Wii U CPU Espresso can use Latte's eDRAM...

[eyeofcore]

Hi there, so I tweeted Marcan about if Wii U's CPU Espresso can use eDRAM from Latte GPU and he confirmed this, I asked him that question because Shin'en hinted it in their interview at hdwarriors website...

https://twitter.com/marcan42/status/387699522459426817

http://hdwarriors.com/general-impression-of-wii-u-edram-explained-by-shinen/

So what are implications since the CPU can use eDRAM of the GPU? Is this in a way AMD's hUMA, sort of just in a different way? Also how much Gflops does Wii U's CPU have since by my rough estimates it should have 15Gflops like the mobile Core 2 Duo T9300 or it is higher?

 

[ShintaiDK]

I get it to 15Gflops as well for 16bit flops on the 64bit FPU. Or 7.5Gflops for regular 32bit SP flops. And in INT it goes completely wrong for the CPU.

The T9300 would do 40gflops for 16bit flops or 20 for SP.

Wii U CPU is much weaker than the Xbox360 CPU.

I assume they used it for backwards compability. Because its a turd of a CPU.

 

[podspi]

It makes you wonder why they didn't just include a hardware emulator on the thing (like the early PS3s). But that really isn't the Nintendo way. They like to make money on the hardware on day 1.

 

[eyeofcore]

I get it to 15Gflops as well for 16bit flops on the 64bit FPU. Or 7.5Gflops for regular 32bit SP flops. And in INT it goes completely wrong for the CPU.

The T9300 would do 40gflops for 16bit flops or 20 for SP.

Wii U CPU is much weaker than the Xbox360 CPU.

I assume they used it for backwards compability. Because its a turd of a CPU.

Wii U's CPU is not weaker, most launch games and ports use 2 cores and they are Xbox 360 ports so games ran poorly and there difference between Wii U's Espresso and Xbox 360's Xenon is different... Also Wii U's Pipeline is like 8 to 10 times shorter and has much more L2 Cache and as Marcan and Shin'en said, it can access Latte's eDRAM so there should be benefits in terms of latency between CPU and GPU.

Here I got Gflops info about C2D T9300 and it says 15Gflops for it;
http://www.overclock.net/t/586994/cpu-gflop-performance-database

Also this;
http://www.cinemablend.com/games/Pr...e-Shadows-Multi-Threaded-Shadowing-59659.html

People only speculate it is three PPC750CL ducktapped together because Marcan said it has similarities though that does not mean that there are not modifications to increase the speed and look at the clocks and it is 4 stage pipeline so it is very efficient.

 

[USER8000]

It seems Watch Dogs better than the XBox360 and PS3 on the Wii U:

http://uk.ign.com/articles/2013/10/08/watch-dogs-on-wii-u-is-better-than-current-gen

"I would say the Wii U version is pretty much in between what the current gen is and what the next gen is from a version standpoint," Morin told CVG.
"It's hard to position it. I would tend to say it's maybe a bit closer to current gen than next gen for certain things but it's a beautiful game on Wii U and it's cool to play it just on the GamePad."

 

[eyeofcore]

It seems Watch Dogs better than the XBox360 and PS3 on the Wii U:

http://uk.ign.com/articles/2013/10/08/watch-dogs-on-wii-u-is-better-than-current-gen

Of course it will ran/look better, Wii U's GPU Latte supports Open GL 4.3/OpenCL 1.2/Shader Model 5.0 and has been proven in one of preview videos of Super Mario 3D World and I am sick of people forcing it is RV7XX, its not;

http://www.youtube.com/watch?v=i3arXxOvp7U

 

[ShintaiDK]

Wii U's CPU is not weaker, most launch games and ports use 2 cores and they are Xbox 360 ports so games ran poorly and there difference between Wii U's Espresso and Xbox 360's Xenon is different... Also Wii U's Pipeline is like 8 to 10 times shorter and has much more L2 Cache and as Marcan and Shin'en said, it can access Latte's eDRAM so there should be benefits in terms of latency between CPU and GPU.

Here I got Gflops info about C2D T9300 and it says 15Gflops for it;
http://www.overclock.net/t/586994/cpu-gflop-performance-database

Also this;
http://www.cinemablend.com/games/Pr...e-Shadows-Multi-Threaded-Shadowing-59659.html

People only speculate it is three PPC750CL ducktapped together because Marcan said it has similarities though that does not mean that there are not modifications to increase the speed and look at the clocks and it is 4 stage pipeline so it is very efficient.

The Wii U CPU is much weaker than the Xenons in Xbox360. Thats just a plain fact. Both are tricores, but the Xbox got 128bit SIMD as well as 3.2Ghz and SMT. The Xenon can also do 76.8Gflops 16bit on paper. While the Expresso is left with a 64bit FPU and a 32bit integer unit running at 1.25Ghz.

