• 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.

WiiU technical discussion (serious discussions welcome)

Single precision floating point SIMD width for Jaguar (2 for Wii U, with paired singles).

Jaguar can operate on 128 bit SIMD (ie 4 x 32 bit), whereas Espresso's limited to 64 bit (2 x 32 bit) "paired singles" SIMD.

Ok. Thanks for clarifying.

Where was this confirmed that the ARM did anything for the OS? and that the WiiU multitasks with games?

I'm not sure if it was ever officially confirmed, though Marcan stated that there is an ARM processor in the system that at least does the tasks that Wii's ARM "Scarlet" did. I believe Wsippel's research led him to unoffically confirm that the ARM is mult-core.

As for the background OS tasks running off the ARM, I honestly think that just makes sense considering that Espresso only has 3 cores to work with and we haven't heard anything about any of those cores being locked up.
 

Kenka

Member
There's a big difference between 360-Wii ports and 720-Wii U ports. In the former case, there's a huge gulf in functionality, and a huge gulf in power. In the latter, there's a relatively small gulf in functionality, and a moderately large gulf in power. The Wii simply couldn't do things the 360 could do (ie programmable shaders), so games like CoD had to be rebuilt from the ground up for Wii. By contrast, a sufficiently scaleable engine should allow a single game to be cross-developed for Wii U, 720 and PS4.

Of course, the more important numbers when it comes to ports are financial numbers. Fundamentally, if a publisher believes that the revenue gained from porting a game to a platform exceed the costs of porting said game to said platform, then they'll pay for the port. If they don't, they won't.
Actually, it's a bit harsher (for Nintendo) that that: if they think the best way to use their resources is to develop games for the WiiU, they will do it. There might be a lot of games that could be profitable for WiiU, but other projects on the twins can prove more profitable even, and thus, no project would be ever greenlit by a big publisher on WiiU because better opportunities are to be sought somewhere else. It's an explanation that would hold particularly true in the actual case of the 3DS; it doesn't get games because S-E, Konami, EA, Ubi, Rockstar and the others think it's more profitable to put their eggs in other baskets. Which in the case of S-E is totally stupid.
 

wsippel

Banned
Imo it's likely that they didn't change much besides adding SMP support. At least up to now nothing hints at that.

Anyway, it's barely debatable that Jaguar's IPC are at least a bit higher. And then we have 8 instead of 3 cores and a 30% higher clock rate which adds up to a factor of 3.5.
I agree with oversitting that the CPU could be crucial when it comes to downports.
Well, they had to change a lot:

No PPC750 ever supported SMP.
No PPC750 was ever manufactured at 45nm.
No PPC750 ever used eDRAM.
No PPC750 supported more than 1MB L2 cache.
No PPC750 was designed to run at clock speeds exceeding 1.1GHz.

Not to mention Gekko and Broadway were no off-the-shelf cores, either. Initial 750s didn't support paired singles, and Nintendo's versions have several dozen additional instructions - they were already more of a superset. And contrary to popular opinion, not even Broadway was just a shrinked Gekko. Some logic units were completely redesigned.
 

Oblivion

Fetishing muscular manly men in skintight hosery
Sorry for the double post, but I had to get this off my chest.

After browsing the "games known for their visuals" thread someone reminded me of the conduit for Wii. For those that don't know, The Conduit was suppose to be a game promising Xbox 360 graphics on the Wii. They even made a video showcasing what their engine could do.
https://www.youtube.com/watch?v=5tovnipDToc

It obviously never matched the 360 1:1 but it was an impressive effort given the API they had to work with.

Now imagine if a developer went through the same effort for Wii U, trying to bring PS4/720 visuals to the console. It's unlikely to be 1:1 but with 3 multicore OOE processors, 1GB of RAM and a 2012 GPU, the difference between Wii U and PS4/720 should be far negligible this time.

While I appreciate HV's work in trying to create some impressive tech when few others bothered on Wii, it's still somewhat depressing to see a developer brag about tech as old as it was.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
It depends on what you are looking at, but in some metrics the CPU performance difference could be >7x.
The above needs the remark that due to x86-64 simd actually not having a MADD op, that ratio is for one cpu doing MADDs while the other does MULs & ADDs, which is not exactly equivalent coding-wise.
 
Well, they had to change a lot:

No PPC750 ever supported SMP.
No PPC750 was ever manufactured at 45nm.
No PPC750 ever used eDRAM.
No PPC750 supported more than 1MB L2 cache.
No PPC750 was designed to run at clock speeds exceeding 1.1GHz.

