Predict: The Next Generation Console Tech

[I.S.T.]

BOOM!!!
http://club.tgfcer.com/thread-6593085-22-1.html

The Sapphire HD7970 poison Edition graphics card PCB front and back each welding 12 from Hynix H5GQ2H24AFR GDDR5 memory particles, formed 6GB/384bit memory specifications.

Some people say that the memory is very expensive, people go to Taobao can search a search the particles, to see how much not Luanxiang, just say the cost of video memory, because I'm not Microsoft, can not see the final product in the end with what

Hey, somebody there with a Nichijou avatar. Fucking awesome.

 

[3dilettante]

Something like "HSA"?

The way HSA is presented, it could be used to transform code to run on SPEs, I think. However, HSA's big benefits include a coherent, unified, and pageable memory space. The SPEs and their local stores do not benefit from it.
Since HSA's memory model is more flexible than using a local store, I think it could be constrained to work on the SPEs, although it would be a project to make the run time and libraries necessary to basically make SPE-suitable code from something that has no concept of their limitations.

At that point, I'd be curious if it's worthwhile, or if the fact that the SPEs are present needs to be exposed to the programmer, which seems to be counterproductive.

 

[patsu]

I'm here without ideas on the subject - if they do include what is essentially a new Cell chip in the console and have dreamed up functional roles for it to play, I will be floored. Why give 200 MP3 streams up as an example though? Just seems strange, especially if it is a DSP. (And why claim otherwise if it's not?) The transistor budget... it just doesn't make sense in my mind.

Hey - I'm ready to be surprised. Sony: surprise me.

I see "Sony why you no surprise me" in your future. ;-)
I am 90% convinced that it will not have Cell or SPURS, 'specially if third parties have (1) equivalent/better "easily portable" PC games for quick bucks, and (2) PS3 games that don't use SPUs effectively. The first parties can port their titles selectively.

The new SIMD engine(s), if any, will need to handle:
* Audio (easy ?)
* H.265 (right ?)
* Physics

Sony engineers had a hard time implementing H.264 on Cell because of CABAC. They had to redo the implementation to finally make it run (very) well. This time round, they will/should see what H.265 needs first. As I understand, H.265 is about double the run-time work than H.264. So they may look for something easier and perhaps more custom/cheaper.

Finally, security. According to the unofficial specs, the OS will use the same main memory as other components. This is very unSony. I suspect Sony will lock down something for security purposes. They locked down the storage card on Vita. They sealed the SPU security kernel on PS3. On PS4, perhaps they will have another special sauce here (Something SPU-lilke but custom ?)

All speculations. Take it with a lot of salt.

EDIT:

The way HSA is presented, it could be used to transform code to run on SPEs, I think. However, HSA's big benefits include a coherent, unified, and pageable memory space. The SPEs and their local stores do not benefit from it.
Since HSA's memory model is more flexible than using a local store, I think it could be constrained to work on the SPEs, although it would be a project to make the run time and libraries necessary to basically make SPE-suitable code from something that has no concept of their limitations.

At that point, I'd be curious if it's worthwhile, or if the fact that the SPEs are present needs to be exposed to the programmer, which seems to be counterproductive.

Yes, I think counterproductive is the right word too.

 

[liquidboy]

DSP's in Durango

It's not a dumb question at all. It just seems very unlikely, unless Sony has found some manner via which to exploit them constantly and for multiple purposes, rather than simply for BC and for tasks that could be accomplished similarly for less (the audio DSP theory).

Something like "HSA"?

Interesting you bring up DSP ... I've been following the DSP discussion for the Durango perspective (all rumors of course)...

From what I can gather MS will have a couple of DSP's that are being sold as a way to "future proof" the system..

We have an Audio DSP sitting on 1 of the 3 socs
We have Hardware DSP's sitting on the other 2 socs.

