Predict: The Next Generation Console Tech

[Rangers]

No, you must understand that games are not going to be made for a console with graphics that have the same workload as a program design to do nothing more than push ALUs to breaking point!
Not unless you want a game to look like furmark!
And even if you did want a game like that you are unlikely to get it with most games being multiplatform.

The fact is that to get an AMD GPU into a ALU limited scenario you would have to cut down on everything else that will limit it first and thus make a very below par looking game.

Furmark simply shows AMD has a shading power edge that is not reflected in most PC gaming benchmarks! In consoles, that spare unused power WOULD be used! That's it.

You dont have to cut down on anything else that limits it, you max everything that's the point.

 

[Prophecy2k]

Yeah i think a better comparison would be a buldozer against the latest Power7 IBM chip (assuming Xbox 720's CPU will be Power7-based).

 

[Blazkowicz]

Perhaps at the eventual 28nm SOI shrink, they'll be able to increase performance another %25 or so and keep the same TPD (or push the TPD a little higher ~125 for the SOC, sub 200W for the whole system)

I've googled "bulldozer 28nm", and the 28nm process won't be SOI (nor the 20nm one)
well, here's the source http://www.fudzilla.com/processors/item/23907-2013-bulldozer-successor-is-28nm

 

[Gubbi]

BD could be a technically sound choice of CPU for a console. Developers would fully utilize FMA4, avoid the I-cache aliasing issues, the store-queue idiosyncracies and the high penalty of rescheduling a thread to a core in a different module. Developers already coped with far worse this gen.

Business-wise it looks as bad as the original Xbox deal with a sole supplier of chips. AFAIK Sony can't just buy the IP and shop around for fabs, since they or the fab would need to hold a x86 license.

Cheers

 

[McHuj]

I've googled "bulldozer 28nm", and the 28nm process won't be SOI (nor the 20nm one)
well, here's the source http://www.fudzilla.com/processors/item/23907-2013-bulldozer-successor-is-28nm

I've looked for that as well. I was under the impression that you needed a SOI type process to be able to hit the very high clock rates, 3-4+ GHz. All the high performance/high clock IBM chips and the AMD ones are on 45/32nm SOI. Even the XCGPU was only shrunk to a 45nm SOI, when 40nm bulk is available. So I'm a little surprised that they'll be able to move to a bulk process and maintain those levels of performance. If they can, then good.

 

[McHuj]

Business-wise it looks as bad as the original Xbox deal with a sole supplier of chips. AFAIK Sony can't just buy the IP and shop around for fabs, since they or the fab would need to hold a x86 license.

Cheers

That's a good point and I've wondered if there is a reasonable amount of customization that could be done to bulldozer/piledriver module so that the IP can be sold without Intel getting in the middle of it. I'm guessing it would involve removing instructions/extensions, but can it be done in such a way that you basically don't have to design a new core?

 

[hoho]

I wonder how complex would it be to remove the x86->internal microcode translation part and have the code run directly on (really ) bare metal.

 

[Prophecy2k]

BD could be a technically sound choice of CPU for a console. Developers would fully utilize FMA4, avoid the I-cache aliasing issues, the store-queue idiosyncracies and the high penalty of rescheduling a thread to a core in a different module. Developers already coped with far worse this gen.

Business-wise it looks as bad as the original Xbox deal with a sole supplier of chips. AFAIK Sony can't just buy the IP and shop around for fabs, since they or the fab would need to hold a x86 license.

Cheers

Do you suspect that Charlie is incorrect or has been misinformed given this?

I can't see Sony making this kind of mistake, especially considering their experience with console cost reduction with their previous consoles. And also with the problems they've had further reducing the PS3's costs this gen, i'm quite sure they'll be intending to design a PS4 with a fully fleshed out and clear cut future cost reduction strategy.

x86 makes little sense to me, unless AMD is practically offering it to them for free.

 

[MrFox]

If it's x86, what are the chances they'd go with a Larrabee derivative, Knights-who-says-Ni or whatever it'll be called?

