NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

[patsu]

In regards to the 4 CU's only giving a minor boost in graphics, it makes sense. There's more to a GPU than CU's so the impact of having those 4 CU's may not be the greatest. With that being said, it would appear that the 4 CU's are more a replacement for SPE's if anything, let's just hope they are as flexible. It will be additional resources used for computing various tasks that would be too much for the CPU's FPU to do it and thus freeing up those resources for other tasks. They may end up being used for graphical tasks in the end since devs are always finding ways to make the most out of a given (closed system) architecture.

According to the same rumors, standard audio, video, decompression have dedicated hardware support in Orbis. For the most part, only AI, natural interface input, second display, physics and graphics jobs are left. Security should be a CPU thing.

The CPU has to do some work, so I'm guessing AI and physics may go there. These 4 offbeat CUs -- if present -- should be used mainly for graphics tasks, perhaps driven by AI and physics input from the CPU. If they are meant to be SPU replacements, there should be a wealth of follow-up ideas and techniques from PS3.

Sony should put 28nm CELL in new [mandarory!] PSEye, connect it to PS4 with Thunderbolt, and use some of its processing power to for various PS4 applications.
/runs away

I don't know... sounds like an overkill to include Cell (add-on or built-in) together with these 4CUs.

DF has a rumor on 22nm Cell, but I am not a believer.

 

[Inuhanyou]

From that article of blocking used games, we also heard of Microsoft actively restricting devs to much more restrictive API whereas Sony was apparently more concerned with devs coding to the metal and getting much more power out of the console.

I wonder why Microsoft would go through the trouble of trying to mitigate all these performance deficiencies through the DME's and ESRAM if their goal was clear to begin with, and they were never going to let devs maximize performance through 720's internal hardware. Its such a clear departure from 360 when everything about the internal hardware says they were going with what worked when designing it as an evolution of 360's hardware

 

[dobwal]

No, you're getting it wrong, 4 compute CUs can do graphics tasks just as well as the 14 other CUs but have an additional scalar ALU that makes them better suited to compute tasks.

CPU:

Orbis contains eight Jaguar cores at 1.6 Ghz, arranged as two “clusters”
Each cluster contains 4 cores and a shared 2MB L2 cache
256-bit SIMD operations, 128-bit SIMD ALU
SSE up to SSE4, as well as Advanced Vector Extensions (AVX)
One hardware thread per core
Decodes, executes and retires at up to two intructions/cycle
Out of order execution
Per-core dedicated L1-I and L1-D cache (32Kb each)
Two pipes per core yield 12,8 GFlops performance
102.4 GFlops for system

Does the extra alu in the 4 CUs rumor come from the bolded part?

Why is this not a referral to the AVX instruction set and the FPU. The FPU's pathway was doubled to 128 bits on the jaguar and double pumping allows one 256 bit AVX operation per clock. It makes sense that this refers to AVX and the FPU as AVX is an advance version of SSE (Streaming SIMD Extensions).

 

[arijoytunir]

From that article of blocking used games, we also heard of Microsoft actively restricting devs to much more restrictive API whereas Sony was apparently more concerned with devs coding to the metal and getting much more power out of the console.

Microsoft is taking the apple direction of hardware . They are using their own APi layer to be used by the devs . Like the iOS devices the actual underlying software is same . By creating a Xbox like this they can bring out new Xbox with minor hardware improvements like the iPad and iPhones within 2-3 years . This will solve the stagnation of console hardware problems . And I think coding to the metal will not work in this model of hardware .

Also by blocking used games they are following while digital model as apple and steam .

 

[east of eastside]

Could the 4 CU's be specialized in any way to play a role in facilitating CELL/SPE emulation/ PS3 backwards compatibility?

 

[B Real]

Nikkei is reporting there will be no cell

http://gematsu.com/2013/02/nikkei-ps4-to-utilize-gaikai-have-expanded-social-features

 

[Nevod]

Does the extra alu in the 4 CUs rumor come from the bolded part

Nope, that came from that "4 CUs for compute tasks, minor boost if used for rendering".

There's more to a GPU than CU's so the impact of having those 4 CU's may not be the greatest.

That makes real sense. ROPs or elsewhat is balanced for 14CUs, and adding other 4 to graphics would just bottleneck some other part.

