By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Nintendo - Wii U GPU new info and more speculation

ninjablade on 11 February 2013
ethomaz said:
Aielyn said:
ethomaz said:
There is no magical or miracle... the GPU is weak... just accept that.

Do you have an alternate theory as to what that 30% of the chip is doing?

No? Then don't try to make out as though you know more than you do. And don't bother going "Hollywood", because at most, that would take up maybe 10% of the chip, and you'll probably find that part of the known 70% includes some of it already.

You understand anything about GPU arch? No. Then be quite.

The Wii U's GPU is not that customized like some guys here says... the Wii U's GPU have a space for eDRAM, other for the GPU ifself, and another unknown. The unknown part is not GPU shaders, it's anyting else like some fixed function or even some part to Wii's BC... because the Wii U needs that for some purpose.

The Wii U's GPU part is knewn already and it have at maximum 320SPs... the other 30% part can free up or help the GPU/CPU with the workload but that part don't add raw power to the GPU... you guy thinks there are "special sauce", "magical" or "unicorn" below Wii U's GPU... but the really is another.

Wii U GPU is weak... about 50% better than PS360 GPU... just face it.

PS. I don't need to enter in the fact that some GPU's needs black space (or unused) due production design.

edit please, anyway i agree with, one of the few people that actually knows what' he talking about.



Around the Network
RazorDragon on 11 February 2013
ninjablade said:
RazorDragon said:
ninjablade said:

Why talk about something, when you don't understand it,  what you stated goes against evrything i read from tech head's at neogaf and beyond3d. at the moment the block that hold the sp's are not even big enough to hold 40sp which is why everybody is confused.

 

 

As HoloDust said:

480:24:8 @40nm GPU is 118mm^2

400:20:8 @ 40nm GPU is 104mm^2

 

And TSMC's 40nm eDRAM should give 37mm^2 (got that from http://forum.beyond3d.com/showthread.php?p=1680372 )

So that would give us a final die size of 155mm^2, which is a little smaller than the 156.21mm^2 die size of Wii U's GPU. Introducing design overheads/redundancies for I/O and also increasing yield, it would get us in the 156.21mm^2 range. Of course, that was all speculation before the photo from Chipworks came, but it was all thought by NeoGAF and Beyond3D. I guess you didn't read those topics enough, because what i said was in the expectations of NeoGaf and Beyond3D users.

http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532

http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533

Read and understand. the link you provided was 3 months ago, the descussion has since progressed since we have GPU pics.


That was the intention. Still, it doesn't make any sense to have 160SPs on a 156mm^2 die size. There would be so much wasted space in that GPU and that simply wouldn't happen, except if Nintendo somehow wanted to lose money creating empty space on the die which would magically explain everything. It's still not known the fabrication process on Wii U's GPU. You can't predict that the SIMD blocks are less dense or denser than Llano or whatever GPU they're comparing, because depending on your GPU design you can change the density for more or less without modifying the fabrication process. Even in that topic some Beyond3D users don't agree with that conclusion you posted:

http://beyond3d.com/showpost.php?p=1703605&postcount=4535

 

For example, removing the double precision extensions can improve density without modifying fabrication processes. As this user said:

http://beyond3d.com/showpost.php?p=1703684&postcount=4544

And looking over the next few pages, nobody posted Redwood's DP-less die shots for comparison, so it's actually not known if the removal from DP extensions would make the Wii U be able to hold 40SPs per block or not.

And, as i said, knowing the eDram die on Wii U takes 38.68mm², which allows for the GPU itself to take a whole 118mm^2, the minimum you can expect from a die with this side are 320 SPUs, even if the GPU was made in a 55nm process, there would still be a lot of space left on the die if there were only 160SPs, for example.



ninjablade on 11 February 2013
ethomaz said:
Aielyn said:
ethomaz said:
There is no magical or miracle... the GPU is weak... just accept that.

Do you have an alternate theory as to what that 30% of the chip is doing?

No? Then don't try to make out as though you know more than you do. And don't bother going "Hollywood", because at most, that would take up maybe 10% of the chip, and you'll probably find that part of the known 70% includes some of it already.

You understand anything about GPU arch? No. Then sshut up.