The following patents from MS

1. "Using ray tracing for real time audio synthesis"
2. "Exposing off-host audio processing capabilities"

Explicitly these patents talk about how "Audio DSP" on a device can be used as a source of audio stream signals that can be pushed to
a) other SoC/Components within the host for processing
b) to a different external host (like the cloud) for processing

(interestingly it uses RayTracing techniques to carry out processing of these signals)

The result of the processing can be returned back to the host, if it was processed outside of it, and the played ...

There are other patents that talk about using the other DSP's (video/HW) as sources of streams that can be orchestrated outside of the Host (cloud etc) ... basically using similar techniques as the audio dsp

I guess that's one way MS is future proofing its console, by adding a dimension of the cloud to augment their hardware via these DSP's...

I know its all rumors but its ive always been interested in the rumors of Durango and its DSP's as being a differentiating point ...

 

[Carl B]

That's just the thing though... I view the idea and discussion revolving around these audio DSPs as a total non-event. As far as chips go, you could not get more straightforward. If there's something more exciting at play with these things, I have a hard time understanding why they are being classed simply as audio DSPs in these rumors.

Now - there is processing power required under these situations, for sure, and by having said chips you free up resources on your primary silicon. But the presence of dedicated audio chips would not elevate anything to the realm of the novel, at least IMO.

 

[(((interference)))]

What makes more sense though is that rumors tossed around back and forth here end up in the articles of the gaming press; I'm not ready to ordain these rumors as truth because they were spat back at us in an article. Let's take our time here still...

With DF at least, I don't think Richard is just regurgitating rumours from GAF/B3D, since he hasn't posted any of the rumoured Durango specs. I think he's trying to double or triple source everything before he posts it - and his Durango sources haven't been as forthcoming as his Orbis ones.

 

[LXFBN]

AndyH on NeoGAF

It's nice to finally have an article from a known website on all of these rumours.

I talked about most of this regarding Durango over the past year here - 8 cores at 1.6ghz, 8GB of slow ram, rough estimation of GPU, Win8 utilisation etc. The 3GB and 2 core dedicated to the OS is real. I guess you will just have to wait and see what they have in store.

Here's a couple things I have talked about before that weren't talked about in the article. Microsoft are encouraging 1080p at 30fps or 720p at 60fps to developers. I am not sure if they are trying to enforce this or not.

Approximately 1GB of RAM can be accessed per frame. Dunno if it was at 30fps or 60fps.

Not sure if this is actually new but I was informed that the cores in the Durango processor have their own FPUs rather than the 2 modules share one FPU that the older AMD processors were using. I haven't looked into whether this is something that is new with Jaguar processors though, so this might be nothing special.

Regarding the RAM difference between the two consoles I don't think there will be too much difference between them in actual games other than maybe stuff like alpha blending operations but this is just speculation on my part.

I don't think I am going to talk about anything that isn't already out there before Microsoft and Sony officially reveal their consoles. I am looking forward to it.

 

[XpiderMX]

AndyH on NeoGAF

What we know about him? is he legit? the "1080p/30 - 720p/60" statement looks weird to me.

 

[patsu]

That's just the thing though... I view the idea and discussion revolving around these audio DSPs as a total non-event. As far as chips go, you could not get more straightforward. If there's something more exciting at play with these things, I have a hard time understanding why they are being classed simply as audio DSPs in these rumors.

Now - there is processing power required under these situations, for sure, and by having said chips you free up resources on your primary silicon. But the presence of dedicated audio chips would not elevate anything to the realm of the novel, at least IMO.

For PS4's 200 concurrent MP3 streams rumor... assuming it's true, may be it's based on FPGA ?
http://www.bdti.com/MyBDTI/pubs/20061101_gspx06_fpgas.pdf

That slide argues that traditional DSP metrics can't be used to measure the performance of equivalent FPGA implementations effectively. As a result, Sony may choose the final outcome (e.g., # of concurrent streams) to communicate/advertise its performance ? It would avoid those dumb/smart FLOP slippery slope discussions. And they can reconfigure it after delivery to improve the implementation, and support new codecs ?