You use Linpack numbers. I use theoretical, just like you did for the Wii U.

 

[Exophase]

It has been known for a long time that the eDRAM was visible on the CPU address space. That by itself means nothing without knowing how fast it is. PSP could access the GPU's eDRAM on the CPU address space, but you never saw anyone making a big deal about it and I doubt it ever made a huge difference in game performance. I tried using that eDRAM to store some critical non-graphical buffers because it was lower latency than main memory, but it didn't make a difference.

I already told you how many FLOPs it has in the other thread with the explanation of how to get it (which somehow turned into "your" estimates, when the numbers you originally gave were much different..) Why are you asking for more confirmation now?

That thing about launch games using two cores sounds totally made up, why would that be true? And you're still saying the eDRAM improves latency between CPU and GPU, I already told you this is nonsense and makes no sense. The latency between CPU and GPU is dictated by the link between them, nothing else.

Marcan hasn't just speculated that it's three 750CLs, he's given very strong evidence and very much refuted people's wishful thinking that it was a more advanced CPU.

Probably won't bother saying anything more on this to you since you don't listen, and you brushed off Beyond3D completely despite having tons of experts posting there because you don't like a couple people speculating that it has 160 SPs (though you have no hard proof that it doesn't)

I get it to 15Gflops as well for 16bit flops on the 64bit FPU. Or 7.5Gflops for regular 32bit SP flops. And in INT it goes completely wrong for the CPU.

The T9300 would do 40gflops for 16bit flops or 20 for SP.

Neither of the processors you mentioned perform 16-bit floating point math.

The Xenon can also do 76.8Gflops 16bit on paper.

Nor does this one.

 

[eyeofcore]

PSP is a handheld and Wii U is a home console so there will be difference, eDRAM is bloody fast and PSP's CPU and GPU are slow so of course there would not be a huge difference versus Wii U where CPU and GPU are much stronger than PSP so using eDRAM as "unified pool of memory" on Wii U will give you some results versus on PSP that might not be designed to use eDRAM as "unified pool of memory" versus Wii U that might do that.

Oh made up, why do you think developers had time to use all three cores?! First of all those were Xbox 360 ports so I doubt they off loaded audio to HD DSP Audio chip so two cores were used for games and a third for sound or it could be that only two cores were used for game and audio together... Games on Xbox 360 only use 2 cores and 3rd is used for audio so how is that a nonsense? Explain me?! Also there is vast difference between Espresso and Xenon and are you going to be in denial about that?

He has done his test in vWii mode not in Wii U mode of the console so the CPU could act as PowerPC750 since the vWii mode would only use parts of the CPU that is Wii using. I want to say that all the improvements would not be used in vWii mode for sake of compatibility and cache usage is limited to 256KB on Core 0 and in Wii U mode it is 512KB so where is your logic? Sir?

It is not 160SP's since the GPU is made at 40nm not 55nm so your theory and theory of those people flies out of the window like a mad person throwing out the TV out of the window and the TV crashes on concrete and shatters into pieces...

It is 320 SPU's because Radeon HD 5550 fits perfectly... In every way. Deal with it!

The Wii U CPU is much weaker than the Xenons in Xbox360. Thats just a plain fact. Both are tricores, but the Xbox got 128bit SIMD as well as 3.2Ghz and SMT. The Xenon can also do 76.8Gflops 16bit on paper. While the Expresso is left with a 64bit FPU and a 32bit integer unit running at 1.25Ghz.

You use Linpack numbers. I use theoretical, just like you did for the Wii U.

Wii U's CPU has 3 times the L2 Cache and 8 to 10 times shorter pipeline so it is more efficient and it is out of order so it is better when properly used and higher Ghz does not mean more performance, megahertz myth and that is why Intel wins over AMD when comes to CPU's.

How many execution units has Xbox 360? 3... Wii U has 5 execution units per core so that's 15 so that is a benefit over Xbox 360 that has one execution unit per core.

Il ignore involving Intel's CPU performance, I think you have gone in defense mode. Waving Intel's flag? >

 

[NTMBK]

It is not 160SP's since the GPU is made at 40nm not 55nm so your theory and theory of those people flies out of the window like a mad person throwing out the TV out of the window and the TV crashes on concrete and shatters into pieces...

Wut

 

[Puppies04]

Wii U's CPU has 3 times the L2 Cache and 8 to 10 times shorter pipeline so it is more efficient and it is out of order so it is better when properly used and higher Ghz does not mean more performance, megahertz myth and that is why Intel wins over AMD when comes to CPU's.

Erm.. really??

 

[insertcarehere]

PSP is a handheld and Wii U is a home console so there will be difference, eDRAM is bloody fast and PSP's CPU and GPU are slow so of course there would not be a huge difference versus Wii U where CPU and GPU are much stronger than PSP so using eDRAM as "unified pool of memory" on Wii U will give you some results versus on PSP that might not be designed to use eDRAM as "unified pool of memory" versus Wii U that might do that.

