• 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)

USC-fan

Banned
If you have a large amount of embedded memory then there are plenty of ways to limit writes to main memory. I really don't know why you're trying to trivialize that fact.

Sure if you have large amounts but you only have 32mb.

So the console has been out for two months now, and still no updates? Ugh. I guess the guy who revealed the CPU clockspeed, hasn't found any new revelations either.
Because it really doesnt matter. We have enough info to lead to a conclusion. There is no secret that going to change anything. Wii-u is what it is, like it or not...
 

Donnie

Member
Sure if you have large amounts but you only have 32mb.

Are we now going to argue for 2 pages about the arbitrary definition of "a large amount"?.. Fact is WiiU's embedded memory is a large amount in comparison to 360's and therefore can be used to limit reads and/or writes much more so than 360's eDRAM could. That was the point of my post.
 

USC-fan

Banned
Are we now going to argue for 2 pages about the arbitrary definition of "a large amount"? Please bring me a worthwhile discussion..

At least that would give us something to talk about in here...

Fact is WiiU's embedded memory is a large amount in comparison to 360's and therefore can be used to limit writes and/or writes much more so than 360's eDRAM could. That was the point of my post.
360 edram doesnt need to do that. Unlike the wiiu there is no advantage to read speed by limiting write as you said. This could be one of the ways to make up for the lack of BW on wiiu.

I think this is the way it was design in fact. It was a way of cost cutting and not for performance when going with edram.
 

Donnie

Member
Because it really doesnt matter. We have enough info to lead to a conclusion. There is no secret that going to change anything. Wii-u is what it is, like it or not...

The public finding out specs doesn't change any console, never has, nor is that the point of knowing those details. Everyone knows you've come to a conclusion and shock horror, a lot of people don't agree with you and are interested in actual facts they don't currently know.
 

Donnie

Member
360 edram doesnt need to do that. Unlike the wiiu there is no advantage to read speed by limiting write as you said. This could be one of the ways to make up for the lack of BW on wiiu.


Didn't say 360 needed to limit writes to main memory in order to save read bandwidth, obviously that's not how its main memory works. Also I meant "reads and/or writes".

I think this is the way it was design in fact. It was a way of cost cutting and not for performance when going with edram.

I'll wait until I know the details on the eDRAM (such as bandwidth and accessibility) before I come to that kind of conclusion.
 

ikioi

Banned
In regards to memory bandwidth etc:

A few developers and i believe Nintendo themselves have said that the Wii U's main memory pool has significantly lower latency then the Xbox 360 and PS3's.

The Wii U also has over 2x the dedicated main memory pool the Xbox 360 and PS3. 1GB of RAM for the Wii U vs less then 512MB for Xbox 360 and PS3. It's also a shared memory pool unlike the 256/256 split on the PS3, and none of that 1GB is consumed by the background opperating system.

The Wii U also has significantly larger cache on the CPU, and increased eDRAM vs Xbox 360.

The Wii U's CPU also has a very very short pipe line

The GPU and CPU also are packaged within the same MCM, so latency between the two should be quite low, and bandwith should be very high.

Is it not possible that with the combination of all of the above factors, developers can overcome the limited bandwidth of the Wii U's main ram pool?

Also i was under the impression that the Xbox 360's bandwidth of > 20Gbps was largely bullshit. The Xbox 360 writes and reads constantly to its memory pool which more then halfs the bandwidth when overheads are taken into account. The figures of it being capable of over 20Gbps was in theoretical situations, and in real world factoring in writes and reads its bandwidth is significantly less.

From where i sit it sounds to me like not enough is known really about the Wii U's architecture to really judge its memory bandwidth limitaitons. Its memory management appears to be significantly different in approach to the Xbox 360's and PS3s.
 

Hiltz

Member
Developer Shin'en (maker of Nano Assault Neo) commented again on their Wii U development:

The Wii U development environment allowed us to go from zero to a great, polished game in only half a year. I think we never ever got so far so fast on a new hardware, and we’ve worked on a lot of platforms in the past decades.

