Predict: The Next Generation Console Tech

[Entropy]

The trouble with this plan (which I like, btw) is that there doesn't seem to be much progress in the PowerPC world nowadays.

PPC doesn't have a market to sell small incremental changes into.
All three of the current consoles use PPC processors and that is no coincidence. If IBM has been tasked with providing the next generation CPUs for the next generation consoles, they will design them after the requirements of the customer and the capabilities in terms of density/price/power of the lithography projected to be available for manufacture, regardless of whatever is sold for maximum profit into the desktop PC channel at the time.

That's the whole point of custom solutions, after all.

 

[Johnny_Physics]

Virtualization.

 

[sebbbi]

My wishes for next generation hardware are maybe a bit graphics hardware centered. As GPGPU gets more and more useful, we can offload more and more stuff to the GPU. This means having a low power CPU is not a big problem on a gaming console, as you always have to optimize your game for the platform.

GPU:
- Similar basic features as Xbox 360 GPU. Unified memory, custom vfetch and fast tfetch on vertex shaders allows reading from arbitrary memory addresses, and memexport lets the GPU write to anywhere to the memory. With more customizable EDRAM resolve patterns, and the ability to read the previous render buffer pixel (from EDRAM) this setup would be perfect for GPGPU and would allow many new rendering techniques to be developed.
- SDK should contain an optimized GPGPU library for math operations, sorting, etc.
- Graphics chip should have enough EDRAM to hold 4xMSAA 720p resolution and 2xMSAA 1080p resolution backbuffer + depth buffer without needing to split the scene.
- Some additional SM4 features should be introduced, but the features should be the ones that allow more flexibility to GPGPU purposes instead of better hardware tessellators or geometry shaders. With fast arbitrary memory reading and writing support, the GPU could do most of the same features possible with these graphics centered extra features, but without the extra transistors needed.

CPU:
- I am perfectly happy with the current console CPUs. Xbox 360 CPU seems to be cheaper to manufacture and still packs enough punch (3 HW cores with hyperthreading + vector unit on each). A six hardware thread CPU is better for gaming purposes than a cheap dual core PC chip. However if in future it's more cost efficient to use a cheap quad core PC chip than a custom design (core 2 quad celeron version maybe), then it could be a good canditate. Maybe even something as silly as quad core (8 logical thread hyperthreaded) Atom

RAM:
- Fast (limited sized) EDRAM chip for the GPU. Saves a lot of bandwidth to main memory.
- 1 GB unified memory. As fast as possible, optimized for GPU usage.

 

[eastmen]

A 4870 equivalent should be pretty to fit in even with a tight budget, if my memory is right I think I read that it could be ~170mm² @40nm.
But I see no point in using a 9 cores waternoose/xenon, any job that is likely to spread well on such a chip may be handled better by the GPU (say a more flexible 4870 with the thoughput).

Yea but will it take away from what the gpu is doing ? I want a jump up in A.I and I think a 9-12 core cpu with more cache and other enhancements (fix things developers didn't like or what not) The 3 core verison had small amounts of l1 cache and 1mb of l2 cache correct ? Mabye they can do a 9 core set up with double the L1 cache for each core , 6MB of l2 cache and mabye using L3 cache like in the phenoms and put a large 20MB of L2 cache. A 9 core waternoose with these changes should be tiny of 40nm which is the largest micron size i think they'd ship an xbox next on.

Then mabye a custom gpu based on 40nm would also get used.

I really hope to see an increase in physics. I'd love to see proper behavior of bullets in fps games where the ame would know what your taking cover behind and have the bullets act properly. For instance a cheap wood door in a house would get ripped up and some bullets would get through much faster than a metal door. Or bullets would get through a desk faster than a wall. Stuff like that.


