Support NeoGAF

[Durante]

Would it?

It totally would. It would bring some interest back to what looks like a terribly boring console generation all around. I'd actually prefer it to the 4 fat cores / 4 Jaguar cores rumour (which I consider far more likely).

As the number of cores increases, efficiency always goes down. It's not a big issue for typical supercomputing number crunching, but games are a very different animal.

I do believe the vast majority of computationally intensive tasks in games are parallelizable to some extent. It's interesting to see people use this argument against having many cores, while at the same time pushing for GPGPU. Are you aware that with the latter it's usually accepted that you need to run in the thousands of independent computations ("threads") to achieve good efficiency?

 

[z0m3le]

Even at 1.6 Ghz, 8 Jaguar cores (which is the absolute minimum they should go with with such a tiny/low power core) would still deliver over 50 GFlops from what I can decipher from the specs. And hey, maybe MS does go with 16 after all. That would be pretty awesome.

Before the PS3 and XB360 came out, the GPUs were 1 TFLOPs, don't believe everything you hear. BTW, where did that 50GFLOPs come from? Bobcat does ~1.5GFLOPs per core @ 1.6GHz and AMD has stated that Jaguar only yields about 20% increase in performance... Even doubling bobcat's GFLOPs, you are looking at about ~24GFLOPs @ 1.6GHz /w 8 cores, and that is being really generous as far as I can tell...

Performance on current-gen ports is a mixed bag trending on the "slower than 6years+ old competition" side of things. In no way do they prove the CPU can keep up.

Besides, I don't even buy into the PS4 Jaguar rumour. It would be a hefty performance regression, which is what you're counting on to make the WiiU CPU look competitive. But this won't happen. Jaguar/APU stuff in consoles, it it materializes at all, will be used for low power media modes, active standby with downloads continuing, that sort of thing. It will not be what runs the games. It can't. It would make no sense to launch a successor machine with those sorts of components at all.

A lot of those 360 side ports run better on Wii U than PS3.

I'm not counting on anything, I'm just speculating on popular rumors, however the likely hood that either console will have Cell's GFLOPs performance this next gen is extremely unlikely.

 

[Durante]

I calculated that from the information about the architecture, which led me to assume 4 MADs per clock. It may just be half of that though, you are right. When you look at die partitioning pictures of Jaguar vis-a-vis Bobcat you'll see that the part allocated to the FPU almost doubled in relative size.

In any case, like Rolf I'm quite doubtful that we'll see a Jaguar-only console. If we do, there will be a lot of them. I feel like this is all quite off topic in this thread.

 

[wsippel]

It totally would. It would bring some interest back to what looks like a terribly boring console generation all around. I'd actually prefer it to the 4 fat cores / 4 Jaguar cores rumour (which I consider far more likely).

I do believe the vast majority of computationally intensive tasks in games are parallelizable to some extent. It's interesting to see people use this argument against having many cores, while at the same time pushing for GPGPU. Are you aware that with the latter it's usually accepted that you need to run in the thousands of independent computations ("threads") to achieve good efficiency?

I am. Massive parallel is certainly a great approach for certain workloads. But I think load balancing with GPGPU is a lot more efficient. You can theoretically tie GPGPU and dynamic resolution together or something. Simply put: shit explodes, resolution goes down. The gamer probably won't even notice. CPU cores idling on the other hand are a waste of silicon.

 

[Durante]

I am. Massive parallel is certainly a great approach for certain workloads. But I think load balancing with GPGPU is a lot more efficient. You can theoretically tie GPGPU and dynamic resolution together or something.

Dynamic resolution is certainly a nice way to achieve scalability -- for rendering. Which is already one of the most easily parallelizable tasks in a game, which is why GPUs are by far the most parallel devices in common use. However, when people say "GPGPU" they usually mean computations other than rendering (that would just be "GPU, no?). Those often aren't that easy to scale up to a massive amount of independent WIs.

 

[wsippel]

Dynamic resolution is certainly a nice way to achieve scalability -- for rendering. Which is already one of the most easily parallelizable tasks in a game, which is why GPUs are by far the most parallel devices in common use. However, when people say "GPGPU" they usually mean computations other than rendering (that would just be "GPU, no?). Those often aren't that easy to scale up to a massive amount of independent WIs.

Yeah, but why not tie the two together? Like I wrote: Shit explodes, you need processing power to calculate the physics, resolution goes down. The more stuff is going on, the less you notice a lower resolution, so that approach should work really well.