Oh made up, why do you think developers had time to use all three cores?! First of all those were Xbox 360 ports so I doubt they off loaded audio to HD DSP Audio chip so two cores were used for games and a third for sound or it could be that only two cores were used for game and audio together... Games on Xbox 360 only use 2 cores and 3rd is used for audio so how is that a nonsense? Explain me?! Also there is vast difference between Espresso and Xenon and are you going to be in denial about that?

He has done his test in vWii mode not in Wii U mode of the console so the CPU could act as PowerPC750 since the vWii mode would only use parts of the CPU that is Wii using. I want to say that all the improvements would not be used in vWii mode for sake of compatibility and cache usage is limited to 256KB on Core 0 and in Wii U mode it is 512KB so where is your logic? Sir?

It is not 160SP's since the GPU is made at 40nm not 55nm so your theory and theory of those people flies out of the window like a mad person throwing out the TV out of the window and the TV crashes on concrete and shatters into pieces...

It is 320 SPU's because Radeon HD 5550 fits perfectly... In every way. Deal with it!

Except there is the big glaring fact that the gpu isn't fabbed at TSMC 40nm, but RENESAS 45nm, and the fact that the memory registers simply don't fit 320sps.

As far as the cpu is concerned, this was absolutely done for backwards compatibility, otherwise, it makes no sense to do such an expensive design, when an ARM based solution would be much cheaper and of similar if not higher power.

 

[Sweepr]

Oh made up, why do you think developers had time to use all three cores?!

Wut

 

[eyeofcore]

Except there is the big glaring fact that the gpu isn't fabbed at TSMC 40nm, but RENESAS 45nm, and the fact that the memory registers simply don't fit 320sps.

As far as the cpu is concerned, this was absolutely done for backwards compatibility, otherwise, it makes no sense to do such an expensive design, when an ARM based solution would be much cheaper and of similar if not higher power.

That is not a fact, people on various forums can't agree... One say TSMC because AMD's GPU's are produced there for years and they have experience while others say RENESAS because of eDRAM though TSMC would be terrible when comes to eDRAM and RENESAS when comes to GPU production since both of these companies lack knowledge of the latter.

Xbox 360 had the longest pipeline in history;
http://www.psu.com/forums/showthrea...i-and-Durango(MisterXmedia-Being-Vindicated-)

Also there is a speculation that Wii U's GPU is cut down Radeon HD e6760 so this GPU that has 480 shaders has size of 118mm^2 and with 320 shaders it would have size of 82,6mm^2 and at 45nm it would have size of below 95mm^2. It should fit easily.

ARM better then PowerPC750CL? Don't make me laugh, current ARM CPU's have IPC of an Pentium 3 so they would need to have much higher clocks and more cache.

 

[Techhog]

It makes you wonder why they didn't just include a hardware emulator on the thing (like the early PS3s). But that really isn't the Nintendo way. They like to make money on the hardware on day 1.

Using a 13-year-old CPU probably cost them profit, actually. It's just a poorly managed company that won't be around in 10 years.

I don't even see the point in discussing an archaic flop that will be out of production within 15 months

 

[dguy6789]

This should be in console gaming. This thread is almost entirely misinformation and at best in some cases just throwing around purely theoretical numbers that do not correlate with real world performance.

The $350 Wii U was still sold at a loss on launch. It's a little crazy to say the least for anyone to truly believe a machine produced in late 2012 that costs over $350 to make could be less powerful than a 2005 box. Sort of reminds me of the people that think the PS2 was more powerful than the Gamecube.

The bottom line is that most multiplatform games do look better on the Wii U. Splinter Cell Blacklist is the most obvious example. Wii U is also getting the next gen version of CoD Ghosts, not the one one that runs on the 360 and PS3. It's clearly a much more powerful system than the 360 and PS3 but nowhere near the One and PS4 obviously.

 

[ShintaiDK]

I dont think anyone said the GPU was slower...

But the CPU certainly is.

The Wii U is expensive due to the gamepad. Not the console itself.

 

[eyeofcore]

This should be in console gaming. This thread is almost entirely misinformation and at best in some cases just throwing around purely theoretical numbers that do not correlate with real world performance.

The $350 Wii U was still sold at a loss on launch. It's a little crazy to say the least for anyone to truly believe a machine produced in late 2012 that costs over $350 to make could be less powerful than a 2005 box. Sort of reminds me of the people that think the PS2 was more powerful than the Gamecube.

The bottom line is that most multiplatform games do look better on the Wii U. Splinter Cell Blacklist is the most obvious example. Wii U is also getting the next gen version of CoD Ghosts, not the one one that runs on the 360 and PS3. It's clearly a much more powerful system than the 360 and PS3 but nowhere near the One and PS4 obviously.

