Predict: The Next Generation Console Tech
- Thread starter Acert93
- Start date
- Status
brain_stew
Regular
In-game web browsing, dur!
Tbf, having this feature built into the Steam overlay is pretty nice and useful at times, so i wouldn't be averse to having the option on consoles. Of course there's the UI problems but since all the machines will probably have some sort of pointing device next generation maybe it won't be too bad. Browsing on the Wii works quite well for instance.
I guess in Microsofts case we're looking at an Arcade and a Premium SKU for cost and the flash memory would be a good unifier between the two SKUs to give consistant performance overall. I don't believe that HDD streaming is very effective relative to the performance of flash as constant as simultainious read/write kills performance and you still have much higher latency.
I think in terms of useability the long initial load times for games this generation is a concern (1-2 minutes?) . It would be a pretty useful technology to be able to load straight into a game you've been playing or the one previous without too much waiting and on top of that being able to suspend and shut down a console at any point and return to your game would also be a bonus.
I would say having 16GB (8-10GB cache + 1-2 redundancy + 4-7GB user) or 32GB (12-20GB cache with 2-4 redundancy and 8-18GB user space) could make sense. You could give the Premium SKU half the quantity of the Arcade SKU just for cache to save money as well.
If you can load data into your pool of ram faster with less latency then you can get away with less ram. System ram in the PC architectures is a big slow pool whereas console architectures have more bandwidth so they don't need as much data in reserve. On a console the system ram represents more what the system is using at present and on the PC it also includes space for data which may or may not be needed. Its a JIT approach vs delivery in bulk and so long as you can control the data delivery the JIT approach is far more efficient.
Not really. Having faster RAM will let you do more computation on what's in memory, but the limiting factor is still media read bandwidth and seek performance. Whether that media is a 5400 RPM hard disk or a Blu-ray Drive, it's still extremely slow relative to system memory.
Data streaming can be a difficult problem. You have to be able to design your engine and production asset pipeline with streaming in mind so you can generate the necessary asset dependency graph for streaming/preloading of content at runtime. Retrofitting an existing game to do this when there are already thousands of assets is extremely difficult, which is why many PC to console ports suffer.
brain_stew
Regular
Yeah, that seems like the perfect sort of solution as it gives the developers the fast cache that they want, still allows Microsoft to sell digitally distributed content to all users and gets rid of that large fixed cost of the HDD from the base unit. It has the potential to deliver a better end user experience whilst costing Microsoft less in the long run, seems potentially like a win-win situation to me.
Which is why I stated that it depends on how fast you can load data into main memory. If you're streaming from flash with 250MB/S or greater throughput and 0.1ms read latency then you can be pretty sure you can get data from cache into working memory quickly.
A developer with access to a system like that would only need the data in memory which was currently relevant to the present needs of the game engine. Data which could be needed within 1-5 frames could still reside on the cache and be called as needed and this is still applicable for engines which run at 16ms frame time or 60hz.
Now I think it begs the question, how much memory would you need if you had a cache big enough and fast enough that only data in use needs to reside in main memory?
@Brainstew: Its good to have someone agreeing with me, thanks.
Lastly as an aside: If the next generation architectures and the current cheaper architectures co-exist side by side does that mean that the console architecture for the first mover can be a loss leader so long as the former generation hardware is still being profitably sold?
I love the Microsoft SKU system because it gives you a lot of room for possibilities. So if they are the first mover of the next generation at the end of 2011 they could possibly release their mainline SKU at $399 for a loss and sell the Arcade for $199 with 10-20GB flash for a profit with Natal then next year release a $299 Arcade and break even overall on both SKUs and drop the previous generation Arcade to $149 the year after in 2012.
Using a fast flash memory as cache would be great, but i still think MS will put a slow, fat hard drive into their next console: digital delivering, and game installing (20+ GB) ask for it. I can see the Arcade SKU with at least 128 Gb (5400rpm) and the Premium SKU with 512 GB(5400rpm).
