• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Rumor: Wii U final specs

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
I didn't mean to imply that the eDRAM wouldn't be used to store the framebuffer and some render targets, but I don't think it's a dedicated framebuffer.
That's a safe bet. I mean, Xenos' immediate offspring Yamato (Adreno) does not use its edram (sram) as a dedicated framebuffer, so I don't see a reason why a design several generations later would suddenly return to an 'edram == fb' formula.
 
Nope, it's simple math...although I'd personally say 3-4 times:

2GB of RAM compared to 512MB.
3MB of CPU cache compared to 1MB.
32MB of eDRAM on the GPU compared to 10MB.
A GPU that's at least 2 generations ahead of the Xenos.
A GPU that's likely to have a superior tesselation unit and General Processing capabilities.
A CPU with Out of Order Execution instead of In Order Execution.
A CPU that isn't going to be bogged down with processing audio, IO and the OS.

The above is why it's not cheap. We're certainly not talking about another Wii situation here.

Sounds right to me too.
 

F#A#Oo

Banned
*some* people sure... but what Dave is getting at is that GAF isn't a gestalt entity, and that for there to actually be hypocrisy you have to find individual posters being hypocrites.

you can't call out GAF for hypocrisy. GAF is what? thousands of individuals all with different tastes. name names or don't bother.

I'll go with the general majority and feel of a thread as a basis for measure...it's about the only decent method available for measuring response on here. I mean really individuals responses really?

In general terms the GAF membership got the Wii wrong, it got the DS wrong, it got the PSP wrong. You can read those threads and get a feel for what's being said...

I mean take this thread as an example...some people are expecting Nintendo/IBM/AMD to perform magic and re-write certain things in technological terms. It's not going to happen.
 

i-Lo

Member
Nope, it's simple math...although I'd personally say 3-4 times:

2GB of RAM compared to 512MB.
3MB of CPU cache compared to 1MB.
32MB of eDRAM on the GPU compared to 10MB.
A GPU that's at least 2 generations ahead of the Xenos.
A GPU that's likely to have a superior tesselation unit and General Processing capabilities.
A CPU with Out of Order Execution instead of In Order Execution.
A CPU that isn't going to be bogged down with processing audio, IO and the OS.

The above is why it's not cheap. We're certainly not talking about another Wii situation here.

What boggles my mind is that the OS requires 1GB (allegedly). I hope the OS blows people's mind with features and presentation.

Also, it'll be interesting to see what type of RAM it is, i.e. GDDR3 or not.
 

i-Lo

Member

Durante

Member
Excactly what are the benefits of OoOE processing? (If its abbreviated like that).
Out of order execution enables a CPU to execute instructions in a different order than the one specified in the program, as long as that is allowed by their dependencies. It uses a number of additional "hidden" (not visible in the ISA) registers and an instruction queue to store operations and intermediate results. It's one of many techniques in a modern CPU to increase throughput by hiding instruction latencies and increasing the amount of instruction-level parallelism (ILP) and thus improve instructions per clock (IPC).

Basically, the worse (in terms of instruction scheduling) the code generated by a compiler (or written by a human) is, the larger the benefit of OOE. However, there are also situations where a CPU capable of OOE can perform better (clock-for-clock) than one without OOE, even if an oracle (perfect) instruction scheduler was available to the compiler.

Conversely, with in-order processing, the quality of the binary code matters far more, since the CPU hardware can not mitigate bad instruction schedules. All of this of course makes the idea that developers are too used to in-order architectures and thus incapable of "optimizing for OOE" hilarious.

what blu is implying is that since GDDR3 is EOL, yes the Wii U will use it.
Ba-dum-ching.
 
What boggles my mind is that the OS requires 1GB (allegedly). I hope the OS blows people's mind with features and presentation.

Also, it'll be interesting to see what type of RAM it is, i.e. GDDR3 or not.

I'd go with the general consensus of opinion and say DDR3 with the eDRAM preventing any bottlenecking issues.