We decided to add a local two-player mode where one player uses the TV and the other one the Wii U GamePad. That meant the complete game world would need to be rendered twice for the Wii U GamePad display. That nearly doubled the burden on the GPU and CPU, but even then we were still able to maintain solid 60FPS without much effort on both displays. Then on top we even added camera streaming, so that the TV player can see the face of the Wii U GamePad player, which adds a lot of fun. We expected the additional camera encoding and streaming would add noticeable strain on the Wii U, but it was almost free. All this proved the system is very well balanced and allows developers without much effort, to use all the unique features. Another benefit of the Wii U is that the system memory is so large that we never have to load an asset again. Everything can be cached and load times are almost gone. - Shin’en’s Manfred Linzner

http://nintendoeverything.com/10999...ev-environment-overall-tech-and-large-memory/
 

USC-fan

Banned
The public finding out specs doesn't change any console, never has, nor is that the point of knowing those details. Everyone knows you've come to a conclusion and shock horror, a lot of people don't agree with you and are interested in actual facts they don't currently know.

Sure and most of those people dont post any more. hmmmm...

Yeah i'm the bad guy....
 

USC-fan

Banned
Well you are... but that's just because I'm looking at your internet history right now.

You need people like me. You need people like me so you can point your fuckin' fingers and say, "That's the bad guy." So... what that make you? Good? You're not good. You just know how to hide, how to lie. Me, I don't have that problem. Me, I always tell the truth.
 
You need people like me. You need people like me so you can point your fuckin' fingers and say, "That's the bad guy." So... what that make you? Good? You're not good. You just know how to hide, how to lie. Me, I don't have that problem. Me, I always tell the truth.

Fuckin' A-right!
 
If you have a large amount of embedded memory then there are plenty of ways to limit writes to main memory. I really don't know why you're trying to trivialize that fact. Or are you referring specifically to my example of being able to get more read bandwidth out of WiiU's memory than 360's? Obviously that would take some very tight optimisations and probably wouldn't be possible in most cases, I did say technically.

If we're talking about really tiny scenes using last-gen assets, then sure, man. Meanwhile we've seen single textures here hit >1MB this gen.
 

Oblivion

Fetishing muscular manly men in skintight hosery
Also storing certain frequently used textures in eDRAM isn't the only option to optimise WiiU's main memory bandwidth. WiiU's main memory has 12.8GB/s that can be used for reads and/or writes. While 360 has 11.2GB/s for reads only and 11.2GB/s for writes only. If you use WiiU's eDRAM to limit writes to main memory to a very small amount you could technically have more read bandwidth from WiiU's main memory than 360's.

Also i was under the impression that the Xbox 360's bandwidth of > 20Gbps was largely bullshit. The Xbox 360 writes and reads constantly to its memory pool which more then halfs the bandwidth when overheads are taken into account. The figures of it being capable of over 20Gbps was in theoretical situations, and in real world factoring in writes and reads its bandwidth is significantly less.

From where i sit it sounds to me like not enough is known really about the Wii U's architecture to really judge its memory bandwidth limitaitons. Its memory management appears to be significantly different in approach to the Xbox 360's and PS3s.

If the 360's bandwidth was split that way, why has no one mentioned it until now?

Seems like a pretty important detail for everyone to overlook.
 

wsippel

Banned
If we're talking about really tiny scenes using last-gen assets, then sure, man. Meanwhile we've seen single textures here hit >1MB this gen.
Why would you store textures in the eDRAM to begin with? Texturing doesn't require much bandwidth. The eDRAM is better suited for render targets and display lists.
 

Oblivion

Fetishing muscular manly men in skintight hosery
It has been mentioned. People were talking about it in this thread around the end of November.

Thought I was following the thread pretty closely during the time, but I guess not.

So does this mean the bandwidth gap isn't as huge as a lot of people thought?
 

Popstar

Member
Thought I was following the thread pretty closely during the time, but I guess not.

So does this mean the bandwidth gap isn't as huge as a lot of people thought?
I think it was around the time the thread got invaded with animated gif console warz so you might have missed it in the noise.

But yeah, it means there probably isn't any significant bandwidth disadvantage.
 

AzaK

Member
In regards to memory bandwidth etc:

A few developers and i believe Nintendo themselves have said that the Wii U's main memory pool has significantly lower latency then the Xbox 360 and PS3's.