 

[Donnie]

It's not an "enigma" at all. It's just a different architecture. The fact that one piece of hadware can be both faster and slower than another depending on the workload is only unusual if you have an overly simplified view of performance.

Generally, there are a few things very wrong with the CPU conversation in this thread:
1) Multipliers applied to the overall performance of a CPU. That just can't work for two architectures which are very different.
2) Using Cell and Xenon interchangeably for comparison. They are very much different. Cell (in PS3) has 7 SPEs, and even a single SPE can offer better performance than a single PPE.
3) People freaking out when you call the CPU "weak" and starting to throw out buzzwords and -sentences which they read somewhere without fully understanding them.

That's one workload. The real question is, in how much of an average game workload does the CPU perform worse/better compared to Xenon? My assumption is that, as long as developers don't put the majority of their FP code on the GPU (which requires significant effort and re-engineering, and takes away GPU performance from graphics), there will be a lot of FP code which performs significantly worse.

Add in another one to:

4) People calling Espresso weak when what they actually mean to say is lower peak floating point performance. As you say there are different work loads that all need to be taken into account when judging the performance of a CPU.

Performance on current-gen ports is a mixed bag trending on the "slower than 6years+ old competition" side of things. In no way do they prove the CPU can keep up.

Yet most of these mixed bag of games perform worse on PS3, which is supposed to have easily the most powerful CPU. I think what z0m3le is saying is that when people are claiming the WiiU CPU is horribly slow and then its putting out launch ports that are either very close or in some cases better than PS3/360 that doesn't add up.

Besides, I don't even buy into the PS4 Jaguar rumour. It would be a hefty performance regression, which is what you're counting on to make the WiiU CPU look competitive. But this won't happen. Jaguar/APU stuff in consoles, it it materializes at all, will be used for low power media modes, active standby with downloads continuing, that sort of thing. It will not be what runs the games. It can't. It would make no sense to launch a successor machine with those sorts of components at all.

A 8 core CPU and GPU inside a console to be used for media and standby downloading?

 

[Durante]

Yeah, but why not tie the two together? Like I wrote: Shit explodes, you need processing power to calculate the physics, resolution goes down. The more stuff is going on, the less you notice a lower resolution, so that approach should work really well.

That's an interesting idea and might work for some scenarios, depending on how well you can multiplex graphical and general purpose computation on a given GPU without impacting the efficiency of either.

Yet most of these mixed bag of games perform worse on PS3, which is supposed to have easily the most powerful CPU. I think what z0m3le is saying is that when people are claiming the WiiU CPU is horribly slow and then its putting out launch ports that are either very close or in some cases better than PS3/360 that doesn't add up.

It totally adds up when you take into account that Xenon is also a terrible CPU by modern standards (outside of FP performance it wasn't that hot 5 years ago either), and that Cell likely isn't really being used in most multiplats (and also has to make up for a severely outdated GPU).

Espresso is a weak CPU by any modern standard. Well, maybe not by the standards of tablets, which it should rightfully be compared to considering its size and power consumption.

A 8 core CPU and GPU inside a console to be used for media and standby downloading?

The "OS Jaguar" rumour actually had 4 fat cores (for games) and 4 Jaguar cores IIRC.

 

[wsippel]

That's an interesting idea and might work for some scenarios, depending on how well you can multiplex graphical and general purpose computation on a given GPU without impacting the efficiency of either.

That's the million dollar question, and if Nintendo was truly serious about the whole GPGPU thing, is hopefully exactly what they focussed on.

What I consider pretty interesting is that the Wii U as a whole might be pretty weird. The R700 line was meant to be a lot of things, but at the end, it was just a competent GPU. Maybe PC conventions and DirectX held it back? AMD was very ambitious back then. R700 was what Havok Cloth was demonstrated on. It was the platform running Froblins, computing pathfinding for hundreds of AI entities on a GPU. It was the first and only GPU to ever use the infamous "2nd generation" AMD tessellation engine that was never used and isn't even properly documented anywhere. All of that was supposed to be a big deal, but it all fizzled out. Was it because the concepts or implementations were bad, or was it simply because the industry as a whole moved in another direction?

Sometimes chips appear that are far ahead of the curve. ArtX' Flipper was one such chip. A lot of the tech those dudes came up with made ATI what it was, and even when we're talking about the Xenos eDRAM, that's actually ArtX technology from back in 1999. Gamecube had it. 3Dc made a huge difference for 360, and I believe that's tech first used in the old X800 GPUs. It took ages until Nvidia implemented it and Microsoft adopted it as part of the D3D standard.