The CPU has to do some work, so I'm guessing AI and physics may go there.

AI would surely be on CPU, GPU isn't suited for such algorithms at all. On the contrary, GPU is way better than CPU for physics, as it's mostly number crunching, and just a single CU is equal to whole CPU in that regard.

I really hope for detailed interactive physics, not just loads of particle effects.

 

[Nevod]

By the way, 1080p60 vs 720p30 was an extreme worst case. I think it will be 1080p60 everywhere vs 1080p30+interpolation on most titles and on some first party 1080p60.

 

[Lycan]

So those DMEs coupled with the SRAM could in effect nullify the performance delta between Orbis and Durango after all? :smile:

 

[Nevod]

Does the extra alu in the 4 CUs rumor come from the bolded part?

Why is this not a referral to the AVX instruction set and the FPU. The FPU's pathway was doubled to 128 bits on the jaguar and double pumping allows one 256 bit AVX operation per clock. It makes sense that this refers to AVX and the FPU as AVX is an advance version of SSE (Streaming SIMD Extensions).

So those DMEs coupled with the SRAM could in effect nullify the performance delta between Orbis and Durango after all? :smile:

I think no. Perhaps Durango will have some 10% better performance per CU, but the PS4's GPU is too powerful for that to compensate. Probably, Xbox will have lower framerate and probably there won't be other significant differences.

DMEs and SRAM mean just that GPU won't be starved despite relatively slow main memory and would be used to the max.

 

[dobwal]

Microsoft is taking the apple direction of hardware . They are using their own APi layer to be used by the devs . Like the iOS devices the actual underlying software is same . By creating a Xbox like this they can bring out new Xbox with minor hardware improvements like the iPad and iPhones within 2-3 years . This will solve the stagnation of console hardware problems . And I think coding to the metal will not work in this model of hardware .

Also by blocking used games they are following while digital model as apple and steam .

Unless the hardware is sold at a relatively cheap price without sacrificing major IQ leaps in performance and features, I doubt the mainstream market would want upgrade on such a cycle and profit would be squeezed by poor adoption rates. Apples relatively short periods between product refreshes is due to the higher volume of unit sales. Apples's iPhone/iPad sells more in a year than than the PS360 do in their lifetime. The smartphone market support far more sales than the console market. Plus combined with the $200 a pop on average in operating profit per hardware unit allows for a more aggressive release schedule.

 

[patsu]

AI would surely be on CPU, GPU isn't suited for such algorithms at all. On the contrary, GPU is way better than CPU for physics, as it's mostly number crunching, and just a single CU is equal to whole CPU in that regard.

I really hope for detailed interactive physics, not just loads of particle effects.

Hmm... I think developers will probably have a choice whether to run physics computation on the GPU or the CPU, or both. I remember there are implementations on the PPU VMX and the SPUs this gen. Otherwise, the CPU's vector units seem underutilized.

 

[(((interference)))]

The studio working on PGR5 is called Lucid Games
http://www.vg247.com/2013/02/06/xbox-720-watch-project-gotham-5-art-unearthed-launching-2013-report/

The teaser image adds more weight that Xbox will also be launching in 2013:

 

[(((interference)))]

I think no. Perhaps Durango will have some 10% better performance per CU, but the PS4's GPU is too powerful for that to compensate. Probably, Xbox will have lower framerate and probably there won't be other significant differences.

Yeah, I don't think the differences will be too big - for multiplatform titles anyway.
It might be like earlier PS3 exclusives compared to their 360 counterparts eg. GTA4, RDR, most of the CODs on PS3 vs 360.

 

[dobwal]

Nope, that came from that "4 CUs for compute tasks, minor boost if used for rendering.

Or it can that the 14 CUs may not be on the same silicon as the 4 CU apu. I not convince that a 20 CU 8 CPU doesn't represent too big of a die to launch inside relatively low priced but mainstream hardware. It could also mean the 4 CUs already have designated CPU oriented responsibilities that wont allow much contribution to gpu tasks. It's a huge leap in logic to automatically conclude such a specific action from such a vague statement.

Where can u place an extra scalar alu in the CU that's great for CPU oriented tasks but basically kill its gpu performance?