And it wouldn't surprise me if Sony and Microsoft go with the same too given the target specs for each machines. Can't see how they're going to squeeze 2-4GB in the PS4 and 6-8GB in the 720 otherwise, unless the density of GDDR5 doubles before they're released. And that much RAM is not going to be cheap either.
 
Out of order execution enables a CPU to execute instructions in a different order than the one specified in the program, as long as that is allowed by their dependencies. It uses a number of additional "hidden" (not visible in the ISA) registers and an instruction queue to store operations and intermediate results. It's one of many techniques in a modern CPU to increase the amount of instruction-level parallelism (ILP) and thus improve instructions per clock (IPC).

Basically, the worse (in terms of instruction scheduling) the code generated by a compiler (or written by a human) is, the larger the benefit of OOE. However, there are also situations where a CPU capable of OOE can perform better (clock-for-clock) than one without OOE, even if an oracle (perfect) instruction scheduler was available to the compiler.

Conversely, with in-order processing, the quality of the binary code matters far more, since the CPU hardware can not mitigate bad instruction schedules. All of this of course makes the idea that developers are too used to in-order architectures and thus incapable of "optimizing for OOE" hilarious.

Ba-dum-ching.

This was actually one of the early problems had with the PS3 and 360. Since they couldn't process out of order, generated code was often really slow. Later better compilers were written with better prediction that helped code execution significantly.
 

Durante

Member
And it wouldn't surprise me if Sony and Microsoft go with the same too given the target specs for each machines. Can't see how they're going to squeeze 2-4GB in the PS4 and 6-8GB in the 720 otherwise, unless the density of GDDR5 doubles before they're released. And that much RAM is not going to be cheap either.
There are €120 GPUs out right now with 2 GB of GDDR5. I'm not saying that they will use it (unlike many I have no claim to any "insider information"), but I really don't see a problem with going for that (or even 4 GB late next year).
 

i-Lo

Member
I'd go with the general consensus of opinion and say DDR3 with the eDRAM preventing any bottlenecking issues.

And it wouldn't surprise me if Sony and Microsoft go with the same too given the target specs for each machines. Can't see how they're going to squeeze 2-4GB in the PS4 and 6-8GB in the 720 otherwise, unless the density of GDDR5 doubles before they're released. And that much RAM is not going to be cheap either.

Oh it's quite possible to attain it with current densities of GDDR5. It's just a matter of development costs and heat management. If there are solid indications of increase in densities a year or so from now, perhaps we may see MS and Sony take some losses out of the gate for future guarantee.

A confirmation of the region locking. I thought it was apparent.


*tips hat*

Yea, I already saw it on the other thread. I was asking about the HW. I thought the link had something to do with RAM.
 
What boggles my mind is that the OS requires 1GB (allegedly). I hope the OS blows people's mind with features and presentation.

Also, it'll be interesting to see what type of RAM it is, i.e. GDDR3 or not.
I've heard the OS also has it's own 512MB flash nand so it doesn't draw from the 8GB/32GB to be used for games. This is a nice change from Wii, which took 256MB for OS/backend from the 512MB flash iirc.

Also mentioned it in another thread, but that 1GB is likely to come down and be given back to developers at some point. It did for 3DS already, going from 64MB reserved to 32MB reserved I believe.
 
There are €120 GPUs out right now with 2 GB of GDDR5. I'm not saying that they will use it (unlike many I have no claim to any "insider information"), but I really don't see a problem with going for that (or even 4 GB late next year).

You're talking about a PC graphics card, we're not going to see a graphics card inside any console.
 
Out of order execution enables a CPU to execute instructions in a different order than the one specified in the program, as long as that is allowed by their dependencies. It uses a number of additional "hidden" (not visible in the ISA) registers and an instruction queue to store operations and intermediate results. It's one of many techniques in a modern CPU to increase throughput by hiding instruction latencies and increasing the amount of instruction-level parallelism (ILP) and thus improve instructions per clock (IPC).

Basically, the worse (in terms of instruction scheduling) the code generated by a compiler (or written by a human) is, the larger the benefit of OOE. However, there are also situations where a CPU capable of OOE can perform better (clock-for-clock) than one without OOE, even if an oracle (perfect) instruction scheduler was available to the compiler.

