It also says three enchanced broadway cores, which may not hold up.
But okay: R700. Why couldn't the customization process account for the difference?
-
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
- Thread starter EatChildren
- Start date
It also says three enchanced broadway cores, which may not hold up.
But okay: R700. Why couldn't the customization process account for the difference?
I hope he's trolling, if not then lol.
flipswitch
Member
Is the 1gb ram shared between system memory and graphics?
That seems true.
Brad Grenz
Member
Whatever customization have been made, we're still seeing it described as "DX10.1" level.
I fully grant they customized the part to a certain number of shaders, to work with embedded DRAM, to integrate with the CPU on the same die or module, to integrate with the hardware audio subsystem, to work with the chosen memory configuration, to reduce power utilization during streaming media playback or home screen use.
What I doubt is there is some miraculous, black box "GPGPU" hardware that has been grafted on.
I fully grant they customized the part to a certain number of shaders, to work with embedded DRAM, to integrate with the CPU on the same die or module, to integrate with the hardware audio subsystem, to work with the chosen memory configuration, to reduce power utilization during streaming media playback or home screen use.
What I doubt is there is some miraculous, black box "GPGPU" hardware that has been grafted on.
DaSorcerer7
Member
bgassassin
Member
No. 1GB for system functions, 1GB for games.
Seconded.
I don't think anyone who has an understanding of it believes that though. I do believe those same people believe they've addressed the inefficiency of said architecture in some fashion.
I'll take your word.
His claim had nothing to do with GPGPU (Bgassassin brought it up because he was notoriously contradictory about that), it was about the Wii U GPU not being capable of 600GFLOPS because of the limitations of a ~40w power consumption.
Brad Grenz
Member
Probably in exactly the same way a R700 based card exceeds the DX10.1 specifications in some way.
Wouldn't it be cheaper and easier to just use a later, DX11 design from AMD than to spend the time and money rooting around in the core architecture to put a bandaid on the DX10.1 hardware? A DX11 design where, presumably, those inefficiencies have already been addressed? And if the operation of the part has been so sufficiently altered as to make an significant difference in GPGPU tasks, why do all the leaks still say, "yeah, it's basically like a R700"?
It is likely a member of the R700 family, but modified.
DaSorcerer7
Member
Does anyone one know when GPU7's development was finalised?
bgassassin
Member
Which leaks do you refer to?
Considering how long they've been working on the GPU it may not have been cheaper for what Nintendo may have been going for. After all didn't you just mention GCN and with that AMD only recently addressed those inefficiencies. And we both know Nintendo would be highly unlikely to go with something that new. So with that said the only real option is to modify a VLIW part from AMD. My take is that looking at when development started they chose that GPU, focused on improving the compute of the GPU, and "tacked on", at worst, capabilities above the 10.1-level due to dev input.
Excluding further tweaks, late Dec./early Jan.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Bah. I still call them Tualatins ;p
You're slightly confusing the nomenclatures here. The eXXX series is Freescale's nomenclature (after they spawned from Moto), IBM never changed theirs - i.e. an IBM's G3 is still a ppc7xx, and a G4 - a ppc7xxx. Freescale's nomenclature is:
- G2 -> e300
- G3 -> e500 (actually, those are well modernized, and basically, not G3 anymore; there are even 64bit versions, but I don't think they ever got a SIMD unit).
- G4 -> e600 (for a while those had their AltiVec amputated, but lately it's been reinstated)
Please note that those are only CPU architectures, the end products might be in various SoCs and trade names (QUICC, QorIQ or whatever moronic acronym somebody at FSL came up with these days).
Anyway, my point being, since Freescale was never a console part supplier (they did supply to Apple though), we should stick to AIM's/IBM's nomenclature.
Brad Grenz
Member
Modified how? Do you even know? Cause you didn't seem to even understand the thing about the CPU you were trying to leak. Was it changed in one of the many ways I just listed that don't fundamentally alter the performance per watt or GPGPU efficiency? If it was altered in such a way as to alter the performance characteristics of the R700 design, are we talking about a 5% gain? 10%? 15%? 25%? 50%? Because a lot of people want to take the word "modified" to mean 50% faster, but it could just as easily mean "we changed some I/O stuff to talk to the PPC cores".
