So is the cell spe's "downgrade" confirmed ?

[nAo]

Bah, accuracy elitist. I'll have you spend a week with clipping on PSP and then you try telling me accuracy matters.

LOL, the clipping curse strikes back!

128 bits seems like overkill for a z-buffer, but..*looks at Faf's post*...(some) geometry? I guess if SPEs are involved somewhere along the way, that might make sense? Or vertices are usually 128-bit anyway, I guess?

I've said FP32, not 128bit One might use single channel FP32 render targets or textures.
Uncompressed vertices position require 96 bits (the 4th component can be automatically set to one)

 

[Shifty Geezer]

I've said FP32, not 128bit One might use single channel FP32 render targets or textures.
Uncompressed vertices position require 96 bits (the 4th component can be automatically set to one)

C'mon, you tease! Give us some examples. What exactly are you rendering? 32 bit depth buffer? 32 bit illumination-channel for blending?

 

[nAo]

C'mon, you tease! Give us some examples. What exactly are you rendering? 32 bit depth buffer? 32 bit illumination-channel for blending?

If I can't tease where all the fun is!?
BTW you guessed it right, we're using a FP32 RT to store a very accurate W buffer.

 

[ban25]

Keep in mind it's really 64 MB + 32 MB. Where would the memory go? How about web browsing? Right now Firefox is using up 72MB on my PC. Without the ability to page out to disk, it would crash if this was the PS3 right now...

How many times have you browsed a page on the PSP and received an "out of memory" error? For me, it happens at least once per session.

From what I removed from the XB360 OS thread, can anyone provide a plausible explanation as to how 96 MB can be used up in operations concurrent with gaming? 32 MB would be 3 1080p screens and 8 MB for applications. 48 MB sounds plenty. 96 MB sounds like an insane waste, reserving resources for features that really aren't going to be used by most people most of the time, that I can see.

 

[mckmas8808]

BTW you guessed it right, we're using a FP32 RT to store a very accurate W buffer.

Okay Shifty since you guessed it right, can you explain in English what using FP32 RT in a W buffer means?

Thanks.:smile:

 

[_phil_]

Okay Shifty since you guessed it right, can you explain in English what using FP32 RT in a W buffer means?

Thanks.:smile:

Very far view distance ! (and still precise)

 

[DeanoC]

Okay Shifty since you guessed it right, can you explain in English what using FP32 RT in a W buffer means?

I'll have a guess at what Marco is talking about ;-)
Having a depth value per pixel is helpful because its all you need to construct the closest surface from the viewer. The camera and view plane give you the x and y coords, and you can use the stored depth to get a complete world space point.

The actual depth buffer is designed to have just enough accuracy for hidden surface determination (whats 'thing' is closer to you) and doesn't really have enough to get an accurate re-construction. When you have an accurate depth value you can use it for things like depth of field, shadows etc.

Storing W versus Z are just the usual alternative ways of measuring distance used in graphics.

 

[mckmas8808]

I'll have a guess at what Marco is talking about ;-)
Having a depth value per pixel is helpful because its all you need to construct the closest surface from the viewer. The camera and view plane give you the x and y coords, and you can use the stored depth to get a complete world space point.

The actual depth buffer is designed to have just enough accuracy for hidden surface determination (whats 'thing' is closer to you) and doesn't really have enough to get an accurate re-construction. When you have an accurate depth value you can use it for things like depth of field, shadows etc.

Storing W versus Z are just the usual alternative ways of measuring distance used in graphics.

Okay so what phil said about having a very far view distance yet precise was correct?

 

[MrWibble]

I'll have a guess at what Marco is talking about ;-)

Nice to see such a natural understanding between team mates

 

[Shifty Geezer]

And Deano's the lead programmer! Seems a bit disconcerting he can only guess at what his underlyings are doing.

mckmas : Yes-ish, about the distance. It's more a matter of accuracy than anything, avoiding errors with things being drawn in the wrong order especially at distance. A W buffer is good for elliminating errors at distance but not so good for near objects. 32 bit representation is higher enough resolution to accomodate all situations. A per pixel distance is a very nice, and I take from the RT that these are being RayTraced which sounds cool!

