4 replies to this topic

[Mike12Nite]

The edram is pretty much elbow to elbow the the GPU logic. Since there hasnt been a configuration like this before, I'm curious what having such a low latency connection to alreafy fast, high bandwidth memory will allow developers to do. The edram speed must be blindingly fast, and the CPU also has a very low latency connection to it. What will be the advantage of this in games?

[NintendoReport]

Thought I would bump one of the many threads here about Wii U Power and more specifically edram.. 

 

I read the Digital Foundary analysis of the Titan Fall 360 port and the devs did a fairly impressive job making that game look decent and run well all things considered.

 

Here is a quote regarding how the devs used the edram within the 360, and full article here http://www.eurogamer...rmance-analysis

 

 

Let's begin with rendering resolution. The Xbox One version ships at 1408x792 with 2x multi-sampling anti-aliasing - a sub-native presentation for a console aimed at the era of the 1080p display. There are few surprises with the Xbox 360 version, which adopts a strategy similar to the Call of Duty titles on older Microsoft hardware. There's a 1040x600 native resolution here, backed up by 2x MSAA. It's a set-up that allows Bluepoint to cram the framebuffer into the 360's 10MB of eDRAM, effectively allowing it to use the hardware anti-aliasing for free with no performance penalty

 

 

I am not entirely sure why I am posting this, but thought spec junkies would consider it a good read and maybe can expand upon the different ways Nintendo and it's devs might utilize the Wii U edram.

[Azure-Edge]

Awaits 3dude to come in with his tech know how.

[Scumbag]

3dude would do a better job of explaining but I think Wii U can do everything the 360 could when using edram.

360 edram is over 3x less and is on a daughter die so Wii U's 32mb is more efficient as in sending and receiving instructions between CPU/GPU/edram because including it right next to GPU has huge benefits.

[3Dude]

Low Latency. Or very fast performing ram. Not necessarily that the ram is super fast, but it can be accessed so readily and rapidly that its a very high performing part.

There are 3 basic attributes to ram: Capacity, how much it can hold, Bandwidth, how much it can move, and latency, or how long it takes to use it.


Bandwidth is a pretty big buzzword these days, rather often confused for speed or performance. Bandwidth is not speed, high bandwidth ram is not necessarily fast ram, although bandwidth can affect data transfer time... However transfer time is not the only time that contributes to latency.



A lot goes into latency, from how its clocked, to how much electricity pushes the data through the bus to how many transistors need to be refreshed, how thin/thick is the conductor you are pushing through... to, how far does it have to travel.Well, travel time/distance is actually a pretty big deal and adds a lot to latency. Since the speed of electrons at regulated voltages is pretty, well, fixed (in proportion to the power used in the current), how far the information has to travel, from ram to destination is a pretty big deal.


As you can see in the picture, all the bandwidth in the world isnt going to help that long wide pipe compete with that short pipe on the bottom. The information has to travel too far, and even if the pipe was wide enough to transfer all the 'stuff in one tiny sliver, since it has to travel at the same speed (you cant just blast power through your processor to speed things up, everything will fry), but much much farther, it can never practically compete with the much closer distances of the bottom pipe despite having much much higher bandwidth.

Thats the advantadge of the large edram cache being so close to the gpu logic.

The 360's Edram, is basically useless in comparison.