The Wii U's GPU is not that customized like some guys here says... the Wii U's GPU have a space for eDRAM, other for the GPU ifself, and another unknown. The unknown part is not GPU shaders, it's anyting else like some fixed function or even some part to Wii's BC... because the Wii U needs that for some purpose.

The Wii U's GPU part is knewn already and it have at maximum 320SPs... the other 30% part can free up or help the GPU/CPU with the workload but that part don't add raw power to the GPU... you guy thinks there are "special sauce", "magical" or "unicorn" below Wii U's GPU... but the really is another.

Wii U GPU is weak... about 50% better than PS360 GPU... just face it.

PS. I don't need to enter in the fact that some GPU's needs black space (or unused) due production design.

and that's the best case scenerio, many at beyond3d are doubting it, even has that at the moment, then you have a weaker cpu and 50% slower bandwidth, i don't see how anybody can find that interesting.



ethomaz on 11 February 2013

ninjablade said:

edit please, anyway i agree with, one of the few people that actually knows what' he talking about.

I overdid a bit and apologize @Aielyn.



ninjablade on 11 February 2013
RazorDragon said:
ninjablade said:
RazorDragon said:
ninjablade said:

Why talk about something, when you don't understand it,  what you stated goes against evrything i read from tech head's at neogaf and beyond3d. at the moment the block that hold the sp's are not even big enough to hold 40sp which is why everybody is confused.

 

 

As HoloDust said:

480:24:8 @40nm GPU is 118mm^2

400:20:8 @ 40nm GPU is 104mm^2

 

And TSMC's 40nm eDRAM should give 37mm^2 (got that from http://forum.beyond3d.com/showthread.php?p=1680372 )

So that would give us a final die size of 155mm^2, which is a little smaller than the 156.21mm^2 die size of Wii U's GPU. Introducing design overheads/redundancies for I/O and also increasing yield, it would get us in the 156.21mm^2 range. Of course, that was all speculation before the photo from Chipworks came, but it was all thought by NeoGAF and Beyond3D. I guess you didn't read those topics enough, because what i said was in the expectations of NeoGaf and Beyond3D users.

http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532

http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533

Read and understand. the link you provided was 3 months ago, the descussion has since progressed since we have GPU pics.


That was the intention. Still, it doesn't make any sense to have 160SPs on a 156mm^2 die size. There would be so much wasted space in that GPU and that simply wouldn't happen, except if Nintendo somehow wanted to lose money creating empty space on the die which would magically explain everything. It's still not known the fabrication process on Wii U's GPU. You can't predict that the SIMD blocks are less dense or denser than Llano or whatever GPU they're comparing, because depending on your GPU design you can change the density for more or less without modifying the fabrication process. Even in that topic some Beyond3D users don't agree with that conclusion you posted:

http://beyond3d.com/showpost.php?p=1703605&postcount=4535

 

For example, removing the double precision extensions can improve density without modifying fabrication processes. As this user said:

http://beyond3d.com/showpost.php?p=1703684&postcount=4544

And looking over the next few pages, nobody posted Redwood's DP-less die shots for comparison, so it's actually not known if the removal from DP extensions would make the Wii U be able to hold 40SPs per block or not.

And, as i said, knowing the eDram die on Wii U takes 38.68mm², which allows for the GPU itself to take a whole 118mm^2, the minimum you can expect from a die with this side are 320 SPUs, even if the GPU was made in a 55nm process, there would still be a lot of space left on the die if there were only 160SPs, for example.

actually that maximum is 320 SP, if you read the last page on beyond3d many think its below 320 SP, the perefromance is not amking sense when even the most bottled necked system the pc, can blow past 360 with an HD5500 clocked at 550mhz.



Around the Network
RazorDragon on 11 February 2013
ninjablade said:
RazorDragon said:
ninjablade said:
RazorDragon said:
ninjablade said:

Why talk about something, when you don't understand it,  what you stated goes against evrything i read from tech head's at neogaf and beyond3d. at the moment the block that hold the sp's are not even big enough to hold 40sp which is why everybody is confused.

