Predict: The Next Generation Console Tech

Status
Not open for further replies.
Does anyone have (roughly) accurate figures on what the transistor count is per MB of IBM eDRAM tech? Power7 is a 567mm2 design at 45nm with 1.2bn transistors. Some die shots here:

http://aceshardware.freeforums.org/power-7-t891-15.html

Seems to be packed quite dense. Fitting ~20MB in a 2bn transistor 22nm console integrated CPU/GPU shouldn't be too difficult assuming it scales near linearly (I'm guessing it won't?). Looks like too big a design win to pass up, but I'd like some accurate figures to work with.

20MB would be enough to fit an entire 1080p framebuffer (assuming no msaa since MLAA and its derivatives will probably be much more popular next generation) and still have a decent amount of space left over to work as a generic CPU L3 cache as well. With MRTs becoming so popular it seems silly to go with much more than that as no amount of L3 cache is ever going to be big enough to fit all your render targets at once, so long as they can each easily be swapped in and out of cache then I think the solution is fine.

Not sure of the die size but its 2bn transistors for 32MB of edram at 45nm. Looking at the die I'd say it's at most 1/3 and probably less of the 567mm2. That'd put it at 189mm2 at 45nm or about 59mm2 at 28nm for 20MB with 1.2bn transistors.
 
Scratch the above transistor count, I read that completely wrong. It's 1.2bn trannies for the entire 8 core Power7, if a 1/3 of that is edram then that would be around 400mln trannies for 32MB or 250mn for 20MB.
 
Scratch the above transistor count, I read that completely wrong. It's 1.2bn trannies for the entire 8 core Power7, if a 1/3 of that is edram then that would be around 400mln trannies for 32MB or 250mn for 20MB.

Yeah, 400m transistors for the eDRAM in Power7 is what I guessed "by eye" as well, I was just wondering if someone had more accurate figures but I guess it doesn't really matter if they do as there's no guarantees you'd be able to pack it as densely at a lower process node or whether having an interface wilth 8 cores (vs. a GPU and CELL) is distorting the figure.

Either way, if we assume 300m as a conservative estimate of how few transistors you'd need for 20MB of shared L3 cache then I think its inclusion becomes a no brainer if you're going to use something like a single ~2bn transistor chip solution @ 22nm.

It'd gave you room left over to fit a revamped CELL with twice as many SPEs and 2 beefier PPEs (something like 2 Power7 cores with a shared 2MB L2 cache would seem appropriate) and still have a >1 billion transistor (likely Nvidia) GPU on chip.With 4ghz SPEs you're looking at lot more compute power (over a half a terabyte for the SPEs alone!) and a GPU that is already much faster RSX before you start considering its access to a big chunk of eDRAM and ~10x the external bandwidth (provided by XDR2/XDR3 over a 128 bit bus). Add all that to a much bigger RAM pool (4x as large) and all the efficiency savings you'd be able to make by having a near zero latency interface between CELL and the GPU and a UMA memory setup and you're starting to look at one hell of a tasty little console.

What's better is that it'd be infinitely cheaper to manufacture at launch than the PS3 was for a number of reasons (BD-ROM drive is super cheap, singe chip solution, only 4 RAM chips, much lower motherboard complexity, no need for a separate chip for BC and a much lower end cooling solution can be used). Considering we're only talking about a doubling of rendering resolution here and graphics on PS3 are already quite close to "enough" for many consumers, I really don't see the justification to go much higher end than this. We're still taking about a fantastic upgrade in graphics here and the removal of most (if not all) of the bottlenecks that held back PS3 development. CELL libraries are already very mature at this point, so developers would be able to hit the ground running especially since the GPU part will be very familiar to anyone following PC developement as well.

Sell it at launch (holidays 2013) in dual SKUs, one at $350 with a 16GB flash chip and one at $400 with a 500GB 2.5" mechanical HDD and Sony should be close to breaking even or possibly even making a nice little profit on each unit sold. Absolutely no chance of suffering massive losses for several years after launch, at the very least and that's despite launching at a pricepoint 1/3 lower than PS3.
 
@ ED-Ram density: Isn't the ED-Ram on 360 100M transistors for both ROPs as well as memory?

I suspect it also has something to do with how it is addressed. If its to be used as a cache for multiple cores it would probably be less dense than if it is used for a single application such as frame-buffer.
 
