Predict: The Next Generation Console Tech

There's really no need for gigabytes of texture memory when you can squeeze four to six layers of unique texels for every pixel in a 2MP frame in less than 100MB.

This is about establishing and managing working sets. PCs have never forced developers to become very good at that. Consoles are much different.

I expect no more than 2GB if UMA, 3GB if split pools happen again. All platform holders will seek to improve margins on the base hardware. Console prices this gen have been too high, even despite unsustainable hardware subsidies.

 

Why? Are you expecting a 128-bit bus? Using 32-bit chips, that's 4 chips minimum, and there's not going to be that much of a price difference between 2Gb and 1Gb chips a few years from now...

 

While I consider 8GB to be too much, 2GB is probably not going be enough either.
Launch titles in particular would suffer significantly, because developers would probably not get enough time to maximize efficiency and develop new streaming and virtual texturing systems (which apparently need to be fitted to the specifics of the hardware).

 

I don't know how much abstraction there's in the unreal sdk, but maybe day one games will use UE3.x with texture streaming and all, hitting the "compile for nextbox
or maybe not"

 

Mainly because optical media speeds haven't scaled fast enough to make good use of so much memory. Load times this generation are bad enough. If we expect 8x~10x BD drives in the next console generation, drive speeds will have scaled only by a factor of 4~5. If you then try to fill 8 to 16 times the memory, it follows that load times would get significantly worse. Not looking forward to ten-minute+ installs either, by the way.

And also because I really don't see the pressure to improve surface fidelity by a factor of 8 or more. 1080p will be the TV standard for a good while. We're not that far off.

And again, this gen was too expensive. If you take current console specs and scale them up in line with litography advances, you're looking at the same outrageous component costs, massive hardware subsidies and unattractive price points all over again. I'd prefer if the platforms are specced a little more conservatively if it makes the market healthier, earlier.

edit: and more to your question, I'm expecting the usual choice of narrow, high-speed interfaces to minimize PCB cost. GDDR, more Rambus shenanigans or custom mojo. Whatever it may be specifically, I don't know.

 

Yeah, but there is a general criticism or rather, a wish, from nearly all developers to have more RAM on the current systems, and upping it to 4GB isn't as complicated or expensive as 8GB would be.

 

It'll be interesting to see what launch prices they target. Growth prospects may have become limited over the course of this generation because of the growth of mobile.

Yes I know core gamers will never settle for mobile games but tablets alone are selling at a rate of over 10 million a quarter. When household discretionary incomes are already being drawn to other electronics devices, consoles and especially $60 games become harder-sells.

Does that mean they use current architectures but maybe drop BC? There's been some fanciful speculation like 50 GB SSDs for caching data. Doesn't seem likely given certain economic realities.

 

I don't understand this mindset at all. More memory decrease the stress put on optical media.

With everything else the same, less memory will require more streaming of textures, smaller buffers for preloading and thus more texture pop-in.

I expect at least 4GB. Whoever launches first of MS and Sony risks the counterpart doubling up on RAM and having to sit through a whole generation of Lens of Truth and Eurogamer VS. reviews declaring the competitor the better choice.

We are talking about less than $20 worth of DRAM for 4 GB for a launch console. In 2005 DRAM was 10% of the $525 BOM of the premium 360.

Cheers

 

What good technologies are there for console to peripheral high bandwidth/low power use wireless networking? Are they going to stick to wireless N for the VR/Tablet etc peripherals or is there something else out there which will be useful?

 

Epic seems to expect the next generation:

http://www.vg247.com/2011/09/06/epics-capps-gears-of-war-3-and-the-next-generation/


“‘What would you like to see?’ I can’t tell you, because I know what’s going on.”

Despite the silence, though, Capps is willing to concede that developers Epic’s engaging with on UE4, the company’s next generation engine, are pumped.

“We’ve only talked to a small number, but the ones we have are very excited because we’re doing a lot of core work that’s going to make it easier for them to scale up, especially with lots of processors,” he says."

 

https://www.power.org/events/2010_ISSCC/Wire_Speed_Presentation_5.5_-_Final4.pdf

Are we sure than A2 cores are good for gaming applications?

I mean, it's better an 8 A2 cores with 32 slow threads or a 4 Power7 cores with 16 faster threads (and out of order architecture)?

By the way, look what i found:
http://realworldtech.com/page.cfm?ArticleID=RWT021511004545&p=3

IBM eDRAM's density is > 11Mbit/mm^2 @ 32nm. Which means that in 90 mm^2 ,the same die-size of Xbox360 daughter die at launch, they could pack 120+ Mb of eDRAM.

 

well point is I don't see how 4 cores would qualify as "lot of cores", what's better? I don't know but I expect 8 Power A2 cores to consumes less than the 4 power7 cores.

The problem is that the process doesn't seem ready, power A2 should launch soon @45nm, blue gene also use IBM 45nm process. I don't expect MS to use this process but IBM/GF standard 32nm process, so no EDRAM. I might be wrong by the way, but I believe MS would easily trade lot of cache or scratch pad memory for more CPU and GPU cores especially if Charlie is right about the system using a SoC/APU. there is also cost I don't think that IBM process allowing for edram is cheap IBM doesn't care on the market they are aiming for with POwer 7 /A2 console manufacturers do. Using this process also implies being stuck with a foundry (IBM) for the whole system life cycle, it's also unclear at which pace this process will advance vs simpler processes which may delay cost reduction through shrinking.

 

RAM & Streaming Cache Implications: http://forum.beyond3d.com/showthread.php?t=60856

 

console usually launch with no brand new process, but considering a 2013/14 date that numbers are interesting
and the daughter die can be on a different process anyway

 

Are you sure the heat allows for that? I thought power and heat were now the biggest reasons why die shrinks give diminishing returns.

 

See this youtube also

http://www.youtube.com/watch?v=6JoU08ecIa8&t=2m34s

Mike Capps hints Gears 3 is the culmination for UE3 and seems to shift focus to UE4. I think next gen is coming 2012-2013...

 

If Wuu is considered next-gen then 2012 is pretty much a sure bet. Only question is when will Sony-MS follow.

 

I have the same thought about 2013.

See this post.

http://forum.beyond3d.com/showpost.php?p=1578664&postcount=7459


I have impressions there is evidence is something expected to happen very important for Sony in 2013, perhaps even a new console, see the statements carefully and probably even some thirdies developers now have at least some information about the reasonably practicable to wait for these consoles (I speak mainly of Epic leader in licenses in the current generation).


I don't think sony and ms will leave the current leader of the generation (in total numbers of consoles reaching 80/90 millions units) out in front with more than one year ahead, even if it is a console (wiiu) supposedly only slightly more powerful than ps360.

 

http://www.engadget.com/2011/09/07/sony-announces-a-4k-projector-for-the-home-at-cedia-prices-hmz/
less than a dollar per column of pixels, great

But about the memory, I know Sony allows users to change the hard drive.
Would it be a real big stretch... to have a DDR5 slot which allows the user to upgrade the cache?
It could be fairly simple for developers to implement, and it would allow users to cut loading times, if they want to pay for that. The console only needs an extra connector/lane/memory controller which is less expensive than the ram itself would be (probably)

 

Doesn't make sense because games need to support baseline hardware which would limit how engines can be designed. It would perhaps work for occasional loadtime reduction and higher res textures/less popup but it wouldn't allow radical engine redesign requiring fast streaming. Also it would add another path that doesn't just need to be coded but also tested. It can add quite significant cost to QA.