The Wii U also has over 2x the dedicated main memory pool the Xbox 360 and PS3. 1GB of RAM for the Wii U vs less then 512MB for Xbox 360 and PS3. It's also a shared memory pool unlike the 256/256 split on the PS3, and none of that 1GB is consumed by the background opperating system.

The Wii U also has significantly larger cache on the CPU, and increased eDRAM vs Xbox 360.

The Wii U's CPU also has a very very short pipe line

The GPU and CPU also are packaged within the same MCM, so latency between the two should be quite low, and bandwith should be very high.

Is it not possible that with the combination of all of the above factors, developers can overcome the limited bandwidth of the Wii U's main ram pool?

Also i was under the impression that the Xbox 360's bandwidth of > 20Gbps was largely bullshit. The Xbox 360 writes and reads constantly to its memory pool which more then halfs the bandwidth when overheads are taken into account. The figures of it being capable of over 20Gbps was in theoretical situations, and in real world factoring in writes and reads its bandwidth is significantly less.

From where i sit it sounds to me like not enough is known really about the Wii U's architecture to really judge its memory bandwidth limitaitons. Its memory management appears to be significantly different in approach to the Xbox 360's and PS3s.
But it has twice+ the BW of Wii U so overall it's as good if not better. It doesn't have bidirectional RAM as I've been told.
 

Donnie

Member
If we're talking about really tiny scenes using last-gen assets, then sure, man. Meanwhile we've seen single textures here hit >1MB this gen.

What's that got to do with using eDRAM to limit writes to main memory though? I didn't suggest storing textures in eDRAM, quite the opposite.

But it has twice+ the BW of Wii U so overall it's as good if not better. It doesn't have bidirectional RAM as I've been told.

What are you referring to here?, 360?, PS3?
 

tipoo

Banned
Those are just two examples from Nintendo's documentation. 32MB is more than enough to hold a 1080p frame and z buffer with space to spare (you'd still have about half the eDRAM free).

Also storing certain frequently used textures in eDRAM isn't the only option to optimise WiiU's main memory bandwidth. WiiU's main memory has 12.8GB/s that can be used for reads and/or writes. While 360 has 11.2GB/s for reads only and 11.2GB/s for writes only.


I'm confused. So the 360 memory can do 11.2GB/s upstream and downstream, not two streams of either, is that what you're saying? And the Wii U can its 12.8GB/s channel for either operation? Why is that advantageous to the Wii U then? It's only a little more read bandwidth, and if it needs to write that bandwidth is taken away from.

Actually I think I just talked myself out of my confusion, you're just saying it's not like the 360 can read in 22GB/s gobs, right? The U can't write as much at the same time as reading, but since there's also more of it maybe that's moot. The 360s memory is better suited to streaming things in and out as needed though.

I wonder how that balances out, which one is more advantageous? You could be reading while writing at the same speed to the 360, while the Wii U could just load a bigger level all at once and rely mostly on reads.
 

Panajev2001a

GAF's Pleasant Genius
I'm confused. So the 360 memory can do 11.2GB/s upstream and downstream, not two streams of either, is that what you're saying? And the Wii U can its 12.8GB/s channel for either operation? Why is that advantageous to the Wii U then? It's only a little more read bandwidth, and if it needs to write that bandwidth is taken away from.

Actually I think I just talked myself out of my confusion, you're just saying it's not like the 360 can read in 22GB/s gobs, right? The U can't write as much at the same time as reading, but since there's also more of it maybe that's moot. The 360s memory is better suited to streaming things in and out as needed though.

I wonder how that balances out, which one is more advantageous? You could be reading while writing at the same speed to the 360, while the Wii U could just load a bigger level all at once and rely mostly on reads.

what he meant is that developers writing Xbox 360 games can use the full 22 GB/s of bandwidth if they carefully batch reads and writes in such a way that they maximize the bus's bandwidth since it does not have a generic 22 GB/s channel but two one-way channels with 11 GB/s each.

Wii U only has a single channel with 12.8 GB/s so, on one side you get to use a bit more bandwidth than some applications do on Xbox 360 (it is not easy to batch your reads and writes and to issue them concurrently to use both channels fully), but you would get to use less bus bandwidth than what some games do on Xbox 360.
 

