Interesting.
If the numbers are good, it could be a decent performer (550-600mm2 @32nm)
Predict: The Next Generation Console Tech
- Thread starter Acert93
- Start date
- Status
Interesting.
If the numbers are good, it could be a decent performer (550-600mm2 @32nm)
*Rough figures*
RSX = 7800GTX
8800GTX ( 2x jump in raw power )
9800GX2 ( 2x jump in raw power )
GTX 295 ( 2x jump in raw power )
GTX 590 ( 2x jump in raw power )
Now I know that all the above configurations are not flat out 2x each other but they're as good as.
Now the recently released AMD 7970 with a slight over clock can match a GTX 590 but we all know that with regards to power consumption and heat generation putting something like a 7970 into a console is never going to happen.
The I got to thinking, The 7970 flat out beats the older AMD 5000 series dual GPU card, The AMD 5970. Now in regards to the newer more power efficient 7000 series what card could offer that kind of performance while being cool and easy on the old electricity?
In my eyes it has to be a 7850 or a 7870, Either of those cards would offer a good 6.5x times the jump in raw power while still being cool and collective,
RSX = 7800GTX
8800GTX ( 2x jump in raw power )
9800GX2 ( 2x jump in raw power )
GTX 295 ( 2x jump in raw power )
GTX 590 ( 2x jump in raw power )
Now I know that all the above configurations are not flat out 2x each other but they're as good as.
Now the recently released AMD 7970 with a slight over clock can match a GTX 590 but we all know that with regards to power consumption and heat generation putting something like a 7970 into a console is never going to happen.
The I got to thinking, The 7970 flat out beats the older AMD 5000 series dual GPU card, The AMD 5970. Now in regards to the newer more power efficient 7000 series what card could offer that kind of performance while being cool and easy on the old electricity?
In my eyes it has to be a 7850 or a 7870, Either of those cards would offer a good 6.5x times the jump in raw power while still being cool and collective,
Last edited by a moderator:
Wow if there is truth to his claims. Lots of thing to be assumed form his statement.
256 bits bus
500mm2 chip it's even bigger than what I was considering for the sake of fitting a 256 bits bus comfortably even after a shrink. Produced at IBM and GF, that lefts eDRAM out of the picture, I think that if true a 256bits bus is given along with average GDDR5. A 256 bit should fit even after a shrink even 2.
2GB of RAM. That's as much RAM you can fit on a 256 bits bus while keeping the number of memory chips under control (8). Basically were looking at ~120GB/s of bandwidth.
Trinity could be used as a ref to determine the amount of CPUs and GPUs cores one could fit in to a big, gtx480 size, chip.
I would bet on 4 OoO 2ways SMT CPU cores backed with potent 256bits wide SIMD. I hope that they would be IBM take on Intel last APUs and that they will have mimicked memory sub system and cache hierarchy. I could see them clocked south of 3GHz, 2.8GHz could be the sweet spot . As far as performances are concern I think that without considering the benefit of SMT K8 level of perf per cycle is what we can expect. I expect that power efficiency to be higher than in K8. Thanks to SMT it will perform really well.
I would appreciate an extra core from IBM embedded line for "house keeping duties" and possibly some accelerators for whatever MS thinks is fit (security and networks cone to mind).
For the GPU I would bet around ~1200sp clocked reasonably.
256 bits bus
500mm2 chip it's even bigger than what I was considering for the sake of fitting a 256 bits bus comfortably even after a shrink. Produced at IBM and GF, that lefts eDRAM out of the picture, I think that if true a 256bits bus is given along with average GDDR5. A 256 bit should fit even after a shrink even 2.
2GB of RAM. That's as much RAM you can fit on a 256 bits bus while keeping the number of memory chips under control (8). Basically were looking at ~120GB/s of bandwidth.
Trinity could be used as a ref to determine the amount of CPUs and GPUs cores one could fit in to a big, gtx480 size, chip.
I would bet on 4 OoO 2ways SMT CPU cores backed with potent 256bits wide SIMD. I hope that they would be IBM take on Intel last APUs and that they will have mimicked memory sub system and cache hierarchy. I could see them clocked south of 3GHz, 2.8GHz could be the sweet spot . As far as performances are concern I think that without considering the benefit of SMT K8 level of perf per cycle is what we can expect. I expect that power efficiency to be higher than in K8. Thanks to SMT it will perform really well.
I would appreciate an extra core from IBM embedded line for "house keeping duties" and possibly some accelerators for whatever MS thinks is fit (security and networks cone to mind).