 

[Rolf NB]

Yet most of these mixed bag of games perform worse on PS3, which is supposed to have easily the most powerful CPU. I think what z0m3le is saying is that when people are claiming the WiiU CPU is horribly slow and then its putting out launch ports that are either very close or in some cases better than PS3/360 that doesn't add up.

Most, really? In any case, underused SPEs should tell you something about the potential of software being offloaded to "GPGPU" calculations. Nintendo fans still hope this will boost the system's performance to more impressive levels, but actually offloading to "GPGPU" is significantly harder, more restrictive and less efficient than offloading to SPEs. If there still remain multiplatform developers that haven't figured out a use case for SPEs, any hope of "GPGPU" creating a performance advantage for the WiiU is folly.

A 8 core CPU and GPU inside a console to be used for media and standby downloading?

If you mean 8 Jaguar cores, put that in the "will not happen" bin as well.

 

[USC-fan]

That's the million dollar question, and if Nintendo was truly serious about the whole GPGPU thing, is hopefully exactly what they focussed on.

What I consider pretty interesting is that the Wii U as a whole might be pretty weird. The R700 line was meant to be a lot of things, but at the end, it was just a competent GPU. Maybe PC conventions and DirectX held it back? AMD was very ambitious back then. R700 was what Havok Cloth was demonstrated on. It was the platform running Froblins, computing pathfinding for hundreds of AI entities on a GPU. It was the first and only GPU to ever use the infamous "2nd generation" AMD tessellation engine that was never used and isn't even properly documented anywhere. All of that was supposed to be a big deal, but it all fizzled out. Was it because the concepts or implementations were bad, or was it simply because the industry as a whole moved in another direction?

Sometimes chips appear that are far ahead of the curve. ArtX' Flipper was one such chip. A lot of the tech those dudes came up with made ATI what it was, and even when we're talking about the Xenos eDRAM, that's actually ArtX technology from back in 1999. Gamecube had it. 3Dc made a huge difference for 360, and I believe that's tech first used in the old X800 GPUs. It took ages until Nvidia implemented it and Microsoft adopted it as part of the D3D standard.

r700 gpgpu performance is just bad. If they did go with a gpgpu to make up for the poor performing CPU, that is just a bad design. Better gpgpu design have been out for years...

 

[z0m3le]

r700 gpgpu performance is just bad. If they did go with a gpgpu to make up for the poor performing CPU, that is just a bad design. Better gpgpu design have been out for years...

Do you remember why that was true mostly? because iirc, Wii U already would solve a big problem with the R700's compute performance.

 

[Donnie]

It totally adds up when you take into account that Xenon is also a terrible CPU by modern standards (outside of FP performance it wasn't that hot 5 years ago either), and that Cell likely isn't really being used in most multiplats (and also has to make up for a severely outdated GPU).

Espresso is a weak CPU by any modern standard. Well, maybe not by the standards of tablets, which it should rightfully be compared to considering its size and power consumption

I certainly don't think the majority of claims on this forum about WiiU's CPU problems are in reference to modern PC CPU's. If you're referring to the developer comment itself then yeah its quite possible that's what they meant, a pretty odd comment though considering they're working with Xenon. But I suppose comments like that are easily taken out of all context by "journalists".

Most, really? In any case, underused SPEs should tell you something about the potential of software being offloaded to "GPGPU" calculations. Nintendo fans still hope this will boost the system's performance to more impressive levels, but actually offloading to "GPGPU" is significantly harder, more restrictive and less efficient than offloading to SPEs. If there still remain multiplatform developers that haven't figured out a use case for SPEs, any hope of "GPGPU" creating a performance advantage for the WiiU is folly.

Yes most, and no its not about GPGPU boosting performance, or creating a performance advantage, at least not in my view. Its just a feature that developers can use WiiU's extra shader performance on if they choose to. Also unlike an SPE its not extra silicon that's going to sit there unused if a developer decides not to do any general purpose processing with it, and I think that's a big part of why Nintendo want to go this route. Anyway my point is, for all the bile spouted by many here about WiiUs CPU its handling quick ports of new 360/PS3 games without offloading CPU tasks to the GPU, I wonder how many of them even offloaded audio to the DSP?

