Predict: The Next Generation Console Tech

[bgassassin]

Are you still going to be saying that when inevitably Wii U turns up 0-30% more powerful than PS360?

Are you going to concede when that isn't the case?

 

[Sonic]

I honestly think if MS is going PowerPC and AMD/ATI in their next generation system they might as well start forward compatibility they year they release the system. I see good and bad business sense in letting devs make games for both 360 and 720 and of course still have exclusives for the new system so people adopt to it. But at least this way a common code base could be used, it's already PowerPC so how much code would need to be changed in order to get games up and running on the new hardware. Give them full effects, 1080P resolution and 4XAA and you have a good upgrade from the current system. On the other end, it could mean that more people wait and don't initially get the new system. In that case a steady supply of new content would be needed for the system to sell.

Nintendo's system better be more than 30% more powerful than the 360. The CPU will most likely be beefier, but am so hoping the bulk of the power will come from the GPU. Nintendo shouldn't skimp any more!

 

[Ninjaprime]

That's easy. Power7. Taking the process (45nm vs 32nm) into account, I don't think many would argue x86 superiority very vigorously.

Actually, SNB beats Power7 pretty soundly in almost any metric.

 

[Shifty Geezer]

I would have to consider that a stretch. From a consumers point of view, they are used to certain time periods for their consoles.

Up to now, but the introduction of mobiles with very rapid upgrade cycles showing different possbilities.

3 years seems an awfully short time for a developer to make back their investment as well.

As I say, future systems would be forwards compatible, so a game written for XB3 runs on XB4 also, and XB mobiles 1 and 2 etc. But this isn't tech talk so I'll stop there.

 

[Entropy]

So you believe there's a power7 CPU out there that would be a faster gaming CPU than say a 2700k? Baring in mind the 2700K is 200 million transistors smaller and includes an intergrated GPU.

You're shifting the goalposts a bit - suddenly I have to come up with a PPC processor that is a better gaming CPU than a 2700K? Let me remind you:

And similarly I'd like to see the PPC that's the same size as a Sandybridge offering anywhere near the performance.

And I gave that example. Mind you, these CPUs aren't targeting the same markets, but even with a lithographic process liability, Power7 is equal to a Corei7 2600 in the SPEC CINT2006 suite, and murders the 2600 in the CFP2006 (floating point) suite of benchmarks. I think that qualifies as "anywhere near".

At the time of launch (and for quite a while before), the fastest and most efficient x86 architecture was the dual core Athlon X2. It wouldn't have been as cheap for MS/Sony as the custom PPC solutions they went with but that's a business decision, not a technical limitation. That's the whole point, PPC is and was the right decision because it's cheaper in the long run, not because it's overall more powerful or more performant per watt or transistor.

I think the above is a beatiful summary.

I will just make the addendum that a customer being able to pick and choose cores and configurations can make going with PPC technically superior to X86 as well, for a specific target application area. For instance, if you can stay mostly in L1 (or in the case of the BBE, local storage), both the Xenon and the BBE will vastly outperform your Athlon X2 example in floating point throughput. I can't say to what extent that is actually the case much in games. For general purpose computing (whatever that is), without particular power constraints, x86 does fine.
I'll agree that it is business reasons that kills x86 in consoles, but it is hard to separate business from technology, because consoles are designed within very tight constraints so whatever choice you make in for instance CPU will reverberate through the entire design, price-over-lifetime and so on of the console.

 

[Entropy]

Actually, SNB beats Power7 pretty soundly in almost any metric.

Well, people using Power7 systems aren't exactly running 3DMark on them, so I haven't actually been able to find many comparison points at all. The ones I did find (at spec.org), showed the Power7 to be roughly equivalent in integer processing, and much stronger in FP, despite being a process node behind.

You're welcome to back up your claims.

However, I'm not sure how relevant this discussion is. Nobody is going Intel for their next console, and for good reasons that have little to do with performance. Nor do I believe that anyone will go with straight Power7 cores, they carry baggage no console is interested in. Thus I can't really see the relevance in a "Predict The Next Generation Console Tech" thread.

 

[hoho]

And I gave that example. Mind you, these CPUs aren't targeting the same markets, but even with a lithographic process liability, Power7 is equal to a Corei7 2600 in the SPEC CINT2006 suite, and murders the 2600 in the CFP2006 (floating point) suite of benchmarks. I think that qualifies as "anywhere near".

But from how many transistors is that Power7 is made up from? Feel free to add the massive external L3 as well

 

[Entropy]

But from how many transistors is that Power7 is made up from? Feel free to add the massive external L3 as well

The 32MB L3 is internal EDRAM in Power7.
Power7 is 1.2billion transistors.
Sandy Bridge (4-core) is 1.16 billion transistors.

