My feelings
Last edited:
Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]
- Thread starter Shifty Geezer
- Start date
- Status
My feelings
Megadrive1988
Veteran
Microsoft and Sony, through AMD, will both be using the same technologies:
Zen CPU architecture
next gen Radeon GPU architecture (whatever is after Arctic Islands)
HBM2 or the 1.5x more efficient HBM2 that was in a recent roadmap
APU/SoC on a 10nm FinFET+ process
HDMI 2.0a or 2.0x
100GB Blu-ray
multi TB HDD
Much like this generation, next gen consoles will have the same architecture. The difference will be in number of CPU cores, number of GPU shader, clockspeed and HBM2 configuration, and possibly the size and type of external off-chip system RAM assuming there is some, not just HBM2.
The real differences will be in controller interface and VR options. I don't see a conventional Dual Shock 5, but something that's partly new. There are reports that Ken Kutaragi is working with SCE on new human interface technologies. Perhaps a new controller that isn't a gimmick, that offers some new approaches to control, but also keeping what works best about dual analog. Not sure. Same for MS.
Sony has PlayStationVR and any future PSVR headset. Microsoft will almost certainly offer Oculus Rift support. I don't think HoloLens has any bearing on the next Xbox.
Zen CPU architecture
next gen Radeon GPU architecture (whatever is after Arctic Islands)
HBM2 or the 1.5x more efficient HBM2 that was in a recent roadmap
APU/SoC on a 10nm FinFET+ process
HDMI 2.0a or 2.0x
100GB Blu-ray
multi TB HDD
Much like this generation, next gen consoles will have the same architecture. The difference will be in number of CPU cores, number of GPU shader, clockspeed and HBM2 configuration, and possibly the size and type of external off-chip system RAM assuming there is some, not just HBM2.
The real differences will be in controller interface and VR options. I don't see a conventional Dual Shock 5, but something that's partly new. There are reports that Ken Kutaragi is working with SCE on new human interface technologies. Perhaps a new controller that isn't a gimmick, that offers some new approaches to control, but also keeping what works best about dual analog. Not sure. Same for MS.
Sony has PlayStationVR and any future PSVR headset. Microsoft will almost certainly offer Oculus Rift support. I don't think HoloLens has any bearing on the next Xbox.
I'm feeling similarly towards raw computing capacity and available memory.
Consoles in essence have become slightly customized pc hardware. Improvements in manufacturing technology have slowed down and new processes cost more per transistor than old ones. Look at the rate pc hardware has improved since move to 28nm manufacturing technology. Even a 5x improvement on raw specs for 2019 release is a stretch on a 399$ 150W box.
I wouldn't even be surprised if cpu cores get scaled down to 2 or 3 cores and they just are more efficient and clock higher(dynamically). That would leave some extra space for gpu.
For memory amount I'm not sure consoles would need more than 12-16GB. How much unique memory there can be used per frame anyway? It's rather possible that some memory is used, freed and then re-used, freed etc. per frame. What amount of data really needs to persist between frames?
Somehow I feel it would be better to use money towards having really fast and functional memory such as 12-16GB HBM2. If then there is some money left over to be spent use it for faster mass storage to accelerate fetching unique content. Intel cross point could be rather unique in what it enables...
Somehow I feel it would be better to use money towards having really fast and functional memory such as 12-16GB HBM2. If then there is some money left over to be spent use it for faster mass storage to accelerate fetching unique content. Intel cross point could be rather unique in what it enables...
Betanumerical
Veteran
Why an FPGA?.
I keep thinking about a single versus dual memory pool. I think if Sony and MS want to optimize performance/watt and performance/cost, dual memory pools are the way to go.
It terms of video memory, I think HBM2 will be the solution. I can't imagine a bus bigger than 256-bits that a GDDR solution would require. That's too much die are for a console and no one will want to put down 24 or 32 ram modules on the mother board. I just don't see GDDR being viable. The question I have is 2 or 4 HBM stacks? I assume the failure rates for getting 4 modules on the interposer working are drastically higher than only two so 2 HBM2 stacks might be optimal in terms of cost. I'd guess 8/16GB of HBM2 at 512 GB/sec.
To that add a single/dual channel for DDR4 with a 32/64GB pool. I know devs wouldn't like it, but anyone porting to PC nowadays has to deal with this already. This effectively makes the consoles look even more PC like so in the end I don't think this is that far out there.
We're already wasting 2-3 GB of memory in the current consoles for OS and App related functionality, I can't imagine they would want to dedicate several GB of HMB memory to that in the future. That seems wasteful and inefficient.
RAM expectations:
1. Conservative: 24GB DDR4 + 8 GB HBM
2. Middle: 32GB DDR4 + 16GB HBM (This should be where mid tier (970-level) PC GPU's are in 3/4 years)
3. Shocking: 32GB HBM only
It terms of video memory, I think HBM2 will be the solution. I can't imagine a bus bigger than 256-bits that a GDDR solution would require. That's too much die are for a console and no one will want to put down 24 or 32 ram modules on the mother board. I just don't see GDDR being viable. The question I have is 2 or 4 HBM stacks? I assume the failure rates for getting 4 modules on the interposer working are drastically higher than only two so 2 HBM2 stacks might be optimal in terms of cost. I'd guess 8/16GB of HBM2 at 512 GB/sec.
To that add a single/dual channel for DDR4 with a 32/64GB pool. I know devs wouldn't like it, but anyone porting to PC nowadays has to deal with this already. This effectively makes the consoles look even more PC like so in the end I don't think this is that far out there.