 

[Donnie]

Do you remember why that was true mostly? because iirc, Wii U already would solve a big problem with the R700's compute performance.

Don't bother, he can't grasp the basic idea of a custom GPU or modifications changing the performance of a part.

 

[lwilliams3]

Do you remember why that was true mostly? because iirc, Wii U already would solve a big problem with the R700's compute performance.

Yes. The 32MB of on-chip local memory with the GPU should help the performance of GPGPU tasks.

 

[Donnie]

Yes. The 32MB of on-chip local memory with the GPU should help the performance of GPGPU tasks.

AMD have also increased register memory which would improve GPGPU performance.

 

[phosphor112]

Because if you compare the 360's 3 core Xenon to the Wii's 1 core broadway, you get this:

That post is silly. When comparing tech, you can't just ignore different aspects of the chips. Just like how you can't just say the Wii U CPU sucks because it's clocked slower... it's OoO. But this guy completely ignores (he even says it himself) the multithreading.

To be fair though, game developers have learned to go multithreading heavy...

This allowed devs to push out all sorts of performance from the devices, even though earlier he completely dismisses and contradicts himself by saying multithreading is:

...and multithreading isn't a be-all end-all solution and isn't a 'one size fits all' magic wand either...

Hell, the PPU's probably are weak. It might explain why 360 barely had games that pushed online player count passed 12, since a strong CPU is required for netcoding and such. I don't know, but comparing 1 Wii cpu and the 3 core Xenon and ignoring the multithreading is one of the stupidest things I've seen. It's just as bad as ignoring the OoO performance that the Wii U CPU has.

 

[Datschge]

The R700 line was meant to be a lot of things, but at the end, it was just a competent GPU. Maybe PC conventions and DirectX held it back? AMD was very ambitious back then. R700 was what Havok Cloth was demonstrated on. It was the platform running Froblins, computing pathfinding for hundreds of AI entities on a GPU. It was the first and only GPU to ever use the infamous "2nd generation" AMD tessellation engine that was never used and isn't even properly documented anywhere. All of that was supposed to be a big deal, but it all fizzled out. Was it because the concepts or implementations were bad, or was it simply because the industry as a whole moved in another direction?

ATi in general and R700 in special suffered on developers (even the in-house ones wrt drivers) not being willing or able to create software running directly on the iron and the abstraction layer (DirectX) heading into an ever so slightly different direction, forcing much of the capability to lay dormant. nVidia always mitigated it by obfuscating the actual processing abilities of their chips through their driver layer (which gives them both a competitive advantage, using the same closed source driver base even on open source systems, as well as the ability to support newer tech on older hardware), while ATi always has been more of an interesting chip designer with a lack of software making good use of it. Console systems give an opportunity to offer new technological approaches to an audience more willing to adapt to it (as the hardware is fixed and coding to the iron can give competitive advantages), your example of 3Dc is an apt example where it worked out well. ArtX' TEV engine on the other hand, while capability-wise being the most powerful at the release and competitive (resolution aside) even through the Wii years, was barely exploited. I'm looking forward to seeing where GPGPU with unified memory address space will be heading.

 

[Thraktor]

As this is the technical thread, I thought I may as well ask this here: what regions are you guys in, and what is the frequency range on your Wii U gamepads? (This is on the label on the back of the gamepad). I ask due to the whole "gamepads are region-locked" thing, and was wondering if it might be related to them transmitting/receiving in different bands in different territories.

Mine's 5150MHz to 5250MHz, and it's a European Wii U.

 

[UltimateIke]

As this is the technical thread, I thought I may as well ask this here: what regions are you guys in, and what is the frequency range on your Wii U gamepads? (This is on the label on the back of the gamepad). I ask due to the whole "gamepads are region-locked" thing, and was wondering if it might be related to them transmitting/receiving in different bands in different territories.

Mine's 5150MHz to 5250MHz, and it's a European Wii U.

Mine is the same on a North American Wii U.

 

[Thraktor]

Mine is the same on a North American Wii U.

Thanks, I wonder if it's the same on a Japanese controller.

 

[McHuj]

I calculated that from the information about the architecture, which led me to assume 4 MADs per clock. It may just be half of that though, you are right. When you look at die partitioning pictures of Jaguar vis-a-vis Bobcat you'll see that the part allocated to the FPU almost doubled in relative size.

In any case, like Rolf I'm quite doubtful that we'll see a Jaguar-only console. If we do, there will be a lot of them. I feel like this is all quite off topic in this thread.