 

[3dilettante]

The difference from before is that AMD is more free to fab its chips as a fabless company, and the role of "most desperate to justify its existence" has shifted from IBM's microelectronics division to the struggling x86 designer.
One of the largest arguments against it from AMD's point of view was that the margins were abysmal and limited capacity would be taken away from those lines.
AMD's performance and feature position today is one that lags Intel so severely that the former is not a safe assumption, and its fabless nature means the capacity argument is not as strong.
In fact, one of the trade-offs for the capacity being freed up is that AMD may always be relegated to the inferior bin.

In that scenario, Sony could potentially wrestle AMD into an arrangement where the designer would be on call to design revisions on a cadence closer to what Sony would want for cost savings.
Additionally, if Sony is going for a lot of package-level or interposer integration of disparate parts, the cost savings from owning the CPU IP would be proportionately less.

As far as designs go, BD is an inferior one on the desktop and server. The console space has made do less than top-drawer performance, and it doesn't need many of those features. In that context BD is decent.

The big risk may be what happens if AMD goes under in the life cycle of the PS4.

 

[Blazkowicz]

I wonder how complex would it be to remove the x86->internal microcode translation part and have the code run directly on (really ) bare metal.

from what I believe to understand, I guess the code would grow bigger, wasting bandwith and causing a lot of L1 cache misses

 

[Arwin]

I don't think that is a risk if Sony gets the rights to produce the exact type of chip that they have AMD design for them, as I think is a no-brainer these days for any such contract.

 

[3dilettante]

That's not within AMD's power to give, and still a detraction from a potential x86 console.

 

[Arwin]

That's not within AMD's power to give, and still a detraction from a potential x86 console.

Why not? Didn't that happen for both the PS3 and 360 GPUs?

 

[Tahir2]

Why not? Didn't that happen for both the PS3 and 360 GPUs?

You can't give what you do not own. Please see Intel for an X86 license.

 

[upnorthsox]

The difference from before is that AMD is more free to fab its chips as a fabless company, and the role of "most desperate to justify its existence" has shifted from IBM's microelectronics division to the struggling x86 designer.

Good point

One of the largest arguments against it from AMD's point of view was that the margins were abysmal and limited capacity would be taken away from those lines.
AMD's performance and feature position today is one that lags Intel so severely that the former is not a safe assumption, and its fabless nature means the capacity argument is not as strong.
In fact, one of the trade-offs for the capacity being freed up is that AMD may always be relegated to the inferior bin.

This goes to Gubbi's point, and as you say, they'd be arraigning their own capacity competition. The one way I could see this turning in their favor is if they got some kind of kickback from GloFo for Sony taking over the capacity AMD is (rumored to be) dropping.

In that scenario, Sony could potentially wrestle AMD into an arrangement where the designer would be on call to design revisions on a cadence closer to what Sony would want for cost savings.

Unless they are offering free design for a 20nm shrink, I don't see this as much incentive at all.

Additionally, if Sony is going for a lot of package-level or interposer integration of disparate parts, the cost savings from owning the CPU IP would be proportionately less.

IMO it's the opposite.

As far as designs go, BD is an inferior one on the desktop and server. The console space has made do less than top-drawer performance, and it doesn't need many of those features. In that context BD is decent.

Whether a console is a better fit for BD or not, at this point, going with a BD cpu would entail a PR hit and good PR is exactly what Sony would want from an x86 cpu. At the minimum it would be called a Piledriver variant even if it was Bulldozer thru and thru.

The big risk may be what happens if AMD goes under in the life cycle of the PS4.

Doubtful, but it could surely go thru a very messy filing/restructuring/selloff. While that may not be a big risk, it is another PR hit that Sony shouldn't want to take.

All in all, I still hope to god they aren't planning to go with a ARM cpu.

 

[upnorthsox]

You can't give what you do not own. Please see Intel for an X86 license.

I wonder if there's any wiggle room in AMD's license? Probably not, so Gubbi's point still stands.