Not to mention Gekko and Broadway were no off-the-shelf cores, either. Initial 750s didn't support paired singles, and Nintendo's versions have several dozen additional instructions - they were already more of a superset. And contrary to popular opinion, not even Broadway was just a shrinked Gekko. Some logic units were completely redesigned.
I'm really curious about that. Would you care to be a bit more specific? I thought that Broadway was a shrinked-overclocked Gekko.
 

Gahiggidy

My aunt & uncle run a Mom & Pop store, "The Gamecube Hut", and sold 80k WiiU within minutes of opening.
What was Wii's imbalance?

While in the vertical orientation, the system could tip over, explaining that dumb acrylic circle included in the box.
 

tipoo

Banned
There is 8 of them as well as they being built on 28nm. The cores inside the Wii-u are probably just as small at 45nm.

Also if the wii-u cpu is based on the ibm 750, that architecture is 15 years old. Jaguar is almost completely new abide it does draw from the same design as bobcat. The speed that technology advanced in those years, its likely the wii-u is missing a lot of cpu.

The Core 2 Duo is "based on" the Pentium 3 processor, which is also just under 15 years old. I don't think the Wii U processor is a performance champ in any measure (just based on the size and what fab process it's on - IBM can't pull miracles), but that criticism is unfair because we don't know how loose "based on" is. This criticism hs been put down repeatedly here.


Also, Espresso contains three cores and is 32.76mm^2. Granted, that's including cache, but IBM uses eDRAM which offers roughly three times the density compared to traditional SRAM.



Not sure which way you're leaning with this? A Core 2 Duo also on 45nm is over 120mm2 (just going by memory, somehwere around there), and I think Cell is still over 100mm2 on 45nm. Yes the eDRAM offers thrice the density, but even if you shrank the cache on a Core 2 Duo down that much it would not be that small (just visually approximating 1MB more cache on this 2MB Core 2 Duo)

intel_skulltrail_part_ii___penryn_die.jpg



Each core is really tiny. The whole die is about the size of a single core Atom. But then again, AMD Jaguar cores are tiny too and supposedly that's what other next gen consoles will use, just many more than 3 of them.

Edit: to clarify the cache is the uniform large squares at the bottom.
 

tipoo

Banned
I thought eDRAM is only on the L3 and the 3MB of L2 is SRAM. 3MB of L2 is substantial, this isn't even counting the rest of the uncore.

Wait now, when did we get to it having 3MB L2 cache? I thought it just had 3MB L3 cache made of eDRAM like other IBM designs. I've never heard of L2 and L3 cache being the same size.
 
Well, they had to change a lot:

No PPC750 was ever manufactured at 45nm.
No PPC750 ever used eDRAM.
No PPC750 supported more than 1MB L2 cache.
No PPC750 was designed to run at clock speeds exceeding 1.1GHz.

I would call those minor changes that don't imply any deeper overhaul of the acutal architecture. Far from exciting from a performance standpoint.

Of course we don't know for sure what else might have changed. It's just my guess that it won't be much, amongst others because the development costs would've been high.
 

wsippel

Banned
I'm really curious about that. Would you care to be a bit more specific? I thought that Broadway was a shrinked-overclocked Gekko.
It was menioned by IBM engineers in Linkedin profiles a while ago. Don't know what exactly was redesigned and why. The only unit specified in one profile was the adder.


Not sure which way you're leaning with this? A Core 2 Duo also on 45nm is over 120mm2 (just going by memory, somehwere around there), and I think Cell is still over 100mm2 on 45nm.
That's pretty much my point. Espresso cores are by no means huge, but neither are other embedded cores like ARM - or Jaguar.

Also, Espresso has no L3 cache. The L2 is eDRAM. Which is also why they could afford having a whopping 3MB. Apparently.


I would call those minor changes that don't imply any deeper overhaul of the acutal architecture. Far from exciting from a performance standpoint.

Of course we don't know for sure what else might have changed. It's just my guess that it won't be much, amongst others because the development costs would've been high.
I wouldn't be surprised seeing changes at least comparable to Gekko, which was a very significant evolution of the "regular" PPC750CXe, extending the ISA by several dozen mosly graphics related instructions and sporting a heavily modified FPU introducing limited SIMD capabilities to the 750 line. Nintendo did it once, I see no reason for them not to do it again. If that leads to a smaller, more efficient chip, it would probably warrant the higher development costs.
 

ikioi

Banned
That's a common misconception. A 100% efficient 70W PSU delivers 70W of power. A 80% efficient 70W PSU also delivers 70W of power only that it must draw 87.5W from the wall. For a good PSU it's efficiency affects it's power bill not it's output.

We're not talking about power draw from the wall. We're talking about rated output, which is afaik what the Wii U's PSU's 75w is labeled as.

But you are 100% correct.