Donnie

Member
Panajev2001a

Good summary, thanks :)

I'm confused. So the 360 memory can do 11.2GB/s upstream and downstream, not two streams of either, is that what you're saying? And the Wii U can its 12.8GB/s channel for either operation? Why is that advantageous to the Wii U then? It's only a little more read bandwidth, and if it needs to write that bandwidth is taken away from.

Actually I think I just talked myself out of my confusion, you're just saying it's not like the 360 can read in 22GB/s gobs, right? It can't write as much at the same time as reading, but since there's also more of it maybe that's moot. Seems like the 360s memory is better suited to streaming things in and out as needed though. Seems like the two would balance out in the end and there isn't a great deal of advantage moving to the more limited 1GB, but I'm no developer so who knows.

Its not a case of one type of memory being better (though 1GB vs 512MB is certainly better), just more suited to a specific system. WiiU has a lot of embedded memory (on both the GPU and CPU) and all chips are very closely connected. Its clearly designed to hit main memory much less so than 360. That's not to say more bandwidth wouldn't have been good, of course it would, but I think the type of memory bus used was the right choice.
 
What's that got to do with using eDRAM to limit writes to main memory though? I didn't suggest storing textures in eDRAM, quite the opposite.



What are you referring to here?, 360?, PS3?

'limit writes to main memory' is ambiguous. and even so when there's a cause/effect on screen are you saying that artists should make smaller effect maps so that theyre tiny enough to fit? Shadow maps are already ridiculously hard on writes and reads. They seem to be of a much lower resolution and a much lower colorspace, and they also tend to take up quite a lot of space in a scene. You don't find it plausible that the small, tiny space left in the EDRam is already occupied? Even if they're hideous compared to other consoles the games would look so much worse without them.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
If the 360's bandwidth was split that way, why has no one mentioned it until now?

Seems like a pretty important detail for everyone to overlook.
RAM BW is split for Xenon (dedicated up and down). It's not split for Xenos, AFAIK. We don't know what the RAM bus setup is with U-CPU - it might or might not be split. Actually, there's little we know about the entire machine, perhaps the CPU being the most 'exposed' part so far, if anything in there is.
 

Datschge

Member
Compressed?
Nintendo recently partnered with a company for some texture compression. Typically 4:1 compression IIRCy. S3 is the format I think.

Actually that premièred in Gamecube's Flipper already, and I don't think it was taken out since. =P
 
Developer Shin'en (maker of Nano Assault Neo) commented again on their Wii U development:


http://nintendoeverything.com/10999...ev-environment-overall-tech-and-large-memory/

The game isn't honestly that impressive. There is also a problem with co-op mode and having your gamepad at <50% charge. I never saw this on any other game we played, but my friend was using the gamepad and the screen would black out for a split second, and it would continue him in the same direction he was holding. He couldn't disconnect and reconnect the controller or anything to fix it.
 

AzaK

Member
What's that got to do with using eDRAM to limit writes to main memory though? I didn't suggest storing textures in eDRAM, quite the opposite.

What are you referring to here?, 360?, PS3?
360


what he meant is that developers writing Xbox 360 games can use the full 22 GB/s of bandwidth if they carefully batch reads and writes in such a way that they maximize the bus's bandwidth since it does not have a generic 22 GB/s channel but two one-way channels with 11 GB/s each.

Wii U only has a single channel with 12.8 GB/s so, on one side you get to use a bit more bandwidth than some applications do on Xbox 360 (it is not easy to batch your reads and writes and to issue them concurrently to use both channels fully), but you would get to use less bus bandwidth than what some games do on Xbox 360.

RAM BW is split for Xenon (dedicated up and down). It's not split for Xenos, AFAIK. We don't know what the RAM bus setup is with U-CPU - it might or might not be split. Actually, there's little we know about the entire machine, perhaps the CPU being the most 'exposed' part so far, if anything in there is.