And for the ram:
http://www.hardware.info/en-US/extc...ZyA/Hynix_develops_first_40nm_2Gb_GDDR5_chip/
16 of these would be plenty enough.
And for the ram:
http://www.hardware.info/en-US/extc...ZyA/Hynix_develops_first_40nm_2Gb_GDDR5_chip/
16 of these would be plenty enough.
upnorthsox
Veteran
So you'd need a 1024 bit bus then, I'm sceptical. Even 4 of them would require a 256 bit bus which is not a given for next gen (even though it should be the min).
well, i didn't consider that.
But with clamshell mode, you could use a 256bit MC and connect to 8 of those chips. But i hoped for more than 2 Gb..
Blazkowicz
Legend
you're off by a factor of 2, and you can easily double the number of chips with clamshell mode.
so that gives 2GB memory on a 128bit bus, using 8 chips, which is a very probable outcome. can be scaled to 4 chips if there are even bigger memory chips down the road.
I agree that bandwith and memory capacity is limited by board complexity. It's another reason I don't believe in a 5 teraflop GPU - you won't be able to feed it
so that gives 2GB memory on a 128bit bus, using 8 chips, which is a very probable outcome. can be scaled to 4 chips if there are even bigger memory chips down the road.
I agree that bandwith and memory capacity is limited by board complexity. It's another reason I don't believe in a 5 teraflop GPU - you won't be able to feed it
Unless you're performing lots of calculations on a small amount of data. eg. Procedural texture creation in shader would eat gigaflops without needing any read bandwidth.I agree that bandwith and memory capacity is limited by board complexity. It's another reason I don't believe in a 5 teraflop GPU - you won't be able to feed it
you're off by a factor of 2, and you can easily double the number of chips with clamshell mode.
so that gives 2GB memory on a 128bit bus, using 8 chips, which is a very probable outcome. can be scaled to 4 chips if there are even bigger memory chips down the road.
I agree that bandwith and memory capacity is limited by board complexity. It's another reason I don't believe in a 5 teraflop GPU - you won't be able to feed it
Oh ok, so i was right before.
Does't seems there are 4 gbit gddr5 chip down the road..
Even on 28nm m.p, a 5 Tflops GPU with the current Ati architecture would require 300 mm^2 and a tdp of 200 W. That's too much for a console. But if they change the architecture, maybe removing some fixed function units, there would be more space for ALU.
But what about backwards compatibily? I don't think MS will do the same error as the first Xbox,and i don't think they will put Xbox360 hardware inside Xbox 3.
How would a Xenos 2 look?
upnorthsox
Veteran
Clamshell only helps your board layout, you still need 32 bits I and 32 O per bank so for 8 banks (or 2x 4 clamshell) that's a 512 bit bus. That's not out of the realm of possible but IMO it'll be a struggle to get them to go with even 256 bit.
Btw, I believe it's 177 pins per bank for GDDR5 or 1416 pin pads for an 8 bank (or 2x 4 clamshell) setup. That's an awful lot of pin pads. I could be wrong on this though.
Btw, I believe it's 177 pins per bank for GDDR5 or 1416 pin pads for an 8 bank (or 2x 4 clamshell) setup. That's an awful lot of pin pads. I could be wrong on this though.
512MB was a bit dated this gen too. The cost envelope is what iit is. Perhaps a realistic expectation is 2GB very fast working memory and 2+ GB slower memory?
What do you mean by slower memory, sir? Are you talking about something Dram based or something that may be flash based or based upon some 'other' architecture?
If you mean using 2 different memory pool with different speed, i think MS will still go with UMA and maybe Sony will do the same.
. 8 gb would be perfect for next gen,imho, but we will be far from that. I like though the i idea of having some GBs of high-speed flash memory, and streaming content to 2 (i still hope more) GB of GDDR5. Maybe we will see 192bit MC and so 3GB configurations..
Was it really in 2005 ? I think I was running 2 gigs system ram and my video card had 256 megs. I'd say 512 wasn't bad at all.