Anyway I'm expecting this

$400 launch price

6-9core waternoose + (bigger L1 , L2 and added L3 - OOE capabilitys) I don't see this as a large gpu. If they go 6 core it should only be slightly larger thantwice the size of the waternoose which will be tiny on 40nm and below. Even a 9core would be small on 40nm


300mm2-400mm2 40nm ati gpu dx 11. With that size i'm expecting some of it for edram or a new type of dram on it. I don't know whats a good size but i'm expecting enough for 4x fsaa. I think the b3d article on xenos said 28.1mb is enough for 4x fsaa + 32bit z and fp10 at 720p . So Mabye we will see 30mb though I expect it to use some of the newer fsaa modes. Also I'm not sure if I understand this right but the xenos used no compresion of the frame buffer unlike modern gpus ? If so perhaps they can add that back in to fit a larger buffer into the 30mb of space

4gigs of ram though i'm hoping for 8

250 gig 2.5 inch drive to 500 gig 2.5inch drive depending on cost

32 gigs of flash inside the console for caching.

I also expect a new scaler chip to replace the one in the xbox. Its one of the great things about the system.


I think ms has noticed the trend this gen where pc games are using the current consoles as the base for their games. MS will use the next xbox to drive home the newest dx verison. Be it dx 11 , 11.1 or even 12. This way developers already have a base to move software on and hopefully it will allow for a faster adoption than dx 10 had since it was largely passed over for dx 9 and xbox 360.


I'd also like to add that amd might want to get fully into the next xbox . They might try and push whatever cpu that have avaliable at really good deals for ms. If they do I see a 6 core verison of whatever comes after the phenom 2. Though I think ms should stick with ibm on this

 

[Silenti]

A 6-9 Core OOE CPU
A 300-400^2mm DX 11+ GPU with 30mb of smart RAM
8GB of RAM
250-500GB 2.5 inch hdd
+32GB of built in Flash Ram
for 400$

Um. If the hdd is optional and the RAM is 4GB. Hell, I'm not even that optimistic. Look at the price tag of all that. It would have to be 2012. At least. I cannot see how that would be possible. Not that I wouldn't welcome it, but...


P.S. - finally took the EVE plunge. Damned glad i did. It is going to take me weeks just to figure everything out!

 

[kyetech]

A 6-9 Core OOE CPU
A 300-400^2mm DX 11+ GPU with 30mb of smart RAM
8GB of RAM
250-500GB 2.5 inch hdd
+32GB of built in Flash Ram
for 400$

Um. If the hdd is optional and the RAM is 4GB. Hell, I'm not even that optimistic. Look at the price tag of all that. It would have to be 2012. At least. I cannot see how that would be possible. Not that I wouldn't welcome it, but...


P.S. - finally took the EVE plunge. Damned glad i did. It is going to take me weeks just to figure everything out!

The only thing I see being doable there is the built in flash. Allow me to adjust it to something reasonable

A 4 Core OOE CPU
A 200-250^2mm DX 11+ GPU.
4GB of RAM
200GB 2.5 inch hdd
+32GB of built in Flash Ram
for 299$

 

[fehu]

previously on a closed thread...


Nextgen spin off: ssd internal drive (and not as a media)

I want to talk about this particular aspect of the next generation of consoles.
Considering that a mass storage is needed, for marketing o real use, and that the instalation/caching will be needed due to inevitably slow storage media, what are the real pros and cons of the inclusion of an ssd?

It must not be the fastest and the biggest, for pricing reasons, but good enought to be faster than a regular hdd.
Think about an api that let you cache the data in a fast large (8GB?) and relatively cheap memory from the BR-Next or from the installation in the regular hdd.

Will be redundant? Or a natural step between hdd/mediadisc and the memory?

 

[eastmen]

The only thing I see being doable there is the built in flash. Allow me to adjust it to something reasonable

A 4 Core OOE CPU
A 200-250^2mm DX 11+ GPU.
4GB of RAM
200GB 2.5 inch hdd
+32GB of built in Flash Ram
for 299$

meh. Well 16gb sd cards is only $20 today. In 2010 i'm sure 32 gigs will be dirt cheap.

The gpu is also doable. We already have gpus that size on 40nm today. In 2010 it shouldn't that difficult or expensive to build it. If it does have edram on it or smart ram (whatever that is ) they can do a seprate daughter die like they use on the xenos.

Current consoles already have 120 gig and 160 gig drives in them so going with 200 gig drive is do able, of course if you hvae 32 gigs built in you may not need to have a drive included in the system.