You're right. It's about 12.8 Gigaflops per jaguar core (51.2 Gigaflops per 4 core CPU). Jaguar can issue two 128-bit (4 float adds + 4 float multiplies) per cycle 1.6 * 8 * 4. Bobcat had half that performance. That's single precision floats, double precision is much worse. It's 3 ops (2 adds/1 mult per cycle).

Both Jaguar and A10 (Piledriver) have the same performance per clock per core in terms of RAW gigaflops: 8 flops/per clock/per core. Piledriver is designed to be clocked higher. 2 Jaguar cores at 1.6 GHz will offer the same theoretical performance as a single 3.2 GHz piledriver, but I believe will be much lower power. I'm not sure Steamroller improves on raw flop performance, I doubt they're planning on doubling the vector units.

 

[shinra-bansho]

WiiUs CPU its handling quick ports of new 360/PS3 games without offloading CPU tasks to the GPU, I wonder how many of them even offloaded audio to the DSP?

Out of curiosity, do we know this to be the case for all of these ports?

 

[bobeth]

Out of curiosity, do we know this to be the case for all of these ports?

This is pure speculation.

 

[wonderdung]

I wonder how much floating point heavy code currently used can be redesigned to work better on a CPU like Espresso. I mean achieving comparable, maybe even better results using different solvers/ algorithms.

At least on Wii, moving vanilla floating point code to paired singles could often yield significant speedups. But PS will always have a rough time catching reasonably optimized VMX. You get half the MD in your SI... it's not rocket science.

It does have one benefit over VMX in that it doesn't have to switch register sets to go between SIMD and regular floating point. This is one thing that slows VMX down a lot in PS3 and Xbox. So you could potentially get gains from more code, but you'll just never get around the fact that the PS can only work with two things at a time. In addition the PS instruction set is missing one or two instructions that would save some work... that's just a personal annoyance though. Could have made cross product a little more elegant, but they didn't.

Really though, PS has no business existing in a CPU released in a commercial product in 2012. It's a huge shame. (At least from a development standpoint, from a hardware design standpoint in a full BC console it's probably an amazing technical decision.)

Edit: I'm maybe being too harsh on PS, but the fact is it's a known quantity as Nintendo developers have been messing with it for over a decade now. Magic new algorithmic speedups aren't going to come out of the woodwork.

 

[phosphor112]

Thanks, I wonder if it's the same on a Japanese controller.

It's the same. That tech is a wifi in the 5ghz range. It won't differ from region to region just because they are region locked.

 

[PetrCobra]

As this is the technical thread, I thought I may as well ask this here: what regions are you guys in, and what is the frequency range on your Wii U gamepads? (This is on the label on the back of the gamepad). I ask due to the whole "gamepads are region-locked" thing, and was wondering if it might be related to them transmitting/receiving in different bands in different territories.

Mine's 5150MHz to 5250MHz, and it's a European Wii U.

I didn't really do any research about this, but someone in the thread about this issue said that according to the source article the issue is most likely related to the Wii U being unable to download a firmware update for a Gamepad from a different region, and that the Gamepad itself can be synced to that Wii U without problems. Just saying, in case anyone would be interested to find out what's really going on with this.

 

[z0m3le]

That post is silly. When comparing tech, you can't just ignore different aspects of the chips. Just like how you can't just say the Wii U CPU sucks because it's clocked slower... it's OoO. But this guy completely ignores (he even says it himself) the multithreading.

Wii's CPU is OoO, so comparing a broadway clocked at 1.24GHz vs Xenon clocked
at 3.2GHz is all I'm doing, yes I'm talking core vs core.

This allowed devs to push out all sorts of performance from the devices, even though earlier he completely dismisses and contradicts himself by saying multithreading is:

multithreading isn't going to matter unless it's used, remember that Wii U has a lot of specific hardware for things Xenon had to do alone, such as audio, OS, and I/O(?).

In fact, if you ask most PC gamers which CPU they would get, the i5 or i7, they will tell you i5, because while hyperthreading (multithreading) is nice for some tasks, it's pretty much irrelevant for 99.9% of today's games.

Hell, the PPU's probably are weak. It might explain why 360 barely had games that pushed online player count passed 12, since a strong CPU is required for netcoding and such. I don't know, but comparing 1 Wii cpu and the 3 core Xenon and ignoring the multithreading is one of the stupidest things I've seen. It's just as bad as ignoring the OoO performance that the Wii U CPU has.