An FPGA implementation would essentially skip/save the FLOP count ? and explain why audio DSP is listed separately in the rumored feature list too.

Otherwise, I have no clue why the Sony CTO mentioned "reconfigurable logic" in his interview.

EDIT: Unfortunately it doesn't tell us anything about the SIMD mentioned in Eurogamer's article.

EDIT 2: For Cell, so far, I think the most compelling argument for PS4 SPU is the security framework. Security is one of those realms where cheaper, faster and better may not work for the management. Everyone in Sony up to Hirai and Stringer has been burnt before. The SPU security framework is more or less proven. They may be able to fix the holes and improve it further. None of the completely new implementation can claim this. Then again, it assumes that the Sony security team can't come up with another proven and better mechanism. Even if they follow a Cell-like approach, it will likely be a stripped down, custom implementation.

 

[Helmore]

What we know about him? is he legit? the "1080p/30 - 720p/60" statement looks weird to me.

Especially as that means it will have around 30-35 GB/s in bandwidth, at least the way he phrased it.

EDIT: Taking all the neXtBox rumors together I get something like this:

-8 Jaguar cores at 1.6 Ghz
-12 CUs (768 ALUs) @ 800 MHz
-32 MB of eSRAM
-8 GB of DDR3 memory at 2166 MHz on a 128-bit bus
-some extra doohickies
It would end up being a chip of around 250 mm² and the entire console would consume around 120 Watt maximum (probably less).

To be honest, I would be a bit disappointing with something like that.

 

[DieH@rd]

Especially as that means it will have around 30-35 GB/s in bandwidth, at least the way he phrased it.

...or ~60 GB/s bandwith.

 

[(((interference)))]

AndyH is legit, see Proelites summary of his statements on GAF:
http://forum.beyond3d.com/showthread.php?p=1691527&highlight=andyh#post1691527

Even bkilian thinks so:
http://forum.beyond3d.com/showpost.php?p=1691543&postcount=17578

...or ~60 GB/s bandwith.

Well thuway and Rangers have independently confirmed it is 68 GB/s or thereabouts.

 

[Solarus]

Interesting you bring up DSP ... I've been following the DSP discussion for the Durango perspective (all rumors of course)...

From what I can gather MS will have a couple of DSP's that are being sold as a way to "future proof" the system..

We have an Audio DSP sitting on 1 of the 3 socs
We have Hardware DSP's sitting on the other 2 socs.

The following patents from MS

1. "Using ray tracing for real time audio synthesis"
2. "Exposing off-host audio processing capabilities"

Explicitly these patents talk about how "Audio DSP" on a device can be used as a source of audio stream signals that can be pushed to
a) other SoC/Components within the host for processing
b) to a different external host (like the cloud) for processing

(interestingly it uses RayTracing techniques to carry out processing of these signals)

The result of the processing can be returned back to the host, if it was processed outside of it, and the played ...

There are other patents that talk about using the other DSP's (video/HW) as sources of streams that can be orchestrated outside of the Host (cloud etc) ... basically using similar techniques as the audio dsp

I guess that's one way MS is future proofing its console, by adding a dimension of the cloud to augment their hardware via these DSP's...

I know its all rumors but its ive always been interested in the rumors of Durango and its DSP's as being a differentiating point ...

How would raytracing audio work? Use the tracing to figure out how sound would bounce? Seems cool, makes more sense than VTE for doing raytracing lighting.

Are we sure there won't be a VTE in Durango? Somenoe said raytracing takes alot of memory maybe this si why there is 8gb of GDRR5 in durango? i think durango will do raytracing, or maybe its just what i really wan to happen

i guess a better question is why WOULDN'T there be VTE in durango for raytracing? if a 680gtx can do raytracing whats to say a vte made just for raytracing wouldn't happen?

or should i am just kidding myself? (be brutally honest guys!)

 

[jonabbey]

It sounds like all the VTE/BTE/multi-SOC talk is just way off.

As far as raycasting for audio processing, it's pretty common to use a technique called beamforming with an array of microphones to do directional focusing, but that has nothing in particularly to do with ray tracing.