For the GPU I would bet around ~1200sp clocked reasonably.
Last edited by a moderator:
DuckThor Evil
Legend
Charlie is using the die size range just to show that even in a "worst" case 10k wafers is a lot of chips. He does not claim or even speculate that the die actually is 550-600mm2.
Last edited by a moderator:
Please take this discussion to this existing RAM thread (which I suggest you read first as it's all said there anyhow
)link said:
Well, it seems pretty concrete that IBM is indeed making some new Xbox SOC with EDRAM, but is it 720 or 360?
If there was some confirmation on die size that would quickly narrow the answer!
My bad english issue I had to read the thing multiple times because it was not that clear at first... all this to still misunderstand
let's have another coffee :|Yes that 's a lot of chip in any case, 200.000 that sounds like a hell lot of devs units to me. Ms may want to test the hell out of it... lol
Last edited by a moderator:
Wouldn't the EDRAM in 720 though have to be huge though and even bigger if they decide to use the tiling method that 360 employs?
With most games moving to deffered rendering the frame buffer requirements are huge compared to forward renderers.
And in my opinion the EDRAM on 360 was not that successful and some times caused more problems then it solved with games having to run at reduce resolution to fit into the 10mb...etc..etc..
With most games moving to deffered rendering the frame buffer requirements are huge compared to forward renderers.
And in my opinion the EDRAM on 360 was not that successful and some times caused more problems then it solved with games having to run at reduce resolution to fit into the 10mb...etc..etc..
OK GF can produce Edram. THe PR is unclear about many thing as they use plurals "chips".
I don't see what "networking" has to do with "gaming and graphic".
I think that there are multiple manufacturers involved and MS could clearly be one of those.
But back to edram I don't think that they will use (if they use it) as in the 360 ie a big pool of EDRAM. EDRAM cell could be a substitute for SRAM cell in the L3 for example, maybe in the GPU as at low clock speed the disadvantage of EDRAM versus SRAM are less relevant (if at all).
Early on Charlie claimed that IBM had the high hand in designing the chip It would be fun if they implemented the CPU and the GPU on the SoC using as much EDRAM cell as possible. It offers nice opportunity to raise even higher the transistor density even without going for a GT480 sized chip ( I read again and I clearly misunderstood his statement on the matter).
I'm still convinced that a 256bits wide bus is the way to go along with a +300 mm2 chip ;-)
east of eastside
Newcomer
Calling 360 mini.
I would not say it was not successful as it more than once responsible for the difference in perf with the PS3. Anyway I'm not sure about how MS could have done things significantly better with the same silicon budget.
Still I agree it's time to pass on EDRAM used as an "underfunctional V-RAM".
Last edited by a moderator:
Also with MLAA and FXAA being pretty much touted by most people as the way forward for anti-aliasing over traditional methods it kind of makes EDRAM even more redundant as it was partly there in 360 to alleviate the bandwidth cost of using MSAA.
MLAA and FXAA have a small cost compared to MSAA in pretty much every situation?
MLAA and FXAA have a small cost compared to MSAA in pretty much every situation?
Yes, but they also have tendency to blur things often, and have problems detecting many edges
Right... a lower amount of pixels has absolutely no impact on the amount of work that can be done (shading per pixel or pixel throughput).
Do you honestly think the games that are sub-HD could be done with the same graphical fidelity or framerates at a higher resolution? Come on...Right... a lower amount of pixels has absolutely no impact on the amount of work that can be done (shading per pixel or pixel throughput).
You're totally taking it out of context, Or maybe I just worded it wrong
Wasn't PGR3 forced to run at a lower resolution to fit into EDRAM without tiling? As were a few other games?
Granted lower resolution meens more free time to do more complex effects and shaders and in general that's why most games do run at Sub-HD but I sure there's the odd few that have had to do it because of the 10mb EDRAM and lack of tiling support in the game engine?
That is a really good point. What proof do we really have that this isn't just a standard hardware revision taken overboard by the rumor mill? Just saying that you have multiple sources isn't good enough if your multiple sources all got the information in the same place.
Right... a lower amount of pixels has absolutely no impact on the amount of work that can be done (shading per pixel or pixel throughput).
True, but the overhead for tiling in some cases I'm sure did prevent what would otherwise not have an issue running at a higher resolution than what would fit in the 10MB buffer.
Especially those trying to fit a 60FPS budget.
- Status
Similar threads
- Locked