 

[hoho]

Power7 is 1.2billion transistors.
Sandy Bridge (4-core) is 1.16 billion transistors.

This table here shows 45nm i7's to have 731M transistors and 32nm 6-core ones at 1.17B. Considering how little performance difference there is between those and SB wouldn't it make sense to use Gulftown when doing perf-per-transistor comparisons?

Also, wasn't there a version of Power7 that had 2B transistors? What version was used in that Spec benchmark? Also kind of related question would be that wasn't SpecFP actually more of a bandwith-bottlenecked test and that 100GB/s connection Power7 has could possibly have something to do with it's supermacy over x86 at around half that throughput?

 

[Gubbi]

But from how many transistors is that Power7 is made up from? Feel free to add the massive external L3 as well

The L3 is on-die, 4MB eDRAM cache per core. It has a total of 1.2 billion transistors.

Power 7 cores are roughly comparable to Sandy Bridge cores, they have similar OOO capabilities, SMT and execution resources. Sandy Bridge seems to be slightly faster on single thread performance, while Power 7 is faster on multi-threaded problems (4-way SMT vs 2-way)

Building a console around a P7 derived CPU core would result in a very capable machine.

Cheers

 

[Gubbi]

This table here shows 45nm i7's to have 731M transistors and 32nm 6-core ones at 1.17B.

Also, wasn't there a version of Power7 that had 2B transistors? What version was used in that Spec benchmark? Also kind of related question would be that wasn't SpecFP actually more of a bandwith-bottlenecked test and that 100GB/s connection Power7 has could possibly have something to do with it's supermacy over x86 at around half that throughput?

Your transistor comparisons are pointless. You're mostly comparing cache transistors, and SRAM vs. eDRAM at that.

Each Power 7 core (with L2) is around 27 mm² in 45 nm. A Sandy Bridge core (with L2) is around 17 mm² in 32 nm. If you assume perfect scaling a P7 core would be 13 mm². Scaling isn't perfect though, so let's just assume they are roughly comparable. So die area is comparable, power consumption is comparable and performance is comparable.

Cheers

 

[french toast]

Wow this thread moves on fast, we are getting off topic a bit so i will make this short.

Entropy seems to have a better knowledge (as many others) and has answered the whole intel v PPC, i think top level chips they are equal..however there must be a reason why microsoft has not gone back to Intel? where are these 'superior' Intel processors showing up in any console?...It seems that there are other factors involved.

PPC obviously has a major advantage somewhere..probably price..as when billion$$ are on the line, blind loyalty doesn't exist.

Ill stick to my assertions that vliw5 is going to be the uarch of next gen consoles..in every form.
Shiftys right, you need decent, well writen software from the get go, apart from the early release, thats what kept the 360 ahead of the ps3..the best software, its no good waiting 2-3 years to start to take advantage, peeps will switch to another console.

 

[3dilettante]

The primary push away from x86 is mostly related to business. Ownership of the IP, and thus the ability to do with the design what they want as they shrink the chip and make cost-cutting measures, was more restricted with x86.

IBM has also historically been very helpful in part because it has its own fab and a microelectronics division that it needs to justify keeping around. It doesn't hurt to have additional products to amortize the costs of these big investments, and at various times it has utilized fab capacity that could have gone idle.

 

[Dominik D]

[T]here must be a reason why microsoft has not gone back to Intel? where are these 'superior' Intel processors showing up in any console?...It seems that there are other factors involved.

Intel doesn't license tech. They can manufacture their chips for you, but you're not going to get their tech so if they don't feel like manufacturing given chip anymore (and at some point, sooner than layer, they will) you're out of luck and out of chips. This is perfect for PC but kinda sucks for consoles.

 

[ERP]

Intel doesn't license tech. They can manufacture their chips for you, but you're not going to get their tech so if they don't feel like manufacturing given chip anymore (and at some point, sooner than layer, they will) you're out of luck and out of chips. This is perfect for PC but kinda sucks for consoles.

There are 2 real issues with not owning the IP or a license to manufacture, the first is the cost of the chip 5-7 years from now, you have little control over the suppliers margins.
The second is combining IP from multiple sources onto a single chip, you simply can't do this without the IP.

This is the primary reason consoles were primarily Mips based, before they were power based, and why arm has a shot in the future if they can get something competitive performance wise into the market.

Of course depending on how AMD are to deal with x86 isn't necessarily ruled out.

 

[Ninjaprime]

Well, people using Power7 systems aren't exactly running 3DMark on them, so I haven't actually been able to find many comparison points at all. The ones I did find (at spec.org), showed the Power7 to be roughly equivalent in integer processing, and much stronger in FP, despite being a process node behind.

You're welcome to back up your claims.