I really hope the CPU and memory bandwidth don't cause issues and Nintendo have found some architecture to get around them both. I just can't see how this system is even going to be capable of exceeding PS3/Xbox 360 level graphics when the CPU and MEM1 bandwidth seem to gimped
 

wsippel

Banned
Isn't L2 going to be much slower and harder to access with eDRAM?
IBM uses that exact technology for two of their latest and greatest performance chips, so apparently not.

Looking at Blue Gene/Q might be interesting in that regard, as that chip does some interesting things with its L2. Multi-versioning, atomic transactions and such.
 

Chronos24

Member
Idea man comments about a next gen game once thought not possible on the WiiU is now not only possible but may be in development for the WiiU.

Which I would love to hear from Ideaman again with a tidbit more info to maybe at least tell us if we are close on some of the great estimates we've seen here.
 

Antiochus

Member
It appears that no one has yet managed to crack open the GPU and see just what R700 (4000 series) model it most resembles, let alone benchmark it at all. It appears an absolute NDA regarding Wii U technical specs exists after all, just no publicly.

But going from the scant info present, it appears the closest R700 GPU it resembles is the 4650/4670, which would be a clone of the 3850/3870 of a generation previously. Assuming being programmed as part of streamlined console would bring a moderate boost, the final performance would, at the most generous metric, be close to that a HD 4830. What does that mean? In most likelihood, we'll see close to max quality Crysis 1(c.2007-2008) visuals near the end of Wii U's lifespan......at 720p. Perhaps a bit higher resolution, but it will never reach max quality at 1080p, while Orbis and Durango most certainly can, right at debut.
 

Chronos24

Member
If we are talking about the CPU just for gaming, we need to take out the 2 cores that Durango was reported to be using for the OS. Wii U has a multi-core ARM processor for that. So that would be x5.33 just for gaming with your calculations.

BTW, what is the "4" in your formula for?

I haven't heard much about the ARM processor. Could someone maybe explain this processor to me and any info on its specs? I keep hearing its multi core as well too so I'm a little confused about it. As always I am trying to raise the limits of my technical knowledge :)
 

AzaK

Member
If we are talking about the CPU just for gaming, we need to take out the 2 cores that Durango was reported to be using for the OS. Wii U has a multi-core ARM processor for that.

And possibly another core for audio, assuming a developer (Likely only Nintendo) uses the DSP.

EDit: Wait, whaaa?!! Do you think the OS is running on the ARM and not the main CPU?
 

ozfunghi

Member
But going from the scant info present, it appears the closest R700 GPU it resembles is the 4650/4670, which would be a clone of the 3850/3870 of a generation previously. Assuming being programmed as part of streamlined console would bring a moderate boost, the final performance would, at the most generous metric, be close to that a HD 4830. What does that mean? In most likelihood, we'll see close to max quality Crysis 1(c.2007-2008) visuals near the end of Wii U's lifespan......at 720p. Perhaps a bit higher resolution, but it will never reach max quality at 1080p, while Orbis and Durango most certainly can, right at debut.

How exactly were you able to deduce this from those scans?
 

wsippel

Banned
And possibly another core for audio, assuming a developer (Likely only Nintendo) uses the DSP.
Every developer using middleware, or at least fmod, for audio also uses the DSP. fmod even supports DSP/ CPU load balancing on Wii U, which is pretty awesome. Basically, if too much happens and the DSP can't keep up, it will overflow to the CPU before dropping stuff - if the CPU has cycles to spare that is.

But I expect all next generation consoles to have dedicated audio DSPs. Sony and Microsoft made a mistake not using DSPs last gen - every game has sound after all, and DSPs are much more efficient at audio processing than any CPU.
 

ultrazilla

Gold Member
It appears that no one has yet managed to crack open the GPU and see just what R700 (4000 series) model it most resembles, let alone benchmark it at all. It appears an absolute NDA regarding Wii U technical specs exists after all, just no publicly.

But going from the scant info present, it appears the closest R700 GPU it resembles is the 4650/4670, which would be a clone of the 3850/3870 of a generation previously. Assuming being programmed as part of streamlined console would bring a moderate boost, the final performance would, at the most generous metric, be close to that a HD 4830. What does that mean? In most likelihood, we'll see close to max quality Crysis 1(c.2007-2008) visuals near the end of Wii U's lifespan......at 720p. Perhaps a bit higher resolution, but it will never reach max quality at 1080p, while Orbis and Durango most certainly can, right at debut.

I'm pretty sure Crytek has said they could port Crysis 3 to the WiiU if they
wanted with little to any visual fidelity loss.

And the very strong rumors that the WiiU can run the Unreal Engine 4 leads
me to believe it'll be a stronger system than most now believe it is.
 
We're not talking about power draw from the wall. We're talking about rated output, which is afaik what the Wii U's PSU's 75w is labeled as.

