Wii U hardware discussion and investigation *rename

Status
Not open for further replies.
I'm thinking he's playing funny games about the "GPRs". everyone would just say "registers" for a start.

Thanks, so it gives 16KB in registers for each 5D ALU, or rather 256KB per pack of 80 shaders. This is huge :), this must be why the AMD architectures before GCN worked so well while looking quite more simple than the nvidia designs. AMD just seemed to pack blocks of SP togethers, eventually reaching 1600, while nvidia had their more complicated SM and floorplan already.

I've found a similar analysis for RV770 and it has the same 16K per 5D unit :
http://www.bit-tech.net/hardware/graphics/2008/09/02/ati-radeon-4850-4870-architecture-review/8

So, there was no difference between the original R700 series and its lone 40nm revision.

I quickly hunted for the 6870 GPU (Barts) then the 5870.
about the 5870 :
http://www.hardware.fr/articles/770-3/rv870-ou-cypress.html
Chaque SIMD dispose de 16384 registres de 128 bits, comme les générations précédentes, de manière à supporter un nombre important de threads et ainsi masquer les latences autant que possible.
bingo! we have 256KB per unit of 80 shaders.

about the 6870 : (it's all french and I explain below)
http://www.hardware.fr/articles/804-2/barts-tessellation.html
They say 6870 is basically a slightly cut down 5870 (a few less shader units and no FP64) with a few minor changes (texturing units, tesselation buffers, outputs).

contrairement à Cayman, l’architecture de Barts reste globalement inchangée par rapport à celle des Radeon HD 5000
"Contrarily to Cayman, Bart's architecture remains globally unchanged from that of the Radeon HD 5000s"

I take this and say I'm pretty confident the register files didn't move.
So, at least every Radeon with a model number bigger or equal than 4350 and lower or equal than 6870 has the same amount of registers per set of units :p
 
at first I wanted to complain about not wanting to do that search, but it's fun
I can't come now, someone is wrong on the internet.
 
So according to Matt the GPU has more general purpose resisters (GPRs) than the R700 series.

Interesting.
 
Last edited by a moderator:
Anyone want to give their estimates for number of transistors, both for the GPU and, assuming Wii U has a SoC design, for the SoC? This would have to include the edram.

For reference, the Vejle chip in Xbox 360 Slim has 372m.
 
The thing is with registers is that if you don't use them, why put them in the hardware? A multiport register file can be a significant contributor to the area and leakage power.

The good news is that by now, there are probably many benchmarks and code profiles from the thousands of games that allow AMD/Nintendo to trim down the register file froma R700 sereies to only the amount that is needed.

The other thing that helps with fewer registers is the EDRAM. By having a fast, low latency RAM you can reduce the number of registers without hurting performance. EDRAM is may actually end up being more useful then more registers and lower power.
 
oh noes, please do not reduce our register budget as if it were fire stations budget.
It should be just unchanged from R700. The thing we've read at the top of the page was made up.
 
Googling found the R740 technical evaluation on this very site. According to the article that chip has a total of 2MBs of GPRs shared by 640 shader units. I don't know enough to tell you the advantages of an increase in the ratio of GPRs per shader unit is. I'm guessing this is BG's secret sauce, though. And yet, even the leaker Matt admits this is still sub-Radeon 6XXX level, which is to say any improvement is unlikely to be game changing in terms of GPGPU (let alone enough to propell the part into the 600GFlop range).

Why do you have secret sauce? And why are you talking in third person?
 
is there some compression issue? McHuj was talking about reducing them.

/edit : comprehension!
 
Last edited by a moderator:
GPU's dynamically allocate registers for running shaders, the number of GPR's dictates how many threads run in parallel and as a result more registers means more threads and generally equates to less stalls on non local accesses, things like texture reads.
 
Sadly I can't find an article going through this properly but I would bet that those GRP refers to the sratch pad memory include in the SIMD.
As I said I could not find an in depth review of the Radeon ( actually may be available here...) but if memory serves right there are two type of "scratch pd memory in those architecture, the local data share and the global data shared. The global data share is (if memory serves right... again) used to share data between the SIMD. I would assume that it's just a bunch of register files.
MSFT enforces a minimum amount of these to be compliant with their direct X specifications.

So as I read it (if true) it means that the GPU GPU is R700 based with more GDS (say the same amount as in HD 6xxx to meet requirement of direct x11).
 
Last edited by a moderator:
NO and it doesnt matter if its 700 glfop vs 1.5gflops. It such a massive difference with everything it doesnt matter. Now we have report that the cpu is weaker than the ps360. Then you have the massive ram differences.....etc.

What is likely is devs will use ps360 engines to make wiiu games. While the main games are built for engines that can run on high end hardware. So the maddens and COD will use ps360 engines that would be perfect for the wiiu.



Where did this come from? Someone is trolling wiiu fans hard! lol
I have a feeling that the guy is telling the truth. Don't know why this is... ?

It's not typically that I lose myself,, or maybe I do. But if AMD are trolling him then... Well... that's just awful. That's about all I can say of that. If I were AMD I wouldn't be happy with an employee being manipulative and making fun of a person.

I guess AMD are just replying a polite question this guy made, although AMD USA replying to an european person just seems a bit odd to me, the rest looks and sounds legitimate.

I would be certainly more suspicious if Angeline Jolie suddenly wrote me an email, but I still believe in great companies like ATi -AMD- having some manners and care about customers.

Btw, welcome aboard intomydimension.

In fact AMD telling him that just matches up with what has been said in some theories about the WiiU GPU, that it might be a E6760 derivative, as mentioned in this article from July based on findings from Nintendo and Green Hill Software -whoever they are-;

http://www.nintengen.com/2012/07/speculation-wii-us-gpu-based-on-amd.html

However, according to this bar chart it's efficiency is nothing to write home about, so there is where the custom changes kick in.

14vih5u.jpg
 
Cyan: It seems to me that would just simply be too good to be true.
Dunno... but to me it just goes to show what a regular user can do by just asking a question directly to the manufacturer, and how the company after such a simple question gave him information journalists would kill for!!!!!!! Yet not a single newspaper or gaming site knew about this nor echoed it.

I am thinking about sending AMD an email and see what happens. I love their style to be honest.
 
Dunno... but to me it just goes to show what a regular user can do by just asking a question directly to the manufacturer, and how the company after such a simple question gave him information journalists would kill for!!!!!!! Yet not a single newspaper or gaming site knew about this nor echoed it.

I'm guessing we'll find out when anandtech (or whoever) does a teardown of the console and reports back their findings.

I for one can't wait.
 
http://www.eurogamer.net/articles/2012-09-21-a-chat-about-the-power-of-the-wii-u-with-the-developer-of-a-wii-u-launch-title



Final nail in the coffin.

I'm starting to believe that anonymous ubisoft employee from awhile back who said that the CPU was on par(a little bit weaker), but that the GPU was 1.5x.

How on earth did you deduce this from that statement? They are porting a current generation game where the PS3 was the lead platform for development. The PS3 CPU takes much of the load off of the GPU. KT are then taking their proprietary engine to an OoOe CPU much more dependant upon the GPGPU. A whole different beast architeturally, & you neglect the most important quote: ""With the Wii U being new hardware, we're still getting used to developing for it, so there are still a lot of things we don't know yet to bring out the most of the processing power."
 
Status
Not open for further replies.
Back
Top