Conversely, with in-order processing, the quality of the binary code matters far more, since the CPU hardware can not mitigate bad instruction schedules. All of this of course makes the idea that developers are too used to in-order architectures and thus incapable of "optimizing for OOE" hilarious.

Ba-dum-ching.

How much time is the CPU wasting to rearrange instructions that are already otimized?
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
That's correct, but there are valid comparisons to be made. What's a console but a video card with CPU, Storage, and IO subsystems bolted on?
In this case? A miserable pile of secrets.
 

Durante

Member
You're talking about a PC graphics card, we're not going to see a graphics card inside any console.
I'm talking about the cost of GDDR5 memory as implied by the end-user sale price of a product that uses it. If you have a better or more direct source (other than "man, console RAM is expensive!") go ahead and post it!

How much time is the CPU wasting to rearrange instructions that are already otimized?
Generally? None.
If you want to talk about the pro/contra of OOE you need to get into the "inefficient" use of transistors that the additional hardware for OOE requires (particularly if most of that hardware is unused since your instructions are already perfectly scheduled). But since transistors are the least of your problems these days in a general purpose CPU I believe OOE is the right choice.
 
Oh it's quite possible to attain it with current densities of GDDR5. It's just a matter of development costs and heat management. If there are solid indications of increase in densities a year or so from now, perhaps we may see MS and Sony take some losses out of the gate for future guarantee.



Yea, I already saw it on the other thread. I was asking about the HW. I thought the link had something to do with RAM.

It's not just development costs and heat management being an issue, we're talking about 8 chips of GDDR5 to make up 4GB and 16 chips to make up 8GB so you've got motherboard complexity and bus issues to take into account too as far as I know, although I will stand to be corrected on that.

Edit: Oops, got my maths wrong, that's 16 chips for 4GB and 32 chips for 8GB, which is even more ridiculous! :Oo
 
Seriously, do people just pull these numbers from thin air or something...

Nope, it's simple math...although I'd personally say 3-4 times:

2GB of RAM compared to 512MB.
3MB of CPU cache compared to 1MB.
32MB of eDRAM on the GPU compared to 10MB.
A GPU that's at least 2 generations ahead of the Xenos.
A GPU that's likely to have a superior tesselation unit and General Processing capabilities.
A CPU with Out of Order Execution instead of In Order Execution.
A CPU that isn't going to be bogged down with processing audio, IO and the OS.

The above is why it's not cheap. We're certainly not talking about another Wii situation here.

Pretty much.

We know that RAM wise, the Wii U is 2-4x ahead of the current generation of consoles (Wii excluded), same for the eDRAM.

The CPU and GPU are guesses, I confess, but I think it's safe to say that they too will be 2-4x better than the HD twins - especially as the CPU will be assisted by various extra transistors non-existent in the 360 as snowdog pointed out.

Speaking of which, is it true that the CPU won't have to deal with the processing audio, IO, and the OS?
 
2GB of RAM compared to 512MB. <--- only 1gb for games (2x)
3MB of CPU cache compared to 1MB. <--- and devs saying weaker cpu
32MB of eDRAM on the GPU compared to 10MB. <--- used on?
 

Brashnir

Member
Nope, it's simple math...although I'd personally say 3-4 times:

2GB of RAM compared to 512MB.
3MB of CPU cache compared to 1MB.
32MB of eDRAM on the GPU compared to 10MB.
A GPU that's at least 2 generations ahead of the Xenos.
A GPU that's likely to have a superior tesselation unit and General Processing capabilities.
A CPU with Out of Order Execution instead of In Order Execution.
A CPU that isn't going to be bogged down with processing audio, IO and the OS.

The above is why it's not cheap. We're certainly not talking about another Wii situation here.

You had me up until the last sentence. What you describe pretty much IS the Wii situation. Where do you think the "2 Gamecubes Duct Taped Together" meme came from?
 
Look what I just noticed in my inbox today...Who is writing these responses? What does he actually know? Is this some sort of conspiracy? Maybe I'll get more emails...