But you are 100% correct.

I really hope the CPU and memory bandwidth don't cause issues and Nintendo have found some architecture to get around them both. I just can't see how this system is even going to be capable of exceeding PS3/Xbox 360 level graphics when the CPU and MEM1 bandwidth seem to gimped

MEM1 is the EDRAM, not what's 12.8GB/s(MEM2)
 
IBM uses that exact technology for two of their latest and greatest performance chips, so apparently not.

Looking at Blue Gene/Q might be interesting in that regard, as that chip does some interesting things with its L2. Multi-versioning, atomic transactions and such.

The L2 is also eDRAM in that? I thought power7 only used eDRAM for the L3 to buffer main memory.
 

wsippel

Banned
cool, did not know that.
Both those chips are quite exotic, so that isn't all that surprising. They're targeting very specific applications. PowerEN is for ultra high end networking equipment, and Blue Gene/Q is a real supercomputer. What's interesting is that Blue Gene/Q not only uses eDRAM for L2, it also does a few unusual things with its L2 as I mentioned.
 

Margalis

Banned
Can anyone compare what we know about the WiiU with what we heard of VGLeaks about Durango and Orbis ? It seems the power gap won't be insurmountable but who am I to talk...

Durango is two WiiU's duct-taped together.

This is actually kind of true.
 

Kenka

Member
OK, the popular opinion is that between the PS2 and the XBOX three was a three times performance gap. I assume the DC -> PS2 gap was like two times. That would imply the XBOX was "roughly" (whatever that means) 6 times more powerful than the DC. Is there any indication that such a gap today would result in different development philosophies ? It certainly didn't back then.

I think we are a majority in this thread who only want to know if we can get the same games as the other players out there. So I assume this is the central question, and it's not all about the tech. Is there any game designer who was in touch with all three different dev kits and who could make a comment about how he/she would tackle development on Durango/Orbis and the WiiU ? I hope the answer is that the approach is similar, that would encourage us in being positive about the future library of the console.
 

Margalis

Banned
I think the reality is that by constantly sabotaging their own products, releasing idiotic "test" games, looking at games like Mad World as barometers, etc, most third party devs have convinced themselves that "normal" games don't sell on Nintendo systems, and thus lack the business case for doing a port even if they are technically able.
 

tipoo

Banned
IBM uses that exact technology for two of their latest and greatest performance chips, so apparently not.

Looking at Blue Gene/Q might be interesting in that regard, as that chip does some interesting things with its L2. Multi-versioning, atomic transactions and such.



IBM uses it for high performance server and workstation chips that optimize for high throughput use. They bet on larger caches cancelling out the fact that eDRAM is slower than SRAM. It works well for them in high performance compute, that's one area where they beat even the mighty Intel.

That said I have absolutely no idea how it would scale to a tiny chip with 3MB of it at 1.2GHz. It's thrice as dense as SRAM, so the alternative would be 1MB of faster SRAM, I have no idea if being three times that capacity makes up for the lower speed in a gaming context.

Durango is two WiiU's duct-taped together.

This is actually kind of true.


Kind of true based on what? 5GB RAM (potentially 7 if it's true they skinnied down the OS) available to games vs 1, 8 cores vs 3, still no idea what the U GPU is like but both have embedded memory to help out and the nextbox may have faster eSRAM.

I don't think it will be the Wii-360 gulf again, but I think 2x is way underestimating it.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
IBM uses it for high performance server and workstation chips that optimize for high throughput use. They bet on larger caches cancelling out the fact that eDRAM is slower than SRAM.
It's not slower than SRAM for certain amounts up.
 

Azure J

Member
Having moved away from a lot of the technical discussions regarding Wii U since the last Speculation Thread and seeing the Durango layout has me feeling a little nostalgic right now. I'm kinda surprised both at how Durango is built as the furthest extreme of the Wii U's architectural design and at how close Nintendo got to "playing with the big boys" completely.

This has managed to not only make me want to see more of Durango's potential but also begun to be very content with Wii U's potential for growth.
 

wsippel

Banned
IBM uses it for high performance server and workstation chips that optimize for high throughput use. They bet on larger caches cancelling out the fact that eDRAM is slower than SRAM. It works well for them in high performance compute, that's one area where they beat even the mighty Intel.

That said I have absolutely no idea how it would scale to a tiny chip with 3MB of it at 1.2GHz. It's thrice as dense as SRAM, so the alternative would be 1MB of faster SRAM, I have no idea if being three times that capacity makes up for the lower speed in a gaming context.
I don't know either, but it was certainly more R&D effort to go with eDRAM and the chip would cost the same if they used 1MB SRAM, so 3MB eDRAM being superior is the most logical conclusion.
 
Top Bottom