scale PS3, because both are great, easily scalable platforms, right? Now PS3 as a system had a great CPU but a poop GPU. If PS3 had Xenos, it'd have roxxorz your soxxorz and easily been the best console of this generation. Ergo with a new, better GPU, and the better CPU, PS4 will beat XB3 hands down. That means MS have to abandon Xenon 2 and use Cell or they lose the performance race. Unless Xenon could be competitive with Cell, at which point Cell is providing silly hoops for devs to jump through for no advantages, so Sony should abandon Cell and go with a more user-friendly CPU architecture.

How do you argue that both should scale their existing CPU architectures as both are ideal solutions?

 

[TheChefO]

So agree BC isn't free? Great. Change your third option to remove the zero cost aspect.

Scaling is a free hardware BC solution. Until I see something which isn't PPC that can be measured favorably, I'll continue to believe it is "free"/zero-performance-hit or near enough to it not to consider throwing the baby out with the bath water.

...Ergo with a new, better GPU, and the better CPU, PS4 will beat XB3 hands down...

Indeed.

Yes, in this theoretical scenario, with ps4 and xb360 both scaling their CPU's and having identical GPU's, the ps4 should have a significant theoretical advantage in CPU performance.

Thing is, I don't think the CPU is all that important for the performance bottom line.

This would still result in a net benefit and ps4 multiplat games having an edge on xb720, but in order for such an advantage's potential to be fulfilled, they can't afford to be late to the dance ...

As for how both can be the "best" or "most efficient", obviously they can't both be. But they do both have their pros and cons. Also, nothing is stopping MS (or Sony) from modifying their CPU's by adding new functionality/registers (~VMX256) or adding execution units, instituting additional SMT threads per core, going OOOE, or adding/modifying cache or local store.

Good argument though. Nice point.

 

[Shifty Geezer]

Scaling is a free hardware BC solution. Until I see something which isn't PPC that can be measured favorably, I'll continue to believe it is "free"/zero-performance-hit or near enough to it not to consider throwing the baby out with the bath water.

Which won't ever change. We have a number of variables for which we have some data and some (lots) of unknowns. Your approach is to ignore the unknowns and make choices on the known data. My approach is to consider the problem unsolvable and will wait until there's enough information to go on, which in this case there will never be. All I can talk about are generalisms. If you won't accept them without these impossible datapoints then that's that. Especially when you've decided BC is just a matter of picking a CPU with a compatible ISA. Why doesn't the GPU get a mention?

Yes, in this theoretical scenario, with ps4 and xb360 both scaling their CPU's and having identical GPU's, the ps4 should have a significant theoretical advantage in CPU performance.

Thing is, I don't think the CPU is all that important for the performance bottom line.

Now you've shifted the goalposts a little. If you agree that Cell is better than Xenon, and PS4 with Cell2+GPU will be better than XB3 with Xenon2+same GPU, then there is a cost for scaling XB360; it'll lose performance.