 

[Love_In_Rio]

How would raytracing audio work? Use the tracing to figure out how sound would bounce? Seems cool, makes more sense than VTE for doing raytracing lighting.

Are we sure there won't be a VTE in Durango? Somenoe said raytracing takes alot of memory maybe this si why there is 8gb of GDRR5 in durango? i think durango will do raytracing, or maybe its just what i really wan to happen

i guess a better question is why WOULDN'T there be VTE in durango for raytracing? if a 680gtx can do raytracing whats to say a vte made just for raytracing wouldn't happen?

or should i am just kidding myself? (be brutally honest guys!)

Well,if UE4 ditched voxels based global ilumination forget about the tracing thing.

 

[Solarus]

is there any logical reason besides raytracing why they would ditch svogi? i don't know aything about graphics engines but id imagine that's not a simple thing to remove like that. especially when developers have ue4 for development use? i couldn't imagine them building their games around svogi then they get a new update and it's gone, unless whatever it's being replaced with is better. looks like raytracing might not be far fetched after all?

It sounds like all the VTE/BTE/multi-SOC talk is just way off.

As far as raycasting for audio processing, it's pretty common to use a technique called beamforming with an array of microphones to do directional focusing, but that has nothing in particularly to do with ray tracing.

Between Liquid, bgassassin and bkilian there has to be some truth to the raytracing VTE angle I'd imagine? doesnt bkilian work for ms? iwhy he'd bring up specialized chips and raytracing patents if there wasn't something more to it.i think durango will do raytracing of some kind. if i had a much ebtter understanding of software engineering i'd laugh at my owns posts probably??? i just think with so much talk about it there has to be a ray of truth to it

 

[anexanhume]

AndyH on NeoGAF

AVX2 I wonder?

 

[Love_In_Rio]

is there any logical reason besides raytracing why they would ditch svogi? i don't know aything about graphics engines but id imagine that's not a simple thing to remove like that. especially when developers have ue4 for development use? i couldn't imagine them building their games around svogi then they get a new update and it's gone, unless whatever it's being replaced with is better. looks like raytracing might not be far fetched after all?



Between Liquid, bgassassin and bkilian there has to be some truth to the raytracing VTE angle I'd imagine? doesnt bkilian work for ms? iwhy he'd bring up specialized chips and raytracing patents if there wasn't something more to it.i think durango will do raytracing of some kind. if i had a much ebtter understanding of software engineering i'd laugh at my owns posts probably??? i just think with so much talk about it there has to be a ray of truth to it

Technical problems with textures and i supposse that using voxels would carry a lot of new ones.I dont know if it is really ditched?

 

[Love_In_Rio]

AVX2 I wonder?

I think in jaguar each core has its fpu attached, not like in bobcat.

 

[Rangers]

I'm assuming bgassassin's looking at paper specs that have significant architectural similarities, hence on paper Orbis should beat Durango. Let's say his sources have told him both are 8 core Jaguars at 2 GHz, 8800 type GPUs, one at 1.2 TFlops and one at 1.8 TFlop (apple to apple comparison), one with 100 GB/s BW and one with 200 GB/s BW. That would point to the second machine being the more powerful, and so comments that there's not much between the machines would be hard to reconcile. At which point, you'd want an explanation, and bg's asking if it could be dev tools. I don't think he'd ask his question if there wasn't good reason to see the hardwares as comparable. If the answer to his question is 'no', then he knows either the development comments are wrong, or the hardware specs he's seeing are wrong, or there's more to the machines than the specs are saying.

I guess this is more for the versus thread but, everybody is completely ignoring that the Durango has significantly more RAM in their rush to proclaim Orbis king.

I mean I see two main differing stats (assuming MS built in enough BW that Durango wont be crippled by lack of, and I'm betting they did).

1.8 TF vs 1.2TF, 50% difference
5GB vs 3.5 GB game usable RAM, 43% difference

But everybody seems to be ignoring the bottom one.