This is what I was told from Beyond3D

No, 360, like any other system using standard commodity DRAMs cannot both read and write to main memory at the same time. Don't get confused by the CPU-to-GPU bus having separate read and write lines, that's just a convenience to avoid having bus turnaround penalties slapping you and increasing what is already quite high main memory latency (from the CPU's standpoint, that is.)

Which says that it's just one way. Who's right?

Actually that premièred in Gamecube's Flipper already, and I don't think it was taken out since. =P

S3TC? That's been the de facto standard for well over a decade now!

Yeah I went and found what I had read and it was an extension of a partnership.

I didn't want to say it. (Also, pretty much every game on every platform uses some form of texture compression)

I realise every game uses texture compression. The point is that we just need to make sure that we don't talk about uncompressed texture when working out how many would fit in a certain amount of RAM which is what I thought was being done.
 

Gahiggidy

My aunt & uncle run a Mom & Pop store, "The Gamecube Hut", and sold 80k WiiU within minutes of opening.
Sure and most of those people dont post any more. hmmmm...

Yeah i'm the bad guy....

To be perfectly honest, now when I see your username on here I reflexively squint my eyes and clinch my fists.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
Which says that it's just one way. Who's right?
The b3d poster is right, but what Pana and I said does not contradict with what that guy says. Xenon's CPU front bus (and other consoles CPU buses before it) is split in an up/down manner, and yes, that does have to do with turn-around latencies. Xenos' mem bus, though, is not split - the GPU has the full unidirectional BW at a time. That's because from the GPU's perspective BW is more important (for reading assets) than latency, and Xenos already has the needed write BW to its edram.
 

tipoo

Banned
My understanding (someone correct me if I'm wrong which is I may quite possibly be) is that both DDR and GDDR only either read or write at one time. With DDR if you want to switch from reading to writing or vice-versa you pay the "bus turnaround penalty" in the post you quoted. With GDDR the bus continuously toggles back and forth between reading and writing, eliminating this penalty at the cost of effectively reserving half your bandwidth for reads and half for writes.


So all discreet graphics cards with GDDR memory are actually reading or writing at about half of the bandwidth they advertise? Like a 140GB/s card really can only read at 70GB/s or so? This sure is news to me.
 

Donnie

Member
'limit writes to main memory' is ambiguous. and even so when there's a cause/effect on screen are you saying that artists should make smaller effect maps so that theyre tiny enough to fit? Shadow maps are already ridiculously hard on writes and reads. They seem to be of a much lower resolution and a much lower colorspace, and they also tend to take up quite a lot of space in a scene. You don't find it plausible that the small, tiny space left in the EDRam is already occupied? Even if they're hideous compared to other consoles the games would look so much worse without them.

Think you might be misunderstanding me. I was replying to a comment someone made about using the remaining space in eDRAM (after frame and Z buffer) to hold repeating textures. My point was simply that IMO that space could be better used for any read/write intensive tasks, including stuff like shadows.


But 360 doesn't have twice WiiU's bandwidth, let alone more than twice.

The b3d poster is right, but what Pana and I said does not contradict with what that guy says. Xenon's CPU front bus (and other consoles CPU buses before it) is split in an up/down manner, and yes, that does have to do with turn-around latencies. Xenos' mem bus, though, is not split - the GPU has the full unidirectional BW at a time. That's because from the GPU's perspective BW is more important (for reading assets) than latency, and Xenos already has the needed write BW to its edram.

I've always understood that GDDR3 being unidirectional means the RAM bus cannot access its full bandwidth (in 360's case 22.4GB/s) for either writes or reads, as in max read bandwidth of 11.2GB/s and max write bandwidth of 11.2GB/s. Looking around the net I can't find anything to contradict that.
 
Only thing I want to know is, with what is known. Can we expect 8192x8192 textures accompanied by a native 1080p resolution, developed from the ground up. What if the bandwidth from the 32mb edram is 500gb/s?
 

AzaK

Member
Think you might be misunderstanding me. I was replying to a comment someone made about using the remaining space in eDRAM (after frame and Z buffer) to hold repeating textures. My point was simply that IMO that space could be better used for any read/write intensive tasks, including stuff like shadows.



But 360 doesn't have twice WiiU's bandwidth, let alone more than twice.