 

[upnorthsox]

Some Radeon HD 7870 ([FONT=verdana,geneva]Pitcairn[/FONT]) info. Looks like 130W for the full board (almost 100W for the 7850). 212mm^2, 2.8B transistors, 1GHz clock, 256bit bus, 1280 stream processors, 80 TMUs, 32 ROPS, 2GB 1.2GHz memory (153.6GB/s), etc the 7850 is 860Mhz and same memory. Adverted power is 175-190W and 130-150W (2nd power tune) but this link from Dave B caught my eye:



Pretty impressive performance given the power envelope--I am curious how much memory power draw also plays into this. I do wish more people would offer some more settings in these reviews, e.g. BF3 with MSAA and w/o (ditto MSSAA+FXAA) would be nice. They mention 4xMSAA easily sucks down a ton of performance ("[FONT=verdana,geneva]4xA MSAA will cost you almost a third to half your framerate")[/FONT], so at 39fps for the 7870 at 1920x1200 does that work out to an average closer to 50fps or 60fps?

http://www.guru3d.com/article/amd-radeon-hd-7850-and-7870-review/1
http://www.guru3d.com/article/amd-radeon-hd-7850-and-7870-review/21
http://www.guru3d.com/article/amd-radeon-hd-7850-and-7870-review/24

Anyhow, I thought it was interesting info. If consoles launch on 28nm (with no goal of a fast shrink to accommodate excessive launch power budgets) it would be hard to expect a lot more than an HD 7870. It basically fits the size and the TDP. Maybe a better memory arrangement (stacked memory? see old SA AMD future GPU nugget) and some maturing/reworking over the next couple years allowing a slightly larger die and architectural evolution, but at < 250W on the 28nm node I think it is safe to say that a console using such isn't going to dwarf a 7870 in raw specs. Definitely possible we could/would see something better but nothing that blows it away unless budgets really changed.

The pricing strategy of AMD right now (pegging closely to NV) with the 2GB 7850 coming in at $249 retail (and you know AMD, the manufacturers, and retailers are all getting solid cuts on AMD's current line up), and based on the size of the die, are good indicators also that this sort of chip should fit into a console budget.

To your point, here's the power costs for Nvidia's memory fetches:

Now if you could stay within 2x L3 that'd be a huge power savings. Even 8x L3 would cut your dram power consumption in half.

 

[3dilettante]

This goes to Gubbi's point, and as you say, they'd be arraigning their own capacity competition. The one way I could see this turning in their favor is if they got some kind of kickback from GloFo for Sony taking over the capacity AMD is (rumored to be) dropping.

I'm not certain any capacity AMD drops is something Sony would want. Even value x86 chips sell for more and tolerate worse yields than consumer electronics.
So if AMD says no thanks to a process, there's some doubt Sony could use it.

Unless they are offering free design for a 20nm shrink, I don't see this as much incentive at all.

It's about the only way to make up for Sony's inability to manufacture the x86 itself.

IMO it's the opposite.

The board-level savings happen whether or not Sony owns the core IP, and the cost of the interposer and stacked components makes the SOP more expensive, regardless of ownership.
The cost of the IC that makes up the CPU is a fixed amount, whereas anything on the board or interposer can vary the cost at Sony's discretion. The CPU/GPU chip is still significant, but it would be one of many knobs Sony can turn.

Whether a console is a better fit for BD or not, at this point, going with a BD cpu would entail a PR hit and good PR is exactly what Sony would want from an x86 cpu. At the minimum it would be called a Piledriver variant even if it was Bulldozer thru and thru.

I'm not sure there's a material PR impact for having a BD core, but I don't see a reason why Sony would go for a literal copy of BD version 1.

 

[Tahir2]

I wonder if there's any wiggle room in AMD's license? Probably not, so Gubbi's point still stands.

The "wiggle" was negotiated when Intel settled with AMD recently - the new arrangement allowed another fab to manufacture processors based on the x86 ISA (edit: for AMD) - I think the previous agreement precluded a non-AMD fab from producing AMD based x86 processors.

If I remember correctly this is why we see "Bobcat" CPUs being manufactured at TSMC now and the umbilical cord with GloFlo being cut in recent developments. AMD is free to fab its x86 CPUs anywhere it pleases now and until this agreement expires.