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
They didn't just 'change some IO stuff to talk to the PPC cores'. Sorry, that's all I can say.
40 W for the overall package (console + communication to controllers) excluding heat waste leaves maybe at most 30 W for the GPU. In order to have a 600 GFLOPS output at this wattage, we are looking at a 20 GFLOPS/W energy efficiency. Is it possible to tell with much certainty that this is realistic.
Hell, how far can we go currently, 30 GFLOPS/W, 100 GFLOPS/W ?
edit: huh, what is this:
http://www.realworldtech.com/compute-efficiency-2012/
edit2: it seems it shows that the RV770 architecture is really efficient at computing (?) - a GPU is good... for computing ?
Hell, how far can we go currently, 30 GFLOPS/W, 100 GFLOPS/W ?
edit: huh, what is this:
http://www.realworldtech.com/compute-efficiency-2012/
edit2: it seems it shows that the RV770 architecture is really efficient at computing (?) - a GPU is good... for computing ?
shinra-bansho
Member
Don't know. Would also be keen to know - as it helps fill in the blanks.
The overarching question is: Does the 40W figure preclude the 600 GFLOP number that's been floated.
Does this imply they can or can't have a 600 GFLOP GPU...?
Brad Grenz
Member
75 watts is the max the power supply is rated for. Nintendo says typical usage is around 45 watts in game.
With even 30 full watts just for the GPU the WiiU's GPU would have to be nearly as efficient as AMD's latest GCN architecture manufactured at 28nm to reach 600GFlops. Given that I think it's unlikely that A) it will be manufactured at 28nm, B) its power usage will be more than about 20 watts, and C) that whatever they've done to the R700 is enough to nearly double its performance per watt I think it's pretty safe to discount 600GFlops as a realistic expectation.
shinra-bansho
Member
At 20W, at something similar to the E6760's 16.5 GFLOPs/W, that would put "GPU7" at like 330 GFLOPs...
Does anyone have equivalent figures for the RSX and Xenos?
I may be jumping to conclusion like an idiot who never learns patience but in my head it's each time more a certainty that Wii U is a Wii 2. 6x in performance gap compared to what is rumored for X360 and PS4 is really big (assuming the FLOPS count on both sides is comparable). I will never get ports from other consoles.
edit: I really need to see how the engraving process is related to energy efficiency, is there a linear/quadratic/exponential relation.
edit: I really need to see how the engraving process is related to energy efficiency, is there a linear/quadratic/exponential relation.
JohnnySasaki86
Member
its probably about 450-500glfops like originally rumored way back when(a few months ago around e3). Then the second rumor came out that said it was between 500-600gflops with it being closer to the 600gflop number. Then people seemed to just start assuming it was 600gflops somehow. Nothing was ever really credible and concrete along the way.
there around 200-250 afaik.
What is the best source for this? I googled for a few minutes and some sources say "45 watts typical power usage, 75 watts max", while others say something like "45 watts idle, 75 watts under load".
shinra-bansho
Member
OK, tx. If that is a full-load figure then 600 GFlops looks a bit unlikely indeed. Unless it's a mobile GPU.
flipswitch
Member
Do we know what type of screen the wii u game pad uses, eg ips etc.
One more thing, does the analog buttons click down, like how you sprint in COD?
One more thing, does the analog buttons click down, like how you sprint in COD?
Brad Grenz
Member
Xenon is about 100GFlops. Cell is around 200GFlops.
walking fiend
Member
Can some1 explain how laptop components has a much lower power consumption than desktop counterparts?
WUT.
Wii's processor was 2.9 GFLOPS according to Wikipedia. Surely, Xenon is not 30 times as quick even though there were three cores ?
The world of tech is more and more fascinating by the minute.
Peak GFLOPS is only a good measure of comparison within the same architecture. It's better than clock rate or number of cores, but doesn't always allow comparisons between different architectures.
All about the flops again
Just checking in ...
Thanks for your contribution.
Brad Grenz
Member
45 watts is the WiiU under load.
Wii's CPU was an overclocked version of the Gamecube CPU. It was single core and ran at something like 720 mhz. Xenon is 3 cores at 3.2 ghz, each with a really big vector unit for floating point calculations. Xenon's transistor and thermal budget was also massively larger than the Wii's.
Wii's CPU was an overclocked version of the Gamecube CPU. It was single core and ran at something like 720 mhz. Xenon is 3 cores at 3.2 ghz, each with a really big vector unit for floating point calculations. Xenon's transistor and thermal budget was also massively larger than the Wii's.
That is how i see it. And i also want to place a big question mark next to the 40 W statement of Iwata. How i see it, and how i think it is most sensible for him as head of Nintendo to share that, is in the sense that less = better. He already knows 40 or 50 or even 75 isn't going to persuade the tech analysts that it is superpowerful hardware. What he can do though, is come across as environmental friendly, cost efficient, energy saving etc... So IMO, he is likely to give a low estimate rather than a high. Also, if in the same line of thought, he is speaking of average consumption, i would guess he includes the amount of time you are in the WiiU menu, on MiiVerse, in the game menu etc... Maybe for each hour on the WIiU you are only stressing the GPU/CPU for 20 to 30 minutes.
So i think we are comparing low estimates (power consumption of WiiU) to high estimates (power consumption of GPU/CPU). This is of course pure speculation on my part, but it could explain for 5 to 10 W.
shinra-bansho
Member
The 40W figure appears to be in reference to during gaming, and dependent on what peripherals are attached.
tkscz
Member
My problem with this is that everyone keeps mentioning DX10.1, rather than saying OpenGL 3.3. Why is this bothering me? Well besides the obvious, Direct X is a Microsoft copyright and wouldn't be used on a Nintendo console anyway, OpenGL 3.3 backports from OpenGL 4. While I can't say if DX10.1 does this, I know that OpenGL 3.3 supports backporting, allowing it to "port" OpenGL 4 code down to weaker hardware, or as much as the weaker hardware can handle. If it were just that, then it would allow for a good enough amount of OpenGL 4 (DX11) features, but like you and ideaman have said many times, it has been tweaked and messed with many times, to the point there is no actual comparison for it with an off the shelf graphics card. So it could support more, or Nintendo might have been banking on that backporting.
How does this "appear"?
lightchris
Member
Just to make this clear: No, not even closely, at least not in relevant game code. Broadway has OoOE and a short pipeline, overall resulting in significantly more IPC. Three Broadway cores @ 1 GHz should already be enough to compete with Xenon at quite some tasks.
Imo: Forget the FLOPS when you are comparing differenct architectures. It just confuses people and leads to wrong conclusions. You can't put CPU performance into a single number.
I wouldnt say that's 100% accurate....
From chesemeisters Japanese translation:
Iwata: "Depending on the game being played and the accessories connected, roughly 40 watts of electrical consumption could be considered realistic."
And remember Nintendo are trying to market this as energy efficient. They could have taken this figure whilst playing a virtual console nes game for all we know!
In a console, where the CPU can freely access GPU data without going through a PCI-Express bus and buffers can be cast into different types without going through conversion, a DX10.1/openGL 3.3/SM4.1 GPU can be very useful at non-graphics tasks. I believe Havok's GPU acellerated (whichinclude cloth, hair and fluids) would work fine on it.
MadeInBeats
Banned
Elecy bills have been a big talking point in the UK the last couple of years. I think it could be a USP with some value here when referred to alongside any potential next MS/Sony machine. I'm no Greenpeace warrior, but I do love t buy energy efficient devices and save where ever I can.
Flops are a very poor way to compare the performance of different CPU's.
As an example the i7 920 is theoretically capable of 42.5Gflops while Xenon is theoretically capable of 100Gflops. But in reality the i7 920 is several times better than Xenon.
It's the downside of the 360 for sure. Unplug that shit when you're done with it, otherwise the PS3 and 360 have a notable energy draw even when they're idling.
Alright, explanations that indicate that the situation is more obscure and complex than imagined. I am out of the thread. Thank you guys to let me understand the topic a bit better.
edit: I edited your edit Donnie. Thanks to clarify things a bit more in detail.