I've always understood that GDDR3 being unidirectional means the RAM bus cannot access its full bandwidth (in 360's case 22.4GB/s) for either writes or reads, as in max read bandwidth of 11.2GB/s and max write bandwidth of 11.2GB/s. Looking around the net I can't find anything to contradict that.

360 has about 22 and Wii U about 12 does it not? I'm talking totals so I was assuming (Possibly wrongly) that if you are reading and writing fulltilt on the 360 you can get 22GB/s but on the Wii U 12 max.

Sure, if you're mainly reading then the actual throughput might be similar which was what I was trying to say earlier. That if a Wii U engine is optimised such that it reads 99% of the time because it doesn't have to do all the resolving to main RAM like the 360, then the BW issues don't really matter as much.
 

ikioi

Banned
But it has twice+ the BW of Wii U so overall it's as good if not better. It doesn't have bidirectional RAM as I've been told.

We don't know that for a fact. We don't know specifically if the Wii U's memory is bidirectional or not, nor do we know how its memory architecture really works. We know the MEM1 chips speed and bus, that's honestly about it.

Also if the Wii U's memory bandwidth really was half that of the Xbox 360/PS3, explain why games like Trine 2 are running at a higher graphical quality on Wii U. Or why Gearbox have said that the Wii U version of Aliens will be the best looking console version. All this doom and gloom about the Wii U's memory bandwidth seems at odds with what we're hearing and seeing.

George Reith: Some developers have been critical of the Wii U&#8217;s slow CPU. Did you have any trouble porting Trine 2 to Wii U because of this?

None whatsoever. The whole architecture is running very well and we were able to ramp the T2 art to a higher degree than with the other consoles.

Then there's all the games that make heavy use of the asymetrical aspect of the console. I can't see how the console can do this if its memory bandwidth was so hamstrung. Rendering one 3D image at 720p 60fps on your TV, and another at 30/60FPS on the controller. If memory bandwidth was an issue, i doubt we'd be seeing such results and capabilities.


360 has about 22 and Wii U about 12 does it not? I'm talking totals so I was assuming (Possibly wrongly) that if you are reading and writing fulltilt on the 360 you can get 22GB/s but on the Wii U 12 max.

I'm sure someone else can confirm, but i don't believe in the real world the Xbox 360's MEM1 pool achieves anywhere near 22GB/s.

Sure, if you're mainly reading then the actual throughput might be similar which was what I was trying to say earlier. That if a Wii U engine is optimised such that it reads 99% of the time because it doesn't have to do all the resolving to main RAM like the 360, then the BW issues don't really matter as much.

This is exactly my thoughts.

Developers need to utelize different memory management techniques on the Wii U to what they've traditionally used on the Xbox 360 and PS3. Developers need to make good use of the increased MEM1 capacity, lower latency access times, GPU and CPU cache, eDRAM, and the likely increased bus speed between the CPU and GPU to limit writes to MEM1. Developers who make use good use of the Wii U's architecture should find its memory bandwidth exceeds its paper specs.

Also just thinkout out loud: I know in the case of the Xbox 360 and PS3, when the CPU and GPU needed to share data they both frequently needed to write it back to MEM1 first. For example Xenon would process its portion of the data, then write the completed result to MEM1, from there Xenos would then retrieve the data and process its portion. With the Wii U appearing to have a direct bus between the GPU and CPU, the larger CPU cache, and the GPU's eDRAM, is it not likely that the GPU and CPU on the Wii U can basically swap data within their own local pools and bus and largely ignore having to write to MEM1 to communicate? If that's the case, this would significantly lower I/O on the memory bus.

I have a strong feeling there's a lot more to the Wii U's memory architecture then many are giving it credit for. It seems to be focused heavily on reduced latency, reduced pipe line stall, reduced I/O and bus stall, and all around high efficiency.
 

AzaK

Member
We don't know that for a fact. We don't know specifically if the Wii U's memory is bidirectional or not, nor do we know how its memory architecture really works. We know the MEM1 chips speed and bus, that's honestly about it.

Also if the Wii U's memory bandwidth really was half that of the Xbox 360/PS3, explain why games like Trine 2 are running at a higher graphical quality on Wii U. Or why Gearbox have said that the Wii U version of Aliens will be the best looking console version. All this doom and gloom about the Wii U's memory bandwidth seems at odds with what we're hearing and seeing.