 

[eastmen]

previously on a closed thread...


Nextgen spin off: ssd internal drive (and not as a media)

I want to talk about this particular aspect of the next generation of consoles.
Considering that a mass storage is needed, for marketing o real use, and that the instalation/caching will be needed due to inevitably slow storage media, what are the real pros and cons of the inclusion of an ssd?

It must not be the fastest and the biggest, for pricing reasons, but good enought to be faster than a regular hdd.
Think about an api that let you cache the data in a fast large (8GB?) and relatively cheap memory from the BR-Next or from the installation in the regular hdd.

Will be redundant? Or a natural step between hdd/mediadisc and the memory?

I think 64 gig ssds are now in the $100 price point
http://www.newegg.com/Product/Product.aspx?Item=N82E16820609331

32 gig sata 2 150MB/s read 100MB/sec write.

I don't really see much of an advantage here except energy costs . Unless these become really cheap it will just be easier to go with a traditional hardrives. Many people just want size an don't care about speed. With an ssd your going to have to go the ms route and only allow your own product on it. Looking at newegg ssd drive are all over the place performance wise.

I think putting the flash directly on the motherboard would be your best bet. Put in a small array of even standard class 6 sdhc chips and you can get high speeds out of it. Each console revision you can add a bigger amount to the console. As long as you keep the partion for installion or caching the same size. So if your first console has 32 gigs of flash you would either do 4 8 gig cache installs or 2 16 gig flash installs. So the system can cache 2 games at once or 4 games.

Or mabye you can go the route i was thinking in the other thread , although not per game but for the console. Create your own custom enclosure and use standard flash in a raid array. SD cards that i've seen can go up to 30MB/s but the spec goes much higher it all depends on clock speeds. But at 30MB/s in a raid array you can put 8 16 gigers in there for a 128 gig drive with write speeds of 240MB/s. An enclosure for such a set up could be tinny. You could actually have two slots for these bad boys in the back of the console. I would also keep the standard hardrive though for dlc and other things.

 

[Squilliam]

Would this work people?

My principles here.

Simple, fast, reliable - choose all 3

Intel indicates off chip bandwidth is expensive

Profitable hardware in the vein of Nintendo, but fewer royalties initially on disks

Including both a HDD and Optical drive in your SKUs will keep the price of the hardware needlessly high or force you to take a big cut

Solid state disks will allow the "Arcade SKU full online capabilities and let you patch your games as well, without a HDD

Im a big fan of Microsoft using Intels LRB architecture and I feel that Intel themselves would be better served by a console (And thus sweeten the deal) if LRB was used for both compute as well as graphics rendering.

CPU Architecure.

4 Core Sandy bridge + 80 Core LRB 2 processor sharing a ring bus, produced on the 32nm process node with an integrated pool of Intel T-ram to act as a framebuffer for the LRB CPU cores. Large, hot, but easiler to cool one chip rather than two.

3*4-8 Gbit ram chips (1.5-3GB total) sitting on a 192 bit bus, probably DDR4 specification and likely yielding over 100GB/Sec bandwidth to the CPU and simplifying the overall console, making it cheaper to manufacture and shrink costs over time.

Lastly a 1GB/Sec connection between the south bridge to a SSD interface, I believe they should use a simple, fast interface and let the media scale with time to give them more bandwidth and thus extend the lifetime of the console somewhat to give subscription services such as Xbox Live time to mature and yield massive revenues.

It probably sounds nuts, but that architecture makes the most sense to me as the idea of losing money of consoles indefinately or launching at a too high price point is likely unappealing for Microsoft, but they still will likely wish to differentiate themselves from Nintendo with a high performance console.

Starting Price -> $250/300 and break even for the Arcade SKU.

 

[DuckThor Evil]

The gpu is also doable. We already have gpus that size on 40nm today. In 2010 it shouldn't that difficult or expensive to build it.

This doesn't make too much sense. Why would a 300-400mm2 chip be any less expensive in 2010. It's still 300-400mm2 of silicon isn't it.

 

[3dilettante]

The wattage numbers on the more ambitious feature sets would be very, very high.