Considering its 2009 and I just bought my gf 12 gigs of ram for $380 bucks I think 2 gigs is way to low. I'm thinking 4-6 gigs would be the sweet spot. We are talking 2011 so we have another 2 years almost of price drops on ram. I would expect 12 gigs of ram to start aproaching the $200 price point by 2011.
I know its diffrent ram and diffrent prices. But I'm thinking 2 gigs of ram wouldn't even be competitive with just graphics ram on 2011 gpu's . The higher end gpus are launching with 1 gig of ram. I don't even see 512 meg varients of the 5850 and 5870. I think you have to go down to hte radeon 5750 ($130 bucks) to find a 512 meg part. I would think with possibly another 2 generations of graphics parts out even the lower end will be at 1 gig.
As for those saying to put in 16 gigs of cache or less. I don't see the point . Unless your talking about filling that cache each time you load the game. Because right now ms has cache on the hardrive but after playing 2-3 diffrent games the cache has to be flushed and redone when the next game comes in.
So if in the example posted 16 gig with 8 usable as cache you'd have to split that into 4 gigs or even less unlesss you only want to be able to cache 1 game at a time.
32 gigs with say 22 or so avalible for caching could work.
however why would you need to have any avalible for user content ? Just go with memory cards again. USB 3 or some private verison of it with cheap flash ram and your set. I can get a 8 gig flash stick today for $15. I'm sure ms wouldn't mind selling the same 8 gigs of space in 2011 for $30. may sound like a rip off but for game saves it will fit alot of them on it along with content games and what not .
In fact depending on how cheap that becomes they might not even need to go with a hardrive.
Put in 32-64 gigs of fast nand (200mb/s read / write ) then use external nand for game saves and content downloads.
Having 16-32 gig cache capacity per title would hide most of a 6-8x bluray drives problems. 8x bluray is 36MB a second i believe. So it would take awhile to fill the whole cache up however after the inital load the rest can be streamed in during the first few minutes of play. Though I guess it still would take 7 minutes to fill the cache.
But with a hardrive you'd still have the intial load to worry about. At 36MB a 25 gig game could take a while to install about 12 minutes I think.
Would be interesting to see. I still would rather they come up with thier own nand based gaming sticks and just do away with bluray . Get cheap nand and put it in a raid you can get a few hundre megs a second read ability. You wouldn'tn eed a pool of flash in the console or even a hardrive at that point .
16 gig stick transfering at 200MB/s would be its own cache.
Amiga RAM was actually split pools like PC. The internal Chip RAM was slow because it was shared with all the processors, meaning the CPU would ahve to wait if the RAM was bust serving the other chips. Fast RAM was only accessible by the CPU, so it could work full rate while the other chips used the chip RAM. So the Chip RAM was kinda like VRAM, and the Fast RAM like normal system memory.
I'm thinking whatever slow, cheap RAM option is available as a work space, whether DRAM or Flash. It wouldn't be split pools like PS3, but UMA with both CPU+GPU accessing directly the 2GB fast RAM with a separate RAM drive for copying in/out chunks of RAM when you're actually using it.
512MBs was substantially less than a new PC was offering, just as 2GBs would be less than new PCs come with. Going by your example, PS360 launched with 2/9ths your PC's RAM. Extrapolating that, 2GBs next-gen would be 2/9ths of 9GBs, which I expect to be as much as PCs come with. TBH you don't need more than 4GBs AFAICS. Don't even need more than 2GBs unless you're working with big data and don't want to trash the HDD!eastmen said:
Which begs the question...
...what the heck is she wanting with 12 GBs RAM?? Most oflk, me included, don't fill a gig even editing large photos, and if it weren't for OS bloat, 1 GB would be serviceable. I've used this 1GB PC for 5 years now without hitting RAM limits except on some rare occassions, with large images and lots of layers. Are you installing entire DVD games to RAM or something??
- Status
Similar threads
- Locked