Wii's CPU is OoO, so comparing a broadway clocked at 1.24GHz vs Xenon clocked at 3.2GHz is all I'm doing, yes I'm talking core vs core.

hopefully that clears it up for you.

Again I am not saying right out that Wii U's CPU is ~50% faster than Xenon in everything, but it's general computation should be somewhere around that figure, and being left behind by a far greater amount in FP operations.

 

[phosphor112]

It's cleared up now, but the way people are posting makes it seems like it's one Wii core vs all of Xenon (3 cores) which is a dumb comparison, because multi-threading IS used. You can't just forget about it.

 

[z0m3le]

It's cleared up now, but the way people are posting makes it seems like it's one Wii core vs all of Xenon (3 cores) which is a dumb comparison, because multi-threading IS used. You can't just forget about it.

Yes that is dumb. As for the multithreading, again it's not really that important unless you have a 4 threads that all need to run at once (this is for a 3 core comparison) Xenon always uses 1/6th of it's power or more for audio, OS takes up processing time too and I'm not sure, but if the i/o doesn't have it's own hardware, then that could blunt Xenon's performance too.

Wii U has separate hardware for all of those things mentioned, so when you compare Wii U to Xenon, it's a bit easier to see why multithreading doesn't matter much, what I haven't guessed at though, is how much more efficient espresso is over broadway, it likely has other advancements that could boost performance as well, one of the reasons I say this is because of the rather large cache the processor has, and as I've pointed out before I really am not sure how important the CPUs are going to be in this next generation, I think most graphical advancements right now are being done in lighting and tessellation, which are both GPU ops.

 

[Fess]

Yes that is dumb. As for the multithreading, again it's not really that important unless you have a 4 threads that all need to run at once (this is for a 3 core comparison) Xenon always uses 1/6th of it's power or more for audio, OS takes up processing time too and I'm not sure, but if the i/o doesn't have it's own hardware, then that could blunt Xenon's performance too.

Wii U has separate hardware for all of those things mentioned, so when you compare Wii U to Xenon, it's a bit easier to see why multithreading doesn't matter much, what I haven't guessed at though, is how much more efficient espresso is over broadway, it likely has other advancements that could boost performance as well, one of the reasons I say this is because of the rather large cache the processor has, and as I've pointed out before I really am not sure how important the CPUs are going to be in this next generation, I think most graphical advancements right now are being done in lighting and tessellation, which are both GPU ops.

Why do you say that multithreading doesn't matter? Isn't multithreading used all the time and basically the only way to get high performance out of the current consoles?

 

[z0m3le]

Why do you say that multithreading doesn't matter? Isn't multithreading used all the time and basically the only way to get high performance out of the current consoles?

Because for gaming, if you look at PCs it doesnt. The reason multithreading matters in current consoles is because of all the extra work they have to do that Wii U has dedicated hardware for. Multiple threads are important, but that can be solved with multiple cores, and PC gamers for the most part don't see a performance increase beyond 3 cores, except for select games.

BTW, AMD has no multithreading option, so if both XB3 and PS4 use AMD CPUs, no next gen console will have multithreading. (This is a fact, not an assumption)

 

[Fess]

Because for gaming, if you look at PCs it doesnt. The reason multithreading matters in current consoles is because of all the extra work they have to do that Wii U has dedicated hardware for. Multiple threads are important, but that can be solved with multiple cores, and PC gamers for the most part don't see a performance increase beyond 3 cores, except for select games.

BTW, AMD has no multithreading option, so if both XB3 and PS4 use AMD CPUs, no next gen console will have multithreading. (This is a fact, not an assumption)

From what I've seen in interviews with first party devs they have been using separate threads for things like physics, ai, water, particles, clothsimulation, streaming, etc. Definitely not just audio and os stuff. Edit: Although, I'm not quite sure how using multithreading on single cores differs from using single threads on multiple cores, seems like the same thing to me as long as you can have the same amount of threads running simultaneously.

 

[SquiddyCracker]

So has anyone tried turning on two Wii U pads in the same room as one Wii U console?
I'm curious as to if there's any sort of recognition of multiple pads in the OS already.

 

[blu]

You're right. It's about 12.8 Gigaflops per jaguar core (51.2 Gigaflops per 4 core CPU). Jaguar can issue two 128-bit (4 float adds + 4 float multiplies) per cycle 1.6 * 8 * 4. Bobcat had half that performance. That's single precision floats, double precision is much worse. It's 3 ops (2 adds/1 mult per cycle).

