Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]

Status
Not open for further replies.
I sort of feel XBX is their next generation console. Shower thoughts:

2019:
XBX: exclusives announced to be delivered 2020. This will make the end of generation for XBO.
XBO ramps down production, smaller form factor maybe, device is still fully supported, developers are free to target the machine as they choose.
XBox Family: full mouse and keyboard
MS Office suite, other video editing software.
Some form of continuum feature proper web browsing.

The objective is to repurpose XBO as an all in one device, a replacement for laptops or PCs as long as you can hook it up to a 1080p monitor. This keeps the users engaged in the ecosystem while bringing new features to he platform. Think Raspberry PI type tinker device. Educational, do some development and scripting.

Becomes a device children and students will grow up with, from play to work. Minecraft to Tinker to Office.

2020-2021: Xbox 2 announced.

XBox Family: full mouse and keyboard happen in 2019 , why not in 2018?
 
I've read here and there that changing memory type may introduce incompatibility. So, if there was to be another mid generation upgrade, the transition from GDDR6 to HBM3 would likely break compatibility.

Clearly, Microsoft have found a way around this. Is there any way or likelihood that Sony could or will do the same?
 
I've read here and there that changing memory type may introduce incompatibility. So, if there was to be another mid generation upgrade, the transition from GDDR6 to HBM3 would likely break compatibility.

Clearly, Microsoft have found a way around this. Is there any way or likelihood that Sony could or will do the same?
Doesn't mean incompatible, it only means it's not a cycle-exact behaviour. Which MS doesn't care about. They used a very high bandwidth at least as fast as the worst case access of read/write edram plus ddr3.

I think gddr5x would have caused problems because of the nasty 64 bytes prefetch which gddr6 solved.

Moving from gddr5 to gddr6 can be a perfectly identical behaviour with half the chips at twice the speed. A bit optimistic, I believe we might even see it on a Pro revision before next gen.

I suppose a transition to HBM would also work depending on how much they want a perfect emulation mode like they did on the Pro, it does seem overkill. We have yet to find a single game which doesn't work in boost mode. As long as it has the same number of banks and channels and more bandwidth it should be equal or better during emulation.
 
I've read here and there that changing memory type may introduce incompatibility. So, if there was to be another mid generation upgrade, the transition from GDDR6 to HBM3 would likely break compatibility.

Clearly, Microsoft have found a way around this. Is there any way or likelihood that Sony could or will do the same?
The memory controllers, cache subsystem, and firmware/OS are expected to keep a consistent model of memory functionality.
Different types of memory may not perform in exactly the same way, but code should normally function.
Chips have had multi-standard memory controllers, or the same cores/ISA have been added to chips with different DRAM, or different chipsets for the same CPUs linked to different memory without problems.

Microsoft's ESRAM was treated as if it were regular memory, addressed normally with accesses routed transparently through a crossbar. The performance characteristics weren't the same, but the accesses as far as the code was concerned had the same end result.

If there's a specific quirk to the system that the software somehow depended on, I suppose there could be an incompatibility. That would assume that the software would be given that level of access, and given the complexity and importance of low-level behaviors most such features would be hidden.
One possible quirk mentioned for Scorpio was how paired memory channels were considered coherent, or so it was relayed during the Hotchips presentation. That seems to be in a GPU context and likely stems from how there are fewer L2 slices than memory controllers, meaning the same L2 can communicate to multiple channels.
If the Xbox One did not have that feature, it wouldn't really introduce any incompatibility. It might in some instances be an issue if a future chip removed that feature if software drilled down to use such a subtle change, although that would be the result of a more total change to the on-die architecture along with possible changes to the memory interface.
 
Sony made PSX's with different RAM characteristics when games had much more access to the hardware.
Alpha blending is significantly faster with SGRAM, but blitting narrow vertical strips of the display is slower than early model VRAM.
 
I sort of feel XBX is their next generation console. Shower thoughts:

2019:
XBX: exclusives announced to be delivered 2020. This will make the end of generation for XBO.
XBO ramps down production, smaller form factor maybe, device is still fully supported, developers are free to target the machine as they choose.
XBox Family: full mouse and keyboard
MS Office suite, other video editing software.
Some form of continuum feature proper web browsing.

The objective is to repurpose XBO as an all in one device, a replacement for laptops or PCs as long as you can hook it up to a 1080p monitor. This keeps the users engaged in the ecosystem while bringing new features to he platform. Think Raspberry PI type tinker device. Educational, do some development and scripting.

Becomes a device children and students will grow up with, from play to work. Minecraft to Tinker to Office.

2020-2021: Xbox 2 announced.
 
All of Sony's and Microsoft's consoles this generation have only had access to a partial amount of their main pool of memory.

Is there any possibility, next generation, of a solution to that, or is a significant portion of main memory usage inevitable?

It's always felt like such a waste, seeing multiple gigabytes of GDDR5 used for menus and Netflix. Even 16GB of main memory with 1GB reserved for the system would feel like a huge boost from this generation, even from X1X.

Also, assuming the PS5 contains a system ARM, what's a strong contender for its memory? I think it's been mentioned in this thread that DDR4 will become cheaper than DDR3 over the years, but will it be the cheapest option that can manage 4K media, or might LPDDR4 suffice and become cheaper?
 
All of Sony's and Microsoft's consoles this generation have only had access to a partial amount of their main pool of memory.

Is there any possibility, next generation, of a solution to that, or is a significant portion of main memory usage inevitable?

It's always felt like such a waste, seeing multiple gigabytes of GDDR5 used for menus and Netflix. Even 16GB of main memory with 1GB reserved for the system would feel like a huge boost from this generation, even from X1X.
I think the way PS4/Pro used paged memory, swapping it to the low speed ddr3, seems to be the most interesting solution yet. They could do it more agressively next time.