The PSU and cooling solution would have to scale to match, and their expense and size do not scale with process node.

 

[function]

I can't see Sony or MS wanting to draw any more power than they did this time round. In fact, I expect they'd like to draw less.

Part of the games console's steady rise in processing power has relied on power consumption going up and up. We now have the PS3, which is massive on a scale even the original Xbox is jealous of, and the 360 which was prone to overheating, had/has noisy fans and has a power brick as big as a Wii.

 

[AlNom]

Part of the games console's steady rise in processing power has relied on power consumption going up and up. We now have the PS3, which is massive on a scale even the original Xbox is jealous of, and the 360 which was prone to overheating, had/has noisy fans and has a power brick as big as a Wii.

And that's after removing OoOE...

 

[eastmen]

This doesn't make too much sense. Why would a 300-400mm2 chip be any less expensive in 2010. It's still 300-400mm2 of silicon isn't it.

Two more years of maturing the process. Also they may not be on 40nm in 2010 they could very well be 32nm chips or they may be right around the corner from 32nm.

he wattage numbers on the more ambitious feature sets would be very, very high.

The PSU and cooling solution would have to scale to match, and their expense and size do not scale with process node.

Sure they do. each time the chips would enter a process revision the power and cooling required would decrease. Its how its been doing this gen so far. Its how its allways worked.

The thing is if we've been on 55nm and 40nm in 2008 by 2010/11 32nm will be right around the corner. We may only be talking about the launch year or launch months on the 40nm process.

 

[Carl B]

Two more years of maturing the process. Also they may not be on 40nm in 2010 they could very well be 32nm chips or they may be right around the corner from 32nm.

Die size is die size; maybe you should be talking in transistor terms instead if you want a metric that benefits fundamentally from shrinks.

 

[eastmen]

Die size is die size; maybe you should be talking in transistor terms instead if you want a metric that benefits fundamentally from shrinks.

your right , but its hard to know transistor counts.


How about 700m tranistors for the 9 core waternoose with more cache and enhancments

1.5b tranistors for the gpu + edram. Perhaps even zram as it supposed to be smaller than edram and faster. I don't know how edram and zram would affect costs and sizes or how much you'd really need. I was thinking 30mb. But perhaps they will want to go with 720p with 8xfsaa and fp16 instead of fp10. I don't know how much you'd need. 60Mb ?

 

[Blazkowicz]

4 Core Sandy bridge + 80 Core LRB 2 processor sharing a ring bus, produced on the 32nm process node with an integrated pool of Intel T-ram to act as a framebuffer for the LRB CPU cores. Large, hot, but easiler to cool one chip rather than two.

very large and power hungry, I would give it maybe 3 + 48 cores and have the framebuffer on the external memory.

Lastly a 1GB/Sec connection between the south bridge to a SSD interface, I believe they should use a simple, fast interface and let the media scale with time to give them more bandwidth and thus extend the lifetime of the console somewhat to give subscription services such as Xbox Live time to mature and yield massive revenues.

they could simply use 600MB/s SATA. Have gig ethernet, USB 3.0, 802.11n on the console.

 

[Squilliam]

very large and power hungry, I would give it maybe 3 + 48 cores and have the framebuffer on the external memory.




they could simply use 600MB/s SATA. Have gig ethernet, USB 3.0, 802.11n on the console.

By 2011 Intel is expected to start transitioning to the 22nm process. So following the same trend of a whopping big chip 32 in the first revision and then down to something a lot more manageable in the next iterations. The early adopters/tech heads won't be too put off by a noisier console. Unfortunately im not sure precisely how big each core will be, so its difficult to judge.

Btw does anyone know the aproximate expected size per core for LRB, Nehalem, Sandy Bridge, Power PC etc at 32nm?

Oh yeah, Sata would do just fine then. I thought it was limited to 200MB/Sec, I was mistaken.

 

[Carl B]

The problem is that things are opaque from a fabrication standpoint beyond 22nm. And unless they clarify in the next year or two, the console makers may become very conservative on their die budgets in an atmosphere of uncertain silicon cost reductions moving forward with the next gen.