 

[Lycan]

The studio working on PGR5 is called Lucid Games
http://www.vg247.com/2013/02/06/xbox-720-watch-project-gotham-5-art-unearthed-launching-2013-report/

The teaser image adds more weight that Xbox will also be launching in 2013:

My mouth is watering

 

[AlphaWolf]

Or it can that the 14 CUs may not be on the same silicon as the 4 CU apu. I not convince that a 20 CU 8 CPU doesn't represent too big of a die to launch inside relatively low priced but mainstream hardware. It could also mean the 4 CUs already have designated CPU oriented responsibilities that wont allow much contribution to gpu tasks. It's a huge leap in logic to automatically conclude such a specific action from such a vague statement.

Where can u place an extra scalar alu in the CU that's great for CPU oriented tasks but basically kill its gpu performance?

If there are two redundant CU's it would make sense to keep the units together so any two bad CU's mean you still have a usable part. Is there some logical (beneficial) reason to split four of the CU's off from the GPU?

 

[Nevod]

Or it can that the 14 CUs may not be on the same silicon as the 4 CU apu. I not convince that a 20 CU 8 CPU doesn't represent too big of a die to launch inside relatively low priced but mainstream hardware. It could also mean the 4 CUs already have designated CPU oriented responsibilities that wont allow much contribution to gpu tasks. It's a huge leap in logic to automatically conclude such a specific action from such a vague statement.

Where can u place an extra scalar alu in the CU that's great for CPU oriented tasks but basically kill its gpu performance?

That is also a version, I've thought that myself. However, don't see any real reason for such a split.

There's no real way to kill performance by adding an ALU into CU, imho. Moreover, such an addition seems to be unlikely as it's quite a change in architecture. It could be that 4 CUs are of GCN2 arch, and 14 of standard GCN, but that is still doubtful, as it seems that PS4 is quite a straightforward machine, and such a split between architectures adds complexity. Most probably it is indeed that the graphic system is balanced in a way that 14 CUs exactly utilise everything to max without any bottlenecks.

 

[Silent_Buddha]

So those DMEs coupled with the SRAM could in effect nullify the performance delta between Orbis and Durango after all? :smile:

Unlikely, but it may help close the gap depending on the workload. Orbis from what is released is relatively standard.

Durango from what is released, at least appears to be focused on increasing the efficiency (utilization is another way to think of it) of system resources.

That will help it to decrease the performance gap in some cases, potentially increase it in corner cases, but is likely to be overall slower than the brute force approach of Orbis.

Also take into account that due to the relatively standard nature of Orbis a developer is likely not to have to put as much effort into getting as much out of it as they can. Doing the same with Durango may result in a similar but potentially slightly higher increase in resource utilization. Unless the process of increasing utilization is transparent, in which case it may increase more. On the other hand it may require developer effort to leverage the hardware in such a way as to reach that higher utilization.

Short answer. We still know far too little to make any reasonable judgement as to whether or how much Durango's focus could close the performance gap.

Likewise we really don't know enough about the 14+4 CU organization that is mentioned in the leak. It's easy enough to just think of it as 18 CUs, but in that case why even mention 14+4. Again, we have too little information.

Anybody claiming to know and then saying what it is, likely doesn't know. Otherwise they'd be under NDA and unable to say what they know.

So, take every single thing said in this thread so far as one huge guess by anyone saying it. And then wait until the consoles come out to see what it might be. And even then we may not get to know what things are really like.

Regards,
SB

 

[arijoytunir]

Unless the hardware is sold at a relatively cheap price without sacrificing major IQ leaps in performance and features, I doubt the mainstream market would want upgrade on such a cycle and profit would be squeezed by poor adoption rates. Apples relatively short periods between product refreshes is due to the higher volume of unit sales. Apples's iPhone/iPad sells more in a year than than the PS360 do in their lifetime. The smartphone market support far more sales than the console market. Plus combined with the $200 a pop on average in operating profit per hardware unit allows for a more aggressive release schedule.

so consoles follow the business model of google nexus and amazon devices . they sell the hardware without taking and profit or loss ;i.e bom prices and make up the profit through their services .

does the profit really that much more on subsidised products like consoles ? like a console is selling for 200 bucks to make even with its manufacturing cost and then the manufacturer is making profit through their cuts in games and services . so if the console is sold at 500 bucks making a profit of 300 bucks - then will the services and games be priced significantly less ? and if so why arent the console companies trying this model ?

ps : sorry for being off topic