Both Jaguar and A10 (Piledriver) have the same performance per clock per core in terms of RAW gigaflops: 8 flops/per clock/per core. Piledriver is designed to be clocked higher. 2 Jaguar cores at 1.6 GHz will offer the same theoretical performance as a single 3.2 GHz piledriver, but I believe will be much lower power. I'm not sure Steamroller improves on raw flop performance, I doubt they're planning on doubling the vector units.

You're right about Jaguar and Bobcat. Speaking of those, they do arguably have one architectural efficiency drawback, though (compared to many other CPUs, ppc included) - they don't do MADDs, which means that they might suffer from increased register pressure during tight MADD loop unrolls (read: the code must make sure the dst reg from the mul phase is preserved until the add phase (which can be arbitrary remote, due to latency issues), which means there's one less register available (or worse - a register spill) in comparison to architectures which have a straight MADD).

 

[lightchris]

BTW, AMD has no multithreading option, so if both XB3 and PS4 use AMD CPUs, no next gen console will have multithreading. (This is a fact, not an assumption)

Multithreading != SMT. I think that's where the confusion came from.

And SMT should decently help Xenon to utilize it's in-order pipeline better.

 

[z0m3le]

Multithreading != SMT. I think that's where the confusion came from.

Which is why I said multiple threads and multithreading separately in my post. I know it's not proper, but I thought it was clear enough that I was talking about SMT with the latter.

And SMT should decently help Xenon to utilize it's in-order pipeline better.

That is also interesting to note, that SMT helps with in-order efficiency. Still considering that most multiplatform games have to target the single PPE on the PS3, it technically only has 2 threads that can perform more general purpose code correct? or am I wrong on the idea of the SPEs?

And like I said, if for instance (not that they will, but just as an example) if PS4 used a 6 core AMD jaguar CPU, it would have no more threads than Xenon. AMD is locked at 1 thread per core. (modules are 2 core chips that are basically Siamese-twins)

 

[Thraktor]

It's the same. That tech is a wifi in the 5ghz range. It won't differ from region to region just because they are region locked.

You've got the causality the wrong way around. I was wondering whether it is region locked because they chose different bands in different regions. While it is based on WiFi, that doesn't necessarily mean it has to remain limited to the frequency bands which are used by the 802.11 standard (which are chosen because they're available almost everywhere), and they may have decided to switch to a different frequency band in one or more regions because it would provide a less congested communications channel.

I didn't really do any research about this, but someone in the thread about this issue said that according to the source article the issue is most likely related to the Wii U being unable to download a firmware update for a Gamepad from a different region, and that the Gamepad itself can be synced to that Wii U without problems. Just saying, in case anyone would be interested to find out what's really going on with this.

Yep, that appears to be the case. I'd still be interested to get confirmation on the frequency range of a Japanese gamepad, though.

 

[phosphor112]

You've got the causality the wrong way around. I was wondering whether it is region locked because they chose different bands in different regions. While it is based on WiFi, that doesn't necessarily mean it has to remain limited to the frequency bands which are used by the 802.11 standard (which are chosen because they're available almost everywhere), and they may have decided to switch to a different frequency band in one or more regions because it would provide a less congested communications channel.

Ah, ok, makes more sense. But yeah, it's the same everywhere. It's part of how the tech works, and the high frequency means less interference.

 

[Donnie]

Out of curiosity, do we know this to be the case for all of these ports?

We haven't heard it from a developer, not that we would necessarily hear about it anyway though.

Moving general purpose code to a GPU isn't a trivial task however, far from it. So the fact these are launch ports of 360 games makes the likelihood of any GPGPU usage extremely low IMO.

 

[cyberheater]

I don't know where to put this. Most of the general WiiU threads get closed.

Is there a fix for the black level issue on Netflix. It's way too high.

 

[Durante]

You're right. It's about 12.8 Gigaflops per jaguar core (51.2 Gigaflops per 4 core CPU). Jaguar can issue two 128-bit (4 float adds + 4 float multiplies) per cycle 1.6 * 8 * 4. Bobcat had half that performance. That's single precision floats, double precision is much worse. It's 3 ops (2 adds/1 mult per cycle).