We're already wasting 2-3 GB of memory in the current consoles for OS and App related functionality, I can't imagine they would want to dedicate several GB of HMB memory to that in the future. That seems wasteful and inefficient.
RAM expectations:
1. Conservative: 24GB DDR4 + 8 GB HBM
2. Middle: 32GB DDR4 + 16GB HBM (This should be where mid tier (970-level) PC GPU's are in 3/4 years)
3. Shocking: 32GB HBM only
For game consoles wouldn't it be smarter to have the opposite ratios of memory types? Assuming the OS and app related memory usage is around 2-3 GB now then 8GB of DDR4 should be enough for to take care of that in the next gen OS functionality and it will probably not cost very much in a few years.
I see something like. 8GB DDR4 + 24GB HBM
Also if Nvidia's gpu roadmap is anything to go by then it's possible a mid-tier card in 2018/19 could have 32GB of memory. Pascal is launching next year will support up to 32GB. Volta launching in 2018 will support up to 64GB.
Currently, we have 8Gbit ram module i think... Do we know if there's going to be higher density in the future? Because i seriously doubt any manufacturer would go to 32 chips to get 32GB DDR4. Hell, 16 might be already an issue.
Megadrive1988
Veteran
HBM2 can go to 1.2TB/s w/ 32GB. Why not use that, and save on costs by not having an external bus and external RAM chips. The HBM interposer will be expensive enough.
Allow total power usage to go upto ~190 watts (still below the launch 360/PS3 models) and go for a good balance of CPU and GPU performance.
Allow total power usage to go upto ~190 watts (still below the launch 360/PS3 models) and go for a good balance of CPU and GPU performance.
My proposal were based on the fact that you are build a $399 system. True HBM can go to 1 TB/sec and 32 GB, but that will require 4 stacks like Fury. I'm assuming costs for placing 4 stacks are more than 4X higher than just putting 1 stack down on the interposer. I wish I knew more on the economics of this.
I was think if it came down to it, they could just use an off the shelf DIMM(s) for the system RAM.
IHS said that at launch Sony paid $100 for PS4 APU and $88 for 16 chips of GDDR5 [entire console costed $381], so we will probably see the repeat of that for next console.
For PS5 Sony would be wise to put everything on a single interposer, since that will also simplify the motherboard, power delivery and cooling. We will have 32GB of HBM2 in professional GPU cards in 2016/early 2017, I think consoles will match that only if release schedule calls for early deployment. If PS5 comes in late 2019 or late 2020, we should get more. As for OS ram pool, they should not add new physical chips for it. HBM is plentiful, there is space for both OS and game apps.
The only way I can see them adding separate chips for OS use is if they for example create beefed up version of VITA package [great ARM with new GPU, awesome video encoder/decoder built-in, stack on that few gigs of ram] and smack that on motherboard [or central interposer itself] to not only run OS but also handle suspend state activity. This would enable entire X86+Radeon+HMB package to be completely dedicated to gaming.
Chances for this are less than 1% IMO.
For PS5 Sony would be wise to put everything on a single interposer, since that will also simplify the motherboard, power delivery and cooling. We will have 32GB of HBM2 in professional GPU cards in 2016/early 2017, I think consoles will match that only if release schedule calls for early deployment. If PS5 comes in late 2019 or late 2020, we should get more. As for OS ram pool, they should not add new physical chips for it. HBM is plentiful, there is space for both OS and game apps.
The only way I can see them adding separate chips for OS use is if they for example create beefed up version of VITA package [great ARM with new GPU, awesome video encoder/decoder built-in, stack on that few gigs of ram] and smack that on motherboard [or central interposer itself] to not only run OS but also handle suspend state activity. This would enable entire X86+Radeon+HMB package to be completely dedicated to gaming.
Chances for this are less than 1% IMO.
Last edited:
How much would 32GB of HBM cost in 2018? I'm guessing more than $88.IHS said that at launch Sony paid $100 for PS4 APU and $88 for 16 chips of GDDR5 [entire console costed $381], so we will probably see the repeat of that for next console.
For PS5 Sony would be wise to put everything on a single interposer, since that will also simplify the motherboard, power delivery and cooling. We will have 32GB of HBM2 in professional GPU cards in 2016/early 2017, I think consoles will match that only if release schedule calls for early deployment. If PS5 comes in late 2019 or late 2020, we should get more. As for OS ram pool, they should not add new physical chips for it. HBM is plentiful, there is space for both OS and game apps.
The only way I can see them adding separate chips for OS use is if they for example create beefed up version of VITA package [great ARM with new GPU, awesome video encoder/decoder built-in, stack on that few gigs of ram] and smack that on motherboard [or central interposer itself] to not only run OS but also handle suspend state activity. This would enable entire X86+Radeon+HMB package to be completely dedicated to gaming.
Nvidia Volta is supposed to launch in 2018 with support of up to 64GB of memory. I suspect if mid (970ish) gpu's are using 32GB of HBM in 2018 then console makers would have no choice but to use that amount.
Nvidia Roadmaps aren't worth the paper they're written on, we're already onto the second rev of Denver if you went by their first announcement. Nvidia and AmD both are hostage to what TSMC and GloFo can provide process wise. I'd expect the next device to be on 14nm which limits how much can be squeezed in.
Nvidia Roadmaps aren't worth the time invested to type it out on keyboard. They change it [delay things] all the time.
As does every company. Engineering is hard.
- Status