Nb0Ap.png
 

Donnie

Member
Still not entirely sold on the CPU being totally up to snuff, and IMO it's where I look for a bottleneck to come 1st. Though I fully admit I could be wrong.

The GPU is light years beyond the 360 and PS3 though. Anything respectable from the 4XXX series onward IMO is a metric ton better than either of the current HD console setups.

I'll be interested to see how the RAM pans out, and if over the life cycle of the system if Nintendo gives devs access to the other 1 gig RAM pool down the road.

I'm also interested in how loud or quiet the system will be when doing some serious tasks and accessing the big disc.

Well as I've said for a while I think the whole CPU issue comes down to differences rather than necessarily a problem with raw performance. One of the developers interviewed recently by Eurogamer about the CPU put the problems they were having down to this. Basically its a new CPU for them and they haven't had time to optimise things, but they expect better performance once they do. Once developers get to grips with the CPU I think its going to have performance very similar to Xenon and probably better in some ways.

But moving away from a direct CPU comparison for a moment its important to remember that, unlike 360, WiiU doesn't have just a main CPU to handle the usual CPU tasks. It has three separate processors. A main triple core IBM CPU with 1 thread per core (each thread effectively 33.33% of the CPU's total performance), an audio processor (DSP) and an I/O processor (most likely an ARM CPU clocked somewhere around 500Mhz). 360's CPU is a triple core IBM CPU with 2 threads per core (each thread is effectively 16.66% of the CPU's performance). More threads doesn't necessarily give you better performance but it can help in efficiency if you expect to use a CPU for many different tasks at once. A typical use for Xenon in game could look something like this:

AI (1 Thread)
physics (2 Threads)
Audio (1 Thread)
I/O (1 Thread)
Game Code (1 Thread)

With WiiU right now you might have:

AI/GameCode (1 core)
Physics (1 core)
Audio/I/O (1 core)

But use the extra hardware and you can go:

AI (1 core)
Physics (1 core)
Game Code (1 core)
Audio (DSP)
I/O (ARM CPU)

So while IMO the CPU's raw performance will be Xenon or better I doubt its as happy as Xenon when handling 5 or 6 different tasks at once due to less threads and that is likely one of the things causing efficiency problems. Use the CPU plus the dedicated audio and I/O hardware and it should leave Xenon behind by a decent margin. Because not only do you leave 3 tasks for 3 threads (which should improve efficiency) but you also take 30%+ load away from the CPU and give it to other tasks like AI and game code).
 

Donnie

Member
2GB of RAM compared to 512MB. <--- only 1gb for games (2x)
3MB of CPU cache compared to 1MB. <--- and devs saying weaker cpu
32MB of eDRAM on the GPU compared to 10MB. <--- used on?

XBox 360 has 480MB available for games. Also while currently 1GB is available for WiiU its very likely at least some of the extra 1GB will be added to that later on, just like PS3 started out with 392MB available for games and ended with 464MB for games.

See my last post RE the CPU

The eDRAM will be used primarily for the frame buffer. More means you can fit a full HD frame in there all at once instead of having to render half, then the second half and attach them in main RAM like 360. Doing it that way is less efficient/slower. It can also be used for some other tasks.
 

Absinthe

Member
With (more than) Double the RAM and processing power, while the 360/PS3 were more in the range of 15-20x the RAM and processing power of Gamecube.

Keep in mind, the jumps in tech have been less and less each generation, so your conclusion is not exactly accurate.
 
XBox 360 has 480MB available for games. Also while currently 1GB is available for WiiU its very likely at least some of the extra 1GB will be added to that later on, just like PS3 started out with 392MB available for games and ended with 464MB for games.

Speculation.

The eDRAM will be used primarily for the frame buffer. More mains you can fit a full HD frame in there all at once instead of having to render half, then the second half and attach them in main RAM like 360. Doing it that way is less efficient/slower.