@ ED-Ram density: Isn't the ED-Ram on 360 100M transistors for both ROPs as well as memory?

I suspect it also has something to do with how it is addressed. If its to be used as a cache for multiple cores it would probably be less dense than if it is used for a single application such as frame-buffer.

The xenos has 4 rops ?

Anyway if 100m tranistors for 10MB of edram you can put a good 50MB in less than 500m transistors space. That be pretty nice for a next gen graphics chip and should allow for no tiling with even 1080p 60fps
 
The xenos has 4 rops ?

Anyway if 100m tranistors for 10MB of edram you can put a good 50MB in less than 500m transistors space. That be pretty nice for a next gen graphics chip and should allow for no tiling with even 1080p 60fps

8 AFAIK. It has 8 ROP units on top of the ED-RAM. I have no idea on how many transistors would constitute a single ROP unit. However if I was to hazard a guess, I would say it would be very small. I think I need to take another look at the ED-RAM diagram to work out a rough proportion.
 
That is bizarre ... Microsoft better not just let people download the free episode from the episodal version from that link, or they will have a bait and switch lawsuit coming.
 
8 AFAIK. It has 8 ROP units on top of the ED-RAM. I have no idea on how many transistors would constitute a single ROP unit. However if I was to hazard a guess, I would say it would be very small. I think I need to take another look at the ED-RAM diagram to work out a rough proportion.

So realitsicly 50megs of edram will be quite cheap next generation. I would think ms + ati would go for it again
 
Well IBM could indeed get their technology into a console after all. This is a rather bullish piece but it does show some promise.

IBM's SOI technology can provide up to a 30 percent chip performance
improvement and 40 percent power reduction, compared to standard bulk silicon
technology. SOI protects the transistors on the chip with a "blanket" of
insulation that reduces electrical leakage, saving power and allowing current
to flow through the circuit more efficiently, improving performance.

IBM has fabricated a test chip with an embedded dynamic random access memory
(eDRAM) technology that features the industry's smallest memory cell, and
offers density, speed and capacity better than conventional on-chip static
random access memory (SRAM) announced in 32nm and 22nm technology, and
comparable to what would be expected of an SRAM produced in 15-nanometer
technology - three technology generations ahead of chips in volume production
today.

IBM's eDRAM cell is twice as dense as any announced 22nm embedded SRAM cell -
including the world's smallest 22-nanometer memory cell announced by IBM in
August 2008 - and up to four times as dense as any comparable 32nm embedded
SRAM in the industry. Higher memory density can lead to chips that are
smaller, more efficient and can process more data, improving system
performance.
http://www.reuters.com/article/idUS123822+18-Sep-2009+PRN20090918

Or course there is always T-Ram:

http://en.wikipedia.org/wiki/T-RAM

On and not to be left out theres some more

http://en.wikipedia.org/wiki/Twin_Transistor_RAM

I think the link says it all.

Anyway. There are options.

However if any fab were to go with ED-Ram or the above they would probably be stuck with an IBM alliance fab of which Global Foundries is the only example due to their using SOI technology.
 
I think Sony needs to go balls out with the PS4 like they did with the PS3... maybe more so, but also leverage the technologies from PS3 that have served them well. If they don't do this I could see them being the next Dreamcast personaly. Yes they may decide a modest upgrade in power is the right way to go so they can bring it in within maybe a 299-399 pricepoint. This I believe is not the right way to go... it's been 4 years from 599, but I think people will accept 499-599 within a year or two, with the cavet that the power and inovation is there to warrant the price... it's also an early adopter fee obvously, and people with pay that price... including me. They will want ps4 to last 10 years again, and for that to happen it will have to be powerfull and inovative.

I'd like to see more than just a revamped and upgraded MOVE at launch... it would be nice for a 720p 3d eye2 to come standard as well... in other words, the best of Kinect made better with augmented 3d(and less lag) along with a new move controller/s.

Music store, movie store, app store, revamped psn from the ground up, go for the gusto. Blow people away.

As for the hardware... Build it around 3D gaming because that is the future.

I'd like to see a couple of power7 cores coupled with a combined 8 beefed up spu's. about 4GB of unified XDR2 ram and a powerful imagination technologies GPU built with 3d in mind.

So that's my vision of what I'd like to see sony do. I don't want a modest increase that will be outdated the day it comes out, I want a ps2 type industry disruptor but with ease of programing in mind.
 