However, I'm not sure how relevant this discussion is. Nobody is going Intel for their next console, and for good reasons that have little to do with performance. Nor do I believe that anyone will go with straight Power7 cores, they carry baggage no console is interested in. Thus I can't really see the relevance in a "Predict The Next Generation Console Tech" thread.

Per core, per clock, SNB is ~20% ahead on SPEC_int and roughly even on SPEC_fp, and as stated above, they are roughly comparable die size wise and transistor size.

http://www.realworldtech.com/page.cfm?ArticleID=RWT040511235825&p=2

However, have you ever considered power consumption? Because the 3.55-3.86Ghz Power7 8 core chip has a power dissapation of about ~200 watts, and the 4-4.14Ghz version nearly brushes 250 watts. A i2700k, as was mentioned above, is about 75 watts without the IGP. A pair would be ~150 watts. A third better power consumption, 20% faster on int, even on fp. Balls pretty much in Intels court there.

Still, I don't know if Power7 or SNB would make good console CPUs. I know Wii U is rumored to be Power7 based, but I suspect its more of loosely based, more of a custom CPU with Power7 roots, because Power7, in raw form, is horribly wasteful as a console CPU, as you say.

Edit: SNB, or IVY-B, might make a good console CPU, if someone went with a radical new design, ala Cell in PS3, and coupled it with KC/LRB3-4 as an accelerator. Sony is really the only undefined one now though, and I'm pretty sure they learned their lesson from Cell, but it would be interesting to see what came of an all x86 general purpose machine with high throughput, as Tim Sweeney has talked about.

 

[east of eastside]

This is the primary reason consoles were primarily Mips based, before they were power based, and why arm has a shot in the future if they can get something competitive performance wise into the market.

Like this ARM Atlus 64-bit 20nm chip scheduled for 2014:

http://forum.beyond3d.com/showpost.php?p=1616697&postcount=9519

(reposting this because I have yet to see ARM's recent 64 bit announcement enter into any discussion)

Of course depending on how AMD are to deal with x86 isn't necessarily ruled out.

There is a possible scenario of AMD acquiring the ARM architectural license, it has long been speculated it will eventually do so, and from their customize, along with MS engineers, an ARM 64 bit multicore processor that can be incorporated in their GCN architecture. (recall the ARM and MS attended conference last July on heterogeneous multicore computing perhaps laying the foundation a development like this)

The move has some business elegance as you eliminate the total partners on the IP and fab side by one by totally cutting out IBM.

On the IP side this would be essentially an all AMD/MS chip with the entire fabrication being handled by Global Foundries (who has been increasing their involvement with ARM).

 

[Rangers]

Edit: SNB, or IVY-B, might make a good console CPU, if someone went with a radical new design, ala Cell in PS3, and coupled it with KC/LRB3-4 as an accelerator. Sony is really the only undefined one now though, and I'm pretty sure they learned their lesson from Cell, but it would be interesting to see what came of an all x86 general purpose machine with high throughput, as Tim Sweeney has talked about.

As I understand it it's licensing, patents, IP's, that sort of thing that make Intel, AMD, and X86 a non starter in console CPU's.

No clue if true but I heard that AMD's X86 licensing agreements preclude their CPU's from being in a console, for example. With Intel it would be more an issue that they dont license out the fabbing rights to their IP, apparently.

It's boring but odds are 90% you're going to see a whole lot of IBM CPU's in next gen, again. There are very good reasons, it's the path of least resistance.

Also, bgassasin is dropping hints over on GAF that IBM let it be known they designed a new custom power PC chip for Wii U, one that presumably might be whored out to form the basis of all 3 next gen consoles. So that's interesting.

Of course BG's bias is that all the consoles will be very similar and none will have a drastic power edge, and this fits perfectly in that, so take it with a grain of salt from that angle.

 

[Proelite]

As I understand it it's licensing, patents, IP's, that sort of thing that make Intel, AMD, and X86 a non starter in console CPU's.

No clue if true but I heard that AMD's X86 licensing agreements preclude their CPU's from being in a console, for example. With Intel it would be more an issue that they dont license out the fabbing rights to their IP, apparently.

It's boring but odds are 90% you're going to see a whole lot of IBM CPU's in next gen, again. There are very good reasons, it's the path of least resistance.

Also, bgassasin is dropping hints over on GAF that IBM let it be known they designed a new custom power PC chip for Wii U, one that presumably might be whored out to form the basis of all 3 next gen consoles. So that's interesting.

Of course BG's bias is that all the consoles will be very similar and none will have a drastic power edge, and this fits perfectly in that, so take it with a grain of salt from that angle.

You mean bgassasin is legit? I've always thought he was a Nintendo fan pushing the parity agenda.

 

[babybumb]

You mean bgassasin is legit? I've always thought he was a Nintendo fan pushing the parity agenda.

Yes he is legit in as trying to act as an insider