PVR to Bring Console Quality Gaming to Mobiles

[Fox5]

So do we know of any PDAs or any other devices that may be using these chips? I'd sure like one.

 

[Simon F]

On tonyh_2_logos.bmp, it seems that it lacks texture perspective correction on the ceiling.
Is that a software or hardware issue? In this case, of course.

It'd have to be a software issue. It believe it was a quick port from some other games console.

 

[Ailuros]

Alright, that's more like it, some specs. The high-end is a little higher in geometry, but what are we talking about in an actual gaming environment? How much of that power can it maintain?

Guess that really depends largely on the CPU coupled with the device, CPU/GPU dependancies, engine design, and concern for power consumption. Could be interresting though, to say the least.



Later

Iridius Dio

I'm not sure about the clock frequencies of the MBX range (MBX Lite to MBX PRO), yet from the fill-rate numbers quoted I'd guess in between 60-150MHz.

The VGP is VS1.1 compliant and looks in detail from an older ARM document like this:

At say 150MHz that would equal 800M FLOPs/sec. The VGP is an optional unit and the MBX Lite models can also be accompanied by a VGP Lite (I'm not sure if that one can too sustain 4M FLOPs/MHz).

According to the following the MBX Lite (without a VGP) running at 66MHz can sustain 30Hz with Vsync on in Q3a in mobile resolutions, and that's really the lowest end part, even more lacking the geometry unit:

http://users.otenet.gr/~ailuros/q3aMBX.jpg

 

[Simon F]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

 

[