499 is a non starter. I think 399 is the spot.

Should be easy, PS3 was only 600 because of Blu Ray, not a problem next gen.

Personally I think Sony will adopt Shifty's tablet console concept :p

And who is imagination technologies? Anything short of Nvidia or ATI is a non starter if you want to be competitive. Although there's rumors Sony will in fact be using a Toshiba GPU next time around.

Also, somebody said earlier in this thread EU regulators wont look kindly on a power sucking console next gen. If so that applies to all consoles. So a monster PS4 may be a non starter.
 
Well IBM could indeed get their technology into a console after all. This is a rather bullish piece but it does show some promise.


http://www.reuters.com/article/idUS123822+18-Sep-2009+PRN20090918

Or course there is always T-Ram:

http://en.wikipedia.org/wiki/T-RAM

On and not to be left out theres some more

http://en.wikipedia.org/wiki/Twin_Transistor_RAM

I think the link says it all.

Anyway. There are options.

However if any fab were to go with ED-Ram or the above they would probably be stuck with an IBM alliance fab of which Global Foundries is the only example due to their using SOI technology.

Doesn't amd use IBM's SOI tech ? I believe they have a cross liscensing deal.

499 is a non starter. I think 399 is the spot.

Should be easy, PS3 was only 600 because of Blu Ray, not a problem next gen.

Personally I think Sony will adopt Shifty's tablet console concept

And who is imagination technologies? Anything short of Nvidia or ATI is a non starter if you want to be competitive. Although there's rumors Sony will in fact be using a Toshiba GPU next time around.

Also, somebody said earlier in this thread EU regulators wont look kindly on a power sucking console next gen. If so that applies to all consoles. So a monster PS4 may be a non starter.

I think he means power vr / video lodgic or whatever the heck they are called now

I don't think we will see more than $300 for next gen consoles.
 
499 is a non starter. I think 399 is the spot.

My pick is $299 for the cut down model and $399 for the more expensive version.

And who is imagination technologies? Anything short of Nvidia or ATI is a non starter if you want to be competitive. Although there's rumors Sony will in fact be using a Toshiba GPU next time around.

Well they are the premiere mobile graphics maker in the world IIRC. Their claim to fame is producing the Apple iPhone GPU and I believe some of their graphics IP goes into making Intel IGP/CGPU chips.

They are also fanstastic and awesome, because Rys works there. :D

Also, somebody said earlier in this thread EU regulators wont look kindly on a power sucking console next gen. If so that applies to all consoles. So a monster PS4 may be a non starter.

That would be me. :)
 
If IBM's eDRAM is that good, how's about Cell with a large pool of eDRAM? This would help feed more SPEs and, dare I say it, wanting my universal processor, address some issues with texturing problems by caching more. In terms of bandwidth, if price be damned, we could have 500-1000 GB/s memory and 30+ MBs on-die fast RAM, for bandwidth-worries-free processing!
 
Thats a pretty good point of speculation there Shifty. I believe the PS3 was originally intended to have a large pool of fast eDRAM attached to the Cell processor. Obviously it would do some good for the efficiency of the processor or otherwise they wouldn't have prototyped it. I don't think its a bad choice by any means, if they can feed fewer SPEs more efficienty with less data going over the bus they'd have better performance/watt, correct?
 
Also, somebody said earlier in this thread EU regulators wont look kindly on a power sucking console next gen. If so that applies to all consoles. So a monster PS4 may be a non starter.

That would be me. :)

I think there have been some regulations conserning stand by power usage, but other than that, there will certainly not be a regulation in the near term that would ban a game console drawing something around 200w, or even more to be honest...
 
Thats a pretty good point of speculation there Shifty. I believe the PS3 was originally intended to have a large pool of fast eDRAM attached to the Cell processor. Obviously it would do some good for the efficiency of the processor or otherwise they wouldn't have prototyped it. I don't think its a bad choice by any means, if they can feed fewer SPEs more efficienty with less data going over the bus they'd have better performance/watt, correct?
What would it bring? (EDram)
It's like the size of LS nAo for example stated multiple times that it's not in the top list of thing that would make SPUs better. SPUs are about streaming and STI made sure they had "plenty" of bandwidth (~25GB/s is/was a lot in 2005).
 
Status
Not open for further replies.
Back
Top