By using this site, you agree to our Privacy Policy and our Terms of Use. Close
jake_the_fake1 said:
curl-6 said:
jake_the_fake1 said:
Cerny makes an example of Edram having a potential bandwidth of 1TB with a small pool of memory paired with a much larger slower pool of memory, and how simply having one pool of high bandwidth memory is the better solution because it removes hurdles which developers have to jump through making their lives easier, hence their decision to with with 1 large and fast pool of memory, But you take this as Cerny stating that the WiiU has 1TB of bandwidth?.....wow....just wow.

For reference, the 360 has 10MB of Edram, it only has a bandwidth of 32GB/s from the GPU to the Edram, but internally the Edram has a bandwidth of 256GB/s to the ROPS.

Currently there is no definitive information about the bandwidth of the Edram in the WiiU, there are only educated guesses people can make, but if we take the whole WiiU design into consideration, one can easily see that although the WiiU has an efficient and somewhat customized architecture, the parts themselves aren't aren't top end, from an old GPU architecture, an even older CPU architecture, to the very slow DDR3 ram used. From this, one can make the assumption that the Edram bandwidth will not be on the top end of the scale rather it may be on the lower end of the scale, perhaps even lower than what is found on the 360, I say this because Edram is very expensive at high capacity and high speeds and the WiiU has 32MB of it, compromises would have to be made to balance the system and speed would most likely be the first thing cut. Why put in Edram with a bandwidth of 1TB/s if the current WiiU GPU will never be able to take advantage of it, kind of a waste. For reference the Nvidia Titan, only has a bandwidth of 288GB/s.

The point is, (technically uninformed) people ignore Wii U's eDRAM, point to its main memory and say, "oh look, it's so slow" without realizing the advantages eDRAM gives. This video nicely demonstrates how such an advantage works.

The whole point of Edram is to compensate for the low bandwidth of the main memory because graphics rendering is bandwidth intensive and low bandwidth DDR3 would cripple it. There is also the business equation to it, spend a more money on small pool of fast ram, then put in larger, slower, cheaper ram with the overall cost being cheaper than just using 1 large pool of high speed ram, the compromise there is the overhead you give developers.

In the end its two solutions for the one problem as described in the video, Cerny's approach was to reduce developer overhead as much as possible which led to him favouring 1 large pool of high speed ram. The thinking I guess is that you spend money now on ram, but reduce the difficulty and thus reduce the cost or at the very least maintain the cost of developing rather than have it exponentially grow again.

 

PS4's approach was probably chosen for being easier to develop for, yeah.

I'm not arguing Wii U's approach is better, I'm just saying that by explaining the value of the eDRAM approach Cerny does show the inaccuracy of claims that point to Wii U's main RAM bandwdith and judge the system on that while pretending its 32MB eDRAM doesn't exist.