I totally agree with the last sentence to the last word...

People at Nintendo would need to be crazy to not make a better console than Xbox 360 and Playstation 3 plus not update their design of the console. Very early development kits of the Wii U from 2011 had Radeon HD 4850 though it seems Nintendo aimed at power consumption levels of Gamecube, so not above 45 watts and the console consumes around 30 watts.

I dont think anyone said the GPU was slower...

But the CPU certainly is.

The Wii U is expensive due to the gamepad. Not the console itself.

Comment #5 About Watch Dogs for Wii U being better than Xbox 360 or Playstation 3 version and I responded that of course it will since the GPU supports Open GL 4.3 that is better API than DirectX bloatware when its about CPU,GPU and RAM overhead.

We don't know if it is slower since most games were Xbox 360 ports so differences between Wii U's CPU Espresso and Xbox 360's CPU Xenon is earth shattering, we can't really judge the CPU of the Wii U since games are not properly optimized for it from 3rd party and extra CPU cache is not being used properly plus Wii U has HD Audio DSP so it does need audio to be ran on CPU core and the information that we know about Wii U's CPU came primarily from hackers that used vWii mode while they did not cracked the primary mode The Wii U mode.

Also please note that Xbox 360 CPU has shared L2 cache versus Wii U CPU that has Core 0 512KB Core 1 2MB and Core 2 512KB of cache, it is not unified so that is an issue, a really big one for Xbox 360 ports and developers that are used to CPU caches being unified on consoles.

If we want to know full potential of the Wii U's CPU then we need to have run benchmarks in its native mode. Since CPU is from IBM then somebody should try and run Mac OSX on it and hopefully someone will put an effort to make its GPU run properly, ultil then we don't know much.

Wii U's gamepad is expensive, it has a resistive screen, a large battery, a microphone, speakers and camera plus a simple gyro also a Wifi chip, it is a multi purpose controller.

 

[monstercameron]

I dont think anyone said the GPU was slower...

But the CPU certainly is.

The Wii U is expensive due to the gamepad. Not the console itself.

wii u gamepad bom is probably $50-90 so it aint that expensive.

 

[NTMBK]

others say RENESAS because of eDRAM

Given that the eDRAM is on a separate die, I'm not sure why that would mean the GPU was at RENESAS.

 

[NTMBK]

This should be in console gaming. This thread is almost entirely misinformation and at best in some cases just throwing around purely theoretical numbers that do not correlate with real world performance.

Nah, it should be in Highly Technical. That would be funny to watch.

 

[insertcarehere]

Given that the eDRAM is on a separate die, I'm not sure why that would mean the GPU was at RENESAS.

The edram is embedded with the gpu actually.

 

[insertcarehere]

ARM better then PowerPC750CL? Don't make me laugh, current ARM CPU's have IPC of an Pentium 3 so they would need to have much higher clocks and more cache.

And so does say, a arm cortex a15 core, which would also have the advantage of having decent floating point performance, ability to clock higher and ability to use more cutting edge fabs i.e tsmc 28nm instead of the IBM 45nm process used now

 

[N-A-N-0]

I get it to 15Gflops as well for 16bit flops on the 64bit FPU. Or 7.5Gflops for regular 32bit SP flops. And in INT it goes completely wrong for the CPU.

The T9300 would do 40gflops for 16bit flops or 20 for SP.

Wii U CPU is much weaker than the Xbox360 CPU.

I assume they used it for backwards compability. Because its a turd of a CPU.

It's a much more powerful console than the PS360 though. That's a fact thanks to the GPU, RAM, etc. Latte is also backwards compatible with the Gamecube and Wii's GPU's like Espresso is for the CPU's so I don't think that's why. Probably cost, even PS4 and XB1 are going with relatively quite weak CPU's.

 

[Exophase]

Given that the eDRAM is on a separate die, I'm not sure why that would mean the GPU was at RENESAS.

It's on the same die.

I don't know why anyone would think that using an old VLIW5 AMD GPU would imply that they're manufacturing at TSMC. They did VLIW5 at Global Foundries, why can't they do it with Renesas given enough financing from Nintendo? They've had plenty of time.

The fact that it says "Renesas Japan" printed on it instead of "Taiwan" is kind of a dead giveaway that it's made by Renesas. That, and that it has integrated eDRAM at all. And how Renesas is known to have had to shut down one of its fabs due to Nintendo not meeting expectations with Wii U orders. And Chipworks, who analyzed the chip, says it's Renesas http://chipworks.force.com/catalog/ProductDetails?sku=NIN-C10234F5&viewState=DetailView I guess the "(c) 2010 Renesas" marking on the die was kind of another clue (yes, there's also (c) IBM, because they licensed the CPU IP)