Speculation.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
With (more than) Double the RAM and processing power..
CPU and GPU were both 1.5x the clock, main pool (1T-sram) was intact, and so was the GPU edram. What was clearly extra in terms of power was the secondary pool of 16MB A-RAM -> 64MB GDDR3, and the fact Hollywood could feed from those.
 
You had me up until the last sentence. What you describe pretty much IS the Wii situation. Where do you think the "2 Gamecubes Duct Taped Together" meme came from?
For it to be a Wii situation, Wii U would need a GPU weaker/older than Xenos and about 40% more RAM than PS360. Instead it has a GPGPU (equivalent to 1-2 "gens" newer) and 4 times the RAM. It's less like Wii vs Xbox and more like Xbox vs Dreamcast.
 
Nope, it's simple math...although I'd personally say 3-4 times:

2GB of RAM compared to 512MB.
3MB of CPU cache compared to 1MB.
32MB of eDRAM on the GPU compared to 10MB.
A GPU that's at least 2 generations ahead of the Xenos.
A GPU that's likely to have a superior tesselation unit and General Processing capabilities.
A CPU with Out of Order Execution instead of In Order Execution.
A CPU that isn't going to be bogged down with processing audio, IO and the OS.

The above is why it's not cheap. We're certainly not talking about another Wii situation here.

Are we really sure about that ? This would be very impressive.
 

prag16

Banned
Well as I've said for a while I think the whole CPU issue comes down to differences rather than necessarily a problem with raw performance. One of the developers interviewed recently by Eurogamer about the CPU put the problems they were having down to this. Basically its a new CPU for them and they haven't had time to optimise things, but they expect better performance once they do. Once developers get to grips with the CPU I think its going to have performance very similar to Xenos and probably better in some ways.

But moving away from a direct CPU comparison for a moment its important to remember that, unlike 360, WiiU doesn't have just a main CPU to handle the usual CPU tasks. It has three separate processors. A main triple core IBM CPU with 1 thread per core (each thread effectively 33.33% of the CPU's total performance), an audio processor (DSP) and an I/O processor (most likely an ARM CPU clocked somewhere around 500Mhz). 360's CPU is a triple core IBM CPU with 2 threads per core (each thread is effectively 16.66% of the CPU's performance). More threads doesn't necessarily give you better performance but it can help in efficiency if you expect to use a CPU for many different tasks at once. A typical use for Xenos in game could look something like this:

AI (1 Thread)
physics (2 Threads)
Audio (1 Thread)
I/O (1 Thread)
Game Code (1 Thread)

With WiiU right now you might have:

AI/GameCode (1 core)
Physics (1 core)
Audio/I/O (1 core)

But use the extra hardware and you can go:

AI (1 core)
Physics (1 core)
Game Code (1 core)
Audio (DSP)
I/O (ARM CPU)

So while IMO the CPU's raw performance will be Xenos or better I doubt its as happy as Xenos when handling 5 or 6 different tasks at once due to less threads and that is likely one of the things causing efficiency problems. Use the CPU plus the dedicated audio and I/O hardware and it should leave Xenos behind by a decent margin. Because not only do you leave 3 tasks for 3 threads (which should improve efficiency) but you also take 30%+ load away from the CPU and give it to other tasks like AI and game code).

Do we know how easy/hard it is for 3rd parties doing ports to offload I/O and audio to their respective secondary processors?

Is it possible that they actually have not done this at all, and are still using the main tricore CPU for all this crap? If so, that's an easy explanation why some may claim it's "weaker".
 

Donnie

Member
Speculation.

Educated speculation, its the norm for at least some OS dedicated memory to be moved over to games as the OS is refined and features are nailed down.



Speculation.

No that isn't speculation, not in the slightest. 360 has to split its frames when rendering at anything above 720p with no AA because 10MB isn't enough memory for anything more. 32MB doesn't have that problem.
 

Xanonano

Member
Look what I just noticed in my inbox today...Who is writing these responses? What does he actually know? Is this some sort of conspiracy? Maybe I'll get more emails...

Nb0Ap.png
The e6760 is an embedded GPU for things like slot machines. There's no way you are going to find them on retail shelves.
 
Top Bottom