Predict: The Next Generation Console Tech

Status
Not open for further replies.
It doesn't have to be a HMC though as I'm quite sure that stacking through TSVs will certainly not be ready by the time the next-gen consoles ship. It'll be 2.5D with an interposer.
If the latest rumor is true (720 in 2014, PS4 in 2013) Maybe HMC could be ready for 2014. And the high power variant of wideIO might be ready for 2013 on the PS4, as it doesn't require TSV, it is fine with microbumps (Vita).
 
If you put one or two stacks of DRAM dies on an interposer (like the AMD/Amkor prototype we have seen), how do you think the dies are connected and the interfacing to the PCB is done? For sure they will use TSVs in that case too (the mentioned prototypes did). Wire bonding does not appear viable for the relatively large dies involved and the fast communication between them one wants to have.
I think they only need TSV to connect the memory dies together, and it's already being done. They don't need to use wirebond, they can connect the memory to the interposer with microbumps which is a mature process, and much higher density than wirebonding. I mean even the PSP was using that.
 
I think they only need TSV to connect the memory dies together, and it's already being done. They don't need to use wirebond, they can connect the memory to the interposer with microbumps which is a mature process, and much higher density than wirebonding. I mean even the PSP was using that.
You need TSVs to connect the DRAM dies (through the lower ones) to the interposer and you need TSVs to connect the APU/SoC or the whole interposer respectively to the package/PCB (i.e. TSVs in the interposer). And you can use TSVs in an interposer already now (with some additional cost of course), Xilinx does it for their high end FPGAs for instance.
There is no real difference for the need of TSVs if you use HMCs or just stacked DRAM dies with a wider and lower speed interface on an interposer. In the latter case you would omit the lowest logic die of an HMC and run a 512 or 1024 bit bus over a short distance on an interposer and connect directly to the wide memory controller on the APU/SoC die (with micro bumps, yes). You move the function of the logic die in each HMC stack into the APU/SoC and save the HMC interface PHYs. It doesn't touch the TSV requirements other than you lose one die in each DRAM stack.
 
Probably best not to get your hopes up, it is not particularly cost effective to ship a box with two GPUs. If they're basing these ideas on alpha kits, it's a bad premise, because alpha kits tend to be a hodgepodge of hardware that can give a broad-strokes idea of the architecture of the final hardware. They often have extra CPUs, extra GPUs and other bits and pieces that will not make it into the final product.

Think of it this way: If they were thinking of using a APU with 10 compute units, and a GPU with 20, it would be a lot cheaper in terms of design, OS work, fabricating et al to just use a GPU with 30 compute units to start with. Same overall power, no multi-GPU integration headaches. (and fewer transistors, since you wouldn't need to duplicate the common sections)

Maybe it's because they wanted the GPGPU in the APU to work as closely with the CPU as possible using the same memory pool while leaving the faster memory all to the bigger GPU for graphics.
 
You need TSVs to connect the DRAM dies (through the lower ones) to the interposer and you need TSVs to connect the APU/SoC or the whole interposer respectively to the package/PCB (i.e. TSVs in the interposer). And you can use TSVs in an interposer already now (with some additional cost of course), Xilinx does it for their high end FPGAs for instance.
There is no real difference for the need of TSVs if you use HMCs or just stacked DRAM dies with a wider and lower speed interface on an interposer. In the latter case you would omit the lowest logic die of an HMC and run a 512 or 1024 bit bus over a short distance on an interposer and connect directly to the wide memory controller on the APU/SoC die (with micro bumps, yes). You move the function of the logic die in each HMC stack into the APU/SoC and save the HMC interface PHYs. It doesn't touch the TSV requirements other than you lose one die in each DRAM stack.
Does the SoC absolutely require TSV to the interposer?
I mean couldn't it be also connected to the interposer with microbumps?
 
Does the SoC absolutely require TSV to the interposer?
I mean couldn't it be also connected to the interposer with microbumps?
How do you get signals from the PCB to the die on top of the interposer through the interposer? Yes, with TSVs! ;)
 
If Xbox 720 will not use an APU, can they add VMX/VSX to the CPU?
To which CPU? If it's an x86 CPU, there will be AVX support of course (even Jaguar has it). And an IBM CPU (which appears highly unlikely at this stage) would probably have it right from the start. So what do you want to add?
 
To which CPU? If it's an x86 CPU, there will be AVX support of course (even Jaguar has it). And an IBM CPU (which appears highly unlikely at this stage) would probably have it right from the start. So what do you want to add?

I was thinking that Jaguar will not have VMX support, but I was wrong.
 
But we are also moving towards longer life cycles.
We don't know that; not at all! We've had one long generation that's been unexpectedly capped with a shift to throwaway technology in portable gadgets. We may see MS and/or Sony planning a progressive architecture, or a 4 year cycle, or a 6 year or a 10 year, or even planning on abandoning fixed hardware altogether after this gen. They may not even have much of a plan for the next-gen and will see what happens over the first two years before thinking about where to go next. I'm pretty sure MS had no idea when 360 launched that they'd introduce a full-body motion controller in its 5th year to rejuvenate platform interest. There are way too many variables to pick hardware based on a business plan.
 
I honestly don't see the rush for 2013 if 2.5D or HMC isn't ready. It's such a potential game changer in my mind it's be stupid to force something out in 2013 just to be out in 2013. Yes people will complain that the gen has dragged on too long some more but you don't want to be on the wrong side fo such a technology shift.
 
Last edited by a moderator:
I honestly don't see the rush for 2013 if 2.5D or HMC isn't ready. It's such a potential game changer in my mind it's be stupid to force something out in 2013 just to be out in 2013. Yes peopel will complain that the gen has dragged on too long some more but you don't want to be on the wrong side fo such a technology shift.

Yeah, but my guess is you'll be waiting until 2015 or later to use those. Then you've lost the generation.

Dont forget this gen has already dragged on way longer than many think is healthy. 2013 will be 8 years of 360 (where past gens were typically 5 years) and sales are falling.
 
If it's late 2015 or later I kind of agree with you. At that point you don't have a choice but if it's any earlier pushing out a machine in 2013 seems to be a mistake if it doesn't have it.
 
I don't think they can ever get rid of the glowing orbs. There's just no alternative that would work as reliably. They can be measures from any angle. They remain in view even when not pointed at the tv/camera. You don't need a super-high resolution camera to get good measurements (important for high frame rate input). Unfortunately, image is important and gamers will always shun the "clown noses" for looking dumb, no matter how much better they work.
 
Status
Not open for further replies.
Back
Top