Digital Foundry Article Technical Discussion Archive [2013]

Nah... It may mean they haven't finalized the numbers, or they don't want to get into the discussion. ^_^

The Chinese dev (you posted) leaked that a game has more than 5GB (i.e., 5GB + system heap). Then again, that was a while ago. Also unofficial source.

 

The numbers weren't confirmed in any capacity. I don't know much about teardowns or what transpired so that we know about Wii U's numbers, but we're going to have to wait for that for PS4 i suppose, because Sony isn't talking.

 

That's what I think.

 

Digital Foundry: the complete Xbox One architects interview
http://www.eurogamer.net/articles/digitalfoundry-the-complete-xbox-one-interview

 

Some parts of that article/interview were very interesting and some parts downright laughable (FUD, PR, Conjecture, etc...). The parts dealing with XB1 architecture as a whole and Jaguar CPU, was really good... something MS should be proud of.

 

Maybe you could detail the bits that are misleading or incorrect?

 

I would say anytime they speak about the details of the PS4.

 

Yes that... lots of Conjecture/PR/FUD dealing with that area. Some of the double talk was annoying at times, but I'm not going to harp on it.... everyone has a point of view, regardless if we like it or not. Anyhow, the article serves it purpose on letting you in further on XB1 design and philosophy for it.

 

Geez... this is really not the forum for random commentary spinning this as the engineers being pr spinners... Christ just let it go man and address the actual Data contained in the article

 

I don't like this interview very much either... They never go deep enough to justify their design goals... It read almost like: We designed this console to be very well balanced, without anything in particular being the bottleneck, and made sure you can get the most out of each component, but then goes to give no answer on why this is so efficient, nor why other designs (read Ps4) are not.

It made sound even like they actually lucked out on the esram being more than they expected, otherwise the system would not be as balanced as they wanted too...

 

Spinning the spin, I see what you did.

Anyhow, I'm not debating XB1 performance numbers or the equations MS used to arrive at them. Because that's fairly pointless, especially when you're stacking the deck for PR needs on offsetting internet noise. If it was "all about the games" as they pleaded, then the PR would speak for itself - through the games.

As I stated before, the article serves it purpose... and was quite interesting to see their approach and methods.

 

What I see from all of this is that Microsoft meant to have the Kinect included on every box since the beginning, the Xbox One development team had a budget taking that into account and I think they did an excellent job with it, regardless of the apparent differences between the PS4 XOne hardware, I think that for most people the differences will be even less noticeable than the PS360 launch games. The problem with all of this is that Microsoft had to invest in a Kinect AAA game to showcase what can be done, I know that it will come at some point, but right now most hardcore gamers don’t see it as a plus.

 

The problem I have with all this "balanced" talk is that it suggest AMD has no idea how to balance it's own GPU's.

If the PS4 has "too much" CU power for a balanced system then what does that say for the R290X? Or for AMD who designed it as their flagship product?

The R9 290X compares to the PS4 as follows based on the currently rumored specs and a 1050Mhz core clock:

ROP Throughput: 1.8x
CU Throughput: 3.21x
Triangle Setup: 2.63
Memory Bandwidth: 1.84 - 2.1x (depending on whether you include system memory)
CPU performance (estimated on high end quad Haswell): 2.5x

So clearly the has even more CU power relative to all other parts of the system and so why would AMD do this if they know the CU's will be bottlnecked by the rest of the system? Why not spend those transistors on fewer CU's and beef up the rest of the GPU instead?

 

In a way they did. The 290 has fewer CU's that do more within them. That architectural change most likely imparts the better functionality and bandwidth utilization. Isn't the 290 only 4 CU's? Maybe I misinterpreted the leaked specs.

 

Why was my post removed? Did I mess up again?

Should I have quoted the engineers answers to support what I was saying?

I can't write it again, I forgot it and it was a masterpiece, it will be lost in the internet aether forever. Farewell, post.

 

290xt = 48 CU vs 8970 = 32 CU.

 

The R9-209x I believe has 44CUs (2816 SP), 64 ROPs and 176 TMUs. So, pjb comment still stands.

 

The way I see it, if what Sony And MS said about balance is correct then AMD would have been much better off keeping CU's at 32, but also bumping up ROPs to 64 and the memory interface to allow for 7Ghz GDDR5. That would have placed the entire GPU (and CPU) at around 2.6x the PS4 with CU's being only around 2.3x. Or in other words very close to the PS4's "sweet spot" balance of 14CU's.

I'm sure that would have worked out transistor wise (it probably would have been a net gain). So that fact that AMD didn't do that is pretty telling.

 

Perhaps they just mean "balanced" w.r.t. the computing budget they put into the GPU. The more budget you put in, the higher the GPU should hit

Again it's a pretty limited/simplistic way to quantify the GPU. The actual execution and run-time policies can be (are) more complex.

In my view, programmers and artists still make the most differences to the outcome.

 

4 CU each with 11 SIMD for 44 SIMD total.

Some others say 11 CU with total 44 SIMD, but I think they are misinterpreting the 11 SIMD per CU and reorting it as 11 CU.


EDIT: Actually, I'm more likely to believe 11 CU's with 4 SIMD per as that makes more sense from a component point of view.

Read more: http://wccftech.com/amd-radeon-r9-2...del-pictured-hot-cooler-design/#ixzz2gsNvoWVA