Then there's all the games that make heavy use of the asymetrical aspect of the console. I can't see how the console can do this if its memory bandwidth was so hamstrung. Rendering one 3D image at 720p 60fps on your TV, and another at 30/60FPS on the controller. If memory bandwidth was an issue, i doubt we'd be seeing such results and capabilities.




I'm sure someone else can confirm, but i don't believe in the real world the Xbox 360's MEM1 pool achieves anywhere near 22GB/s.



This is exactly my thoughts.

Developers need to utelize different memory management techniques on the Wii U to what they've traditionally used on the Xbox 360 and PS3. Developers need to make good use of the increased MEM1 capacity, lower latency access times, GPU and CPU cache, eDRAM, and the likely increased bus speed between the CPU and GPU to limit writes to MEM1. Developers who make use good use of the Wii U's architecture should find its memory bandwidth exceeds its paper specs.

Also just thinkout out loud: I know in the case of the Xbox 360 and PS3, when the CPU and GPU needed to share data they both frequently needed to write it back to MEM1 first. For example Xenon would process its portion of the data, then write the completed result to MEM1, from there Xenos would then retrieve the data and process its portion. With the Wii U appearing to have a direct bus between the GPU and CPU, the larger CPU cache, and the GPU's eDRAM, is it not likely that the GPU and CPU on the Wii U can basically swap data within their own local pools and bus and largely ignore having to write to MEM1 to communicate? If that's the case, this would significantly lower I/O on the memory bus.

I have a strong feeling there's a lot more to the Wii U's memory architecture then many are giving it credit for. It seems to be focused heavily on reduced latency, reduced pipe line stall, reduced I/O and bus stall, and all around high efficiency.

Well we know the model of the RAM chips which gives us an upper bounds on speed. I think upper might have been 17, and average about 12. Anyway the last bit you quoted of mine explains that I don't necessarily think it's a bad thing if Wii U is 1/2 RAM BW of 360.
 

Donnie

Member
360 has about 22 and Wii U about 12 does it not? I'm talking totals so I was assuming (Possibly wrongly) that if you are reading and writing fulltilt on the 360 you can get 22GB/s but on the Wii U 12 max.

Sure, if you're mainly reading then the actual throughput might be similar which was what I was trying to say earlier. That if a Wii U engine is optimised such that it reads 99% of the time because it doesn't have to do all the resolving to main RAM like the 360, then the BW issues don't really matter as much.

Yeah I just mean that 22.4GB/s (360 mem bandwidth) isn't twice as much as 12.8GB/s (WiiU's mem bandwidth) as you said 360 had twice+ the bandwidth.
 

tipoo

Banned
I wonder if any of this made it into the Wii U? It uses unified system memory, and Nintendo specifically mentioned latency between chips being important to them. Not the APU + GPU part, obviously.

http://forum.beyond3d.com/showpost.php?p=1692947&postcount=18108


Hey guys,

I have a few questions concerning the rumour of an APU and GPU combination concerning the new PlayStation, if that is agreeable. Maybe someone can bring a little light into my darkness.

I stumbled upon this slide from the Fusion Developer Summit which took place in June 2012. The slide deals with GPGPU algorithms in video games. There are a couple of details that are probably somewhat interesting when speculating about a next generation gaming console.

As far as I understand, AMD argues that today GPGPU algorithms are used for visual effects only, for example physics computations of fluids or particles. That is because developers are facing an insurmountable bottleneck on systems that use homogeneous processor architectures. AMD calls it the copy overhead. This copy overhead originates from the copy work between the CPU and the GPU that can easily take longer than the processing itself. Due to this problem game developers only use GPGPU algorithms for visual effects that don't need to be sent back to the CPU. AMD's solution for this bottleneck is a unified adress space for CPU and GPU and other features that have been announced for the upcoming 2013 APUs Kabini (and Kaveri).