Next gen, the OS and permanent services and apps could stay around 2GB, and anything above for apps would be swapped out to the ddr3 pool when returning to the game, then the game can use that amount of memory back.

With this trick:
24GB gddr6 + 4GB ddr3
Game memory while playing: 22GB (4.5x PS4, 4x the Pro)
Effective App and OS memory when the game is in background: 6GB

The only disadvantage is the additional hassle for the devs dealing with the game pause/restore.
 
I think the way PS4/Pro used paged memory, swapping it to the low speed ddr3, seems to be the most interesting solution yet. They could do it more agressively next time.

Next gen, the OS and permanent services and apps could stay around 2GB, and anything above for apps would be swapped out to the ddr3 pool when returning to the game, then the game can use that amount of memory back.

With this trick:
24GB gddr6 + 4GB ddr3
Game memory while playing: 22GB (4.5x PS4, 4x the Pro)
Effective App and OS memory when the game is in background: 6GB

The only disadvantage is the additional hassle for the devs dealing with the game pause/restore.
I don't think they have to do it for the Pro. From what I gathered it should be transparently done by the OS. and that's not how I imagined the whole procedure. I thought half of that 1GB ddr3 were constantly reserved for the OS apps. The OS allocates the DDR3 for the external apps if they need it. Games never use the DDR3 on Pro.

24GB gddr6 + 4GB ddr3
That was my idea anyway ! :yep2: How much does DDR3 cost ? 4GB / 8GB DDR3 ? won't the DDR4 be less expensive in 2, 3 or 4 years ? I think for the long term it could be more economical to use DDR4.
 
It might be troublesome or not worth it to have multiple memory busses running from the main SOC. I guess it depends on how flexible the memory controller is. *shrug*
 
If there was any more room for I/O on the main SoC, maybe they should simply use it to make the main ram wider. The ddr3 on ps4 and pro is different because it's going through the 4x pcie which is necessary to connect the south bridge anyway. So in terms of I/O it comes for free.
 
I think the way PS4/Pro used paged memory, swapping it to the low speed ddr3, seems to be the most interesting solution yet. They could do it more agressively next time.

Next gen, the OS and permanent services and apps could stay around 2GB, and anything above for apps would be swapped out to the ddr3 pool when returning to the game, then the game can use that amount of memory back.

With this trick:
24GB gddr6 + 4GB ddr3
Game memory while playing: 22GB (4.5x PS4, 4x the Pro)
Effective App and OS memory when the game is in background: 6GB

The only disadvantage is the additional hassle for the devs dealing with the game pause/restore.

As Globalisateur says, and after a little bit of Googling, DDR4 will soon become cheaper than DDR3. I can't find any information on the price of LPDDR4, but I've encountered one reference to it being more expensive. So the low power versions are out of the window.

Whilst Googling for some DDR4 information, I kept getting results for 4K Android TV boxes with 4GB of DDR4, so that's presumably sufficient for streaming 4K media. If that's so, what's the likelihood that the PS5 would beef up its secondary CPU to handle media apps and the XMB?

Might that require too expensive of a CPU upgrade to be worthwhile, or would freeing up a couple of GB of main memory be justification enough?

Personally, I'd quite like a beefed up secondary CPU that would handle all apps; all system functions such as chat, streaming, and video recording; and picture-in-picture functionality for any of these, which I'd particularly like when using PSVR.
 
More memory bandwidth is definitely needed next generation. This gen at a 1080p res I found the abundant use of 1024x1024 smeary textures and low AF in so many games unappealing. Alot of developers had difficulty implementing larger textures and decent filtering in games. Namely because of memory contention since the consoles went the single monolithic die (APU) route this gen. Some devs like Naughty Dog and Guerrilla Games got around the memory bandwidth issues by using material shaders in place of textures for a variety of surfaces like clothing on Nate and environment surfaces.

Hopefully they don't go with 256bit bus + GDDR6 and call it a day. +300bit width is needed or it will hold back numerous triple AAA games for another 5-7 years. Devs will be forced make a compromise between memory bandwidth demanding visuals or cpu demanding complex worlds like Fallout 4 where the cpu has to access main memory alot.
 
Last edited:
Beefing up the Arm chip for multimedia and xmb ... I dont know if it would be worth all the engineering rework (software and OS as well as hardware) for the payoff of lower power system when not gaming.
Probably the most important counter point. If AMD were to design this into their designs then I see it moving into the consoles. But I don’t see either MS or Sony willing to invest in customizing the chip to do this.
 
Not for lower power, but for the sake of freeing up a large amount of more expensive memory without having to burden developers and users with a couple of GB swapping in and out of main memory. Maybe lower power would come as a result of this, but doesn't the X1X use less power in standby than the PS4Pro?

In reference to engineering work, would there be that much, considering the PS4 already operates like this to some extent? I know it's a gross oversimplification, but would it be much effort to take the existing PS4 design, replace the 1GB of DDR3 with 4GB of DDR4, attach some video decoding blocks and whatever else is needed to run 4K menus?

GDDR6 or HBM3 will be relatively expensive and reserving 2 or 3 GB of it in case someone wants to watch an episode of Rick and Morty seems like pissing away money.
 
In reference to engineering work, would there be that much, considering the PS4 already operates like this to some extent? I know it's a gross oversimplification, but would it be much effort to take the existing PS4 design, replace the 1GB of DDR3 with 4GB of DDR4, attach some video decoding blocks and whatever else is needed to run 4K menus?

Yes. Much work. Hardware. OS. Software.
 
Status
Not open for further replies.
Back
Top