 

[Carl B]

C'mon now, Marco is no one's underling; he's a one man tour de force!

 

[Titanio]

A per pixel distance is a very nice, and I take from the RT that these are being RayTraced which sounds cool!

As in ray-casting for visible surface determination?

 

[Bobbler]

A per pixel distance is a very nice, and I take from the RT that these are being RayTraced which sounds cool!

I think the RT was for render target.

 

[Titanio]

I think the RT was for render target.

That does make a little more sense

 

[Shifty Geezer]

I think the RT was for render target.

Ahhhh.

 

[superguy]

Phil that part was MADE UP. Understand that article was created from conversations here at B3D. Nobody here (developer wise) has ever said that a second SPE will be used for special OS calls. That job would go to the first and only SPE reserved for the O.S.

What's your proof of this?

The PS3Portal article does not mention any details about it's sources.

For now it's PS3portal against Mckmass as far as that goes. I think again for now it's a legit bit of speculation (to be taken with a grain of salt).

Also: http://arstechnica.com/news.ars/post/20060413-6600.html

 

[one]

What's your proof of this?

The PS3Portal article does not mention any details about it's sources.

For now it's PS3portal against Mckmass as far as that goes. I think again for now it's a legit bit of speculation (to be taken with a grain of salt).

Also: http://arstechnica.com/news.ars/post/20060413-6600.html

http://www.beyond3d.com/forum/showpost.php?p=737728&postcount=280
So you posted in the very same page as this post by Barbarian and still missed it
http://www.beyond3d.com/forum/showthread.php?t=29818&page=12
http://www.beyond3d.com/forum/showpost.php?p=737779&postcount=288

 

[Mmmkay]

What's your proof of this?

The PS3Portal article does not mention any details about it's sources.

For now it's PS3portal against Mckmass as far as that goes. I think again for now it's a legit bit of speculation (to be taken with a grain of salt).

Also: http://arstechnica.com/news.ars/post/20060413-6600.html

Okay, firstly we're talking about a blog article, so lets put things into perspective. Secondly, most of the information was harvested from two threads on this forum. Beyond3D is the source, for better or worse.

The 2 SPE info was sourced from this very thread and the person who provided the information is not known for being a PS3 insider (or being even remotely objective when it comes to Sony). The Ars article at least acknowledges that the claims were refuted by a PS3 developer (again, in this thread).

 

[superguy]

Ok, point taken. Looks like no second SPE taken by OS (Barbarian is a dev I assume)

Okay, firstly we're talking about a blog article, so lets put things into perspective. Secondly, most of the information was harvested from two threads on this forum. Beyond3D is the source, for better or worse.

Okay I'm believing now, nonetheless I dont see any proof the PS3portal article was culled from B3D.

Just two people claiming it. Not even reasons how they know this.

The Ars article at least acknowledges that the claims were refuted by a PS3 developer

It doesn't say they were refuted by a PS3 developer. It says:

Update: There's a claim in a newer Beyond3D thread that I'd missed to the effect that Sony is not considering having the OS periodically demand a second SPE.

In fact if the blogwriter was using B3d as his source..that makes no sense because while B3d had SPECULATED about the SPE reserved, it had never been confirmed here. In fact Titanio and the like would always be sure to call it "only speculation" or something like that.

It was only confirmed here AFAIK AFTER the blog article more or less spilled the beans!

 

[Crossbar]

It doesn't say they were refuted by a PS3 developer. It says:

Update: There's a claim in a newer Beyond3D thread that I'd missed to the effect that Sony is not considering having the OS periodically demand a second SPE.

In fact if the blogwriter was using B3d as his source..that makes no sense because while B3d had SPECULATED about the SPE reserved, it had never been confirmed here.

Now the interesting question: Who is the tail in this tail-recursive loop?