But these features alone are only good for eliminating the copy overhead. Developers still have to deal with another bottleneck, namely the saturated GPU. This problem is critical for GPGPU in video games since the GPU has to deal with both, game code and GPGPU algorithms at once. I'm not sure whether this bottleneck only exists for thick APIs like DirectX or if it also limits an APU that is coded directly to the metal. Anyway, AMD claims that a saturated GPU makes it hard for developers to write efficient GPGPU code. To eliminate this bottleneck AMD mentions two solutions: Either you can wait for a 2014 HSA feature that is called Graphics Pre-Emption, or you can just use an APU for the GPGPU algorithms and a dedicated GPU for graphics rendering. The latter is what AMD recommends explicitly for video gaming and they even bring up the similarities to the PlayStation 3, which renownedly uses SIMD co-processors for all kinds of tasks.


I would like to know what you guys think about these slides.


What if AMD was building an 28nm APU for Sony that is focused solely on GPGPU, for example four big Steamroller cores with very fast threads in conjunction with a couple of MIMD engines? Combine it with a dedicated GPU and a high bandwidth memory solution and you have a pretty decent next gen console.

I would also like to know if an APU + GPU + RAM system in package is possible with 2.5D stacking, which was forecasted by Yole Development for the Sony PlayStation 4, for IBM Power8 and Intel Haswell.

And since Microsoft is rumoured to have a heavily customized chip with a "special sauce", could that mean they paid AMD to integrate the 2014 feature Graphics Pre-Emption in the XBox processor, so they can go with one single ultra-low latency chip instead of a FLOP-heavy system in package?
 

IdeaMan

My source is my ass!
http://www.notenoughshaders.com/2013/01/17/wiiu-memory-story/


really good read up on memory discussion; I created a thread about it too (not very good with the whole thread creation thing) but I think this thread is better suited to the discussion

very much in line with thraktor and blu, and what many others in this thread have been speculating. its a good read



(thanks @ gahiggidy for helping me find this thread)

Yeah, i want to thank blu, Alstrong, but also Thraktor, Durante, Popstar and Quaz51 for their insight that helped me writing this article (one month ago, but then i had PC problems and kinda busy irl ><) :)

I hope you'll enjoy it, even if the tech-aware won't learn revolutionary things.
 

ikioi

Banned
ideaman are you able to confirm with your sources if the Wii U's memory is bidirectional, and if the memory controller is running the ram in single, dual, or triple channel?

Some are saying it's single, others bidirectional.

I'd be more inclined to believe it's bidirectional given Nintendo do have a major love with their console's memory configurations.
 

IdeaMan

My source is my ass!
ideaman are you able to confirm with your sources if the Wii U's memory is bidirectional, and if the memory controller is running the ram in single, dual, or triple channel?

Some are saying it's single, others bidirectional.

I'd be more inclined to believe it's bidirectional given Nintendo do have a major love with their console's memory configurations.

If i happen to learn something about that, i'll let you know :) All my sources are focusing on "real performance in game" and they haven't witnessing problems with the whole memory chain.
 

Donnie

Member
It means what we've suspected for a long while. That the next XBox and Playstation won't be near the kind of jump over WiiU that PS3 and 360 were over the original Wii.

What it'll mean as far as ports to and from WiiU and these systems is still up in the air. All we can say for certain is porting something down from XBox3/PS4 to WiiU will be easier than porting something down from 360/PS3 to Wii was.
 

ozfunghi

Member
It means what we've suspected for a long while. That the next XBox and Playstation won't be near the kind of jump over WiiU that PS3 and 360 were over the original Wii.

What it'll mean as far as ports to and from WiiU and these systems is still up in the air. All we can say for certain is porting something down from XBox3/PS4 to WiiU will be easier than porting something down from 360/PS3 to Wii was.

Well, they didn't provide any hard numbers on the 720 GPU, but if the rumors are true, about it being 1.2 TFlops, i wonder how lherre thinks about that, since i believe he claimed the gap between WiiU and the next xbox would be larger than that between Dreamcast and Xbox. Unless the U-GPU ends up being nowhere near 500 GFlops.

Edit: also, PS4 reserving 512MB for OS... they can't expand later on, so they must feel comfortable with that number or else they would have reserved more and streamlined it later on. An indication that Nintendo can trim their OS while remaining "next-gen"?
 
Top Bottom