Both Jaguar and A10 (Piledriver) have the same performance per clock per core in terms of RAW gigaflops: 8 flops/per clock/per core. Piledriver is designed to be clocked higher. 2 Jaguar cores at 1.6 GHz will offer the same theoretical performance as a single 3.2 GHz piledriver, but I believe will be much lower power. I'm not sure Steamroller improves on raw flop performance, I doubt they're planning on doubling the vector units.

Thanks for the confirmation, those were my assumptions from reading the specs. So in terms of (single-precision) FLOPs per Watt/area Jaguar is actually pretty efficient for a CPU. I like the idea of having a large number of Jaguar cores more and more.

 

[Thraktor]

Thanks for the confirmation, those were my assumptions from reading the specs. So in terms of (single-precision) FLOPs per Watt/area Jaguar is actually pretty efficient for a CPU. I like the idea of having a large number of Jaguar cores more and more.

It definitely seems that, if you've got X mm^2 and Y watts to play with, lots of Jaguar cores would be a better bet than a few Bulldozer cores. That's unless there's going to be anything that specifically needs the sort of per-thread performance of Bulldozer over that of Jaguar, but given that Bulldozer hardly has exceptional per-thread performance in the first place, and you'd have to underclock it significantly to prevent it melting the console, I honestly doubt that's the case.

 

[OryoN]

Not sure how related this is, but since Wii U's CPU is described as "enhanced broadway" cores, I guess it's relevant...

Despite how "weak" Wii was compared to other consoles, I was surprised that it did had games that focused mainly on physics - which struck me as a bit odd. Right out of the gate, there was Elebits, then Boom Blox(and it's sequel) came along. Some will write those off as "not impressive" simply based on simplistic visuals, but they did a really good job off showcasing what Broadway could do for physics.

Below is a video - of Elebits' editor mode - actually boasting about this very thing.
http://www.youtube.com/watch?v=3xJXvFqhCk0


This makes me anxious to see what would be possible(if devs pushed it) now that Wii U have higher clocked multi-core CPUs, with extra large caches, low-latency access to the GPU's Edram, and who knows what other "enhancements."

What's strange, is that for all the power PS360 packed in their CPU's, we didn't see physics & AI taken to "the next-level" as developers promise each generation. Of course, it could be argued that this generation was more focused on pushing GPUs than anything else. Console CPU's this gen were in some ways over-powered, but were somewhat inefficient and spent a lot of time doing tasks that dedicated chips could have been used for. It does seem like this gen was a failed experiment hoping that a ridiculous amount of CPU FLOPs would bring significant advances in Physics and AI.

That didn't quite pan out, so now it appears that next-gen consoles CPUs will be trading a lot of those FLOPs in favor of efficiency, large caches, plus dedicated silicon and/or enough cores to handle various tasks. That combined with a strong focus on GPGPU capabilities...maybe this is the better formula for the results we are promised each generation?

 

[Popstar]

.

 

[phosphor112]

Good physics and AI aren't simply a programming / CPU resource problem. They're also a design problem. If you improve them and showcase them in your game, by their nature you take away control from the designer. I think the move towards more tightly controlled and scripted experiences this passing generation is as responsible as any technical limitation.

everythingwasuncharted.gif

Pretty much. There are some games that simply don't need crazy physics calculations. BF3? Yeah, doesn't need super crazy physics, though it does have a lot running already, including momentum.

LBP is a game with "full" physics calculations though and is impressive as well.
http://www.youtube.com/watch?v=LyRoWdpxiLo
Sure it isn't 60fps, but this game has a lot going for it.

Also, GT5 and Forza 4 are games with phenomenal physics engines that both run at 60fps.

Halo also runs pretty impressive physics calculations at once.

 

[pottuvoi]

We certainly saw decent jump in physics when comparing to previous generation.

 

[saunderez]

We certainly saw decent jump in physics when comparing to previous generation.

Watching Euphoria do its stuff in Backbreaker is like poetry in motion. So good.

 

[phosphor112]

Watching Euphoria do its stuff in Backbreaker is like poetry in motion. So good.

The NFL Blitz remake should have been blitz 64 with new graphics and euphoria for tackles. That's it. Too bad it's not though. =[

 

[wsippel]

Watching Euphoria do its stuff in Backbreaker is like poetry in motion. So good.

I despise the NaturalMotion shit in most games. It introduces a massive disconnect between the player's actions and the actions performed by the character. Controls (and by proxy animations) need to be tight, precise and 100% predictable. Realism is a bonus. Uncharted was a great example of an animation system getting in the way.