 

 

As HoloDust said:

480:24:8 @40nm GPU is 118mm^2

400:20:8 @ 40nm GPU is 104mm^2

 

And TSMC's 40nm eDRAM should give 37mm^2 (got that from http://forum.beyond3d.com/showthread.php?p=1680372 )

So that would give us a final die size of 155mm^2, which is a little smaller than the 156.21mm^2 die size of Wii U's GPU. Introducing design overheads/redundancies for I/O and also increasing yield, it would get us in the 156.21mm^2 range. Of course, that was all speculation before the photo from Chipworks came, but it was all thought by NeoGAF and Beyond3D. I guess you didn't read those topics enough, because what i said was in the expectations of NeoGaf and Beyond3D users.

http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532

http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533

Read and understand. the link you provided was 3 months ago, the descussion has since progressed since we have GPU pics.


That was the intention. Still, it doesn't make any sense to have 160SPs on a 156mm^2 die size. There would be so much wasted space in that GPU and that simply wouldn't happen, except if Nintendo somehow wanted to lose money creating empty space on the die which would magically explain everything. It's still not known the fabrication process on Wii U's GPU. You can't predict that the SIMD blocks are less dense or denser than Llano or whatever GPU they're comparing, because depending on your GPU design you can change the density for more or less without modifying the fabrication process. Even in that topic some Beyond3D users don't agree with that conclusion you posted:

http://beyond3d.com/showpost.php?p=1703605&postcount=4535

 

For example, removing the double precision extensions can improve density without modifying fabrication processes. As this user said:

http://beyond3d.com/showpost.php?p=1703684&postcount=4544

And looking over the next few pages, nobody posted Redwood's DP-less die shots for comparison, so it's actually not known if the removal from DP extensions would make the Wii U be able to hold 40SPs per block or not.

And, as i said, knowing the eDram die on Wii U takes 38.68mm², which allows for the GPU itself to take a whole 118mm^2, the minimum you can expect from a die with this side are 320 SPUs, even if the GPU was made in a 55nm process, there would still be a lot of space left on the die if there were only 160SPs, for example.

actually that maximum is 320 SP, if you read the last page on beyond3d many think its below 320 SP, the perefromance is not amking sense when even the most bottled necked system the pc, can blow past 360 with an HD5500 cloed at 550mhz.


Agreed. And i know the maximum is 320 SPs in Wii U's case, however PC cards like the HD 4770 with a die size of 118mm^2 had 480 SPs, which doesn't happen on the Wii U, of course, because of different achitectures. Anyway, Wii U's  case can't be analyzed by only looking at it performance-wise.  Games on the market now are launch titles, and the system may have bottlenecks not known by us and developers that go beyond the SP count of the GPU.



ninjablade on 11 February 2013
RazorDragon said:
ninjablade said:
RazorDragon said:
ninjablade said:
RazorDragon said:
ninjablade said:

Why talk about something, when you don't understand it,  what you stated goes against evrything i read from tech head's at neogaf and beyond3d. at the moment the block that hold the sp's are not even big enough to hold 40sp which is why everybody is confused.

 

 

As HoloDust said:

480:24:8 @40nm GPU is 118mm^2

400:20:8 @ 40nm GPU is 104mm^2

 

And TSMC's 40nm eDRAM should give 37mm^2 (got that from http://forum.beyond3d.com/showthread.php?p=1680372 )

So that would give us a final die size of 155mm^2, which is a little smaller than the 156.21mm^2 die size of Wii U's GPU. Introducing design overheads/redundancies for I/O and also increasing yield, it would get us in the 156.21mm^2 range. Of course, that was all speculation before the photo from Chipworks came, but it was all thought by NeoGAF and Beyond3D. I guess you didn't read those topics enough, because what i said was in the expectations of NeoGaf and Beyond3D users.

http://forum.beyond3d.com/showpost.php?p=1703471&postcount=4532

http://forum.beyond3d.com/showpost.php?p=1703534&postcount=4533

Read and understand. the link you provided was 3 months ago, the descussion has since progressed since we have GPU pics.


That was the intention. Still, it doesn't make any sense to have 160SPs on a 156mm^2 die size. There would be so much wasted space in that GPU and that simply wouldn't happen, except if Nintendo somehow wanted to lose money creating empty space on the die which would magically explain everything. It's still not known the fabrication process on Wii U's GPU. You can't predict that the SIMD blocks are less dense or denser than Llano or whatever GPU they're comparing, because depending on your GPU design you can change the density for more or less without modifying the fabrication process. Even in that topic some Beyond3D users don't agree with that conclusion you posted:

http://beyond3d.com/showpost.php?p=1703605&postcount=4535

 

For example, removing the double precision extensions can improve density without modifying fabrication processes. As this user said:

http://beyond3d.com/showpost.php?p=1703684&postcount=4544

And looking over the next few pages, nobody posted Redwood's DP-less die shots for comparison, so it's actually not known if the removal from DP extensions would make the Wii U be able to hold 40SPs per block or not.

And, as i said, knowing the eDram die on Wii U takes 38.68mm², which allows for the GPU itself to take a whole 118mm^2, the minimum you can expect from a die with this side are 320 SPUs, even if the GPU was made in a 55nm process, there would still be a lot of space left on the die if there were only 160SPs, for example.

actually that maximum is 320 SP, if you read the last page on beyond3d many think its below 320 SP, the perefromance is not amking sense when even the most bottled necked system the pc, can blow past 360 with an HD5500 cloed at 550mhz.


Agreed. And i know the maximum is 320 SPs in Wii U's case, however PC cards like the HD 4770 with a die size of 118mm^2 had 480 SPs, which doesn't happen on the Wii U, of course, because of different achitectures. Anyway, Wii U's  case can't be analyzed by only looking at it performance-wise.  Games on the market now are launch titles, and the system may have bottlenecks not known by us and developers that go beyond the SP count of the GPU.

yes that's possible, it's just seems wierd to me it had such an advantage in GPU power they would have to lower the resolution on the wii u version of sonic racing compared to 360/ps3, the samething for COD blops 2, i mean bad frame rate is understandable but the least you can do is up the resolution if they had 320 sp and in sonic case they had to lower it but i agree with you for the most part.



Aielyn on 11 February 2013
ethomaz said:
You understand anything about GPU arch? No.  Then be quite.

I don't need to know much about GPU architecture to know that, when people who DO know what they're talking about can't identify 30% of the chip, it is presumptive to assume that the part they do understand is all that matters.

What if one of the sections of the GPU is actually some new logic that is going to find its way into new GPU chips in the future, with AMD putting it into the Wii U as an early use of the logic? I'm not saying that's what has happened, but the fact that we don't know is the point - until the chip is actually understood fully, we shouldn't be making assumptions. There's a big difference between suggesting that people not get their hopes up, and making absolute statements like the ones you've made.



ninjablade on 11 February 2013
Aielyn said:
ethomaz said:
You understand anything about GPU arch? No.  Then be quite.

I don't need to know much about GPU architecture to know that, when people who DO know what they're talking about can't identify 30% of the chip, it is presumptive to assume that the part they do understand is all that matters.

What if one of the sections of the GPU is actually some new logic that is going to find its way into new GPU chips in the future, with AMD putting it into the Wii U as an early use of the logic? I'm not saying that's what has happened, but the fact that we don't know is the point - until the chip is actually understood fully, we shouldn't be making assumptions. There's a big difference between suggesting that people not get their hopes up, and making absolute statements like the ones you've made.

we undertsand enough to know, the fact is the gpu si weak even in its best case scenerio, you think nintendo after using such a weak CPU and crippled memory bandwidth put some alien tech in the gpu, if you know anything about gpu's then you would know even at 320SP the GPU is gimped by the memory band banndwidth already, they are over doing with 320 SP as it is.



Aielyn on 11 February 2013
ninjablade said:
yes that's possible, it's just seems wierd to me it had such an advantage in GPU power they would have to lower the resolution on the wii u version of sonic racing compared to 360/ps3, the samething for COD blops 2, i mean bad frame rate is understandable but the least you can do is up the resolution if they had 320 sp and in sonic case they had to lower it but i agree with you for the most part.

You're forgetting to factor in the Upad screen.

Assuming the Digital Foundry claim regarding Sonic Racing resolutions are correct, the pixels made on the three systems for TV only are:

Wii U: 1024x576 =  589,824
PS3: 1280x720 = 921,600
360: 1152x544 = 626,688

Right? But the Wii U also has to generate the Upad screen. That's another 854x480 = 409,920, bringing the Wii U up to 999,744 pixels. And it has been noted that the Wii U FPS is more stable than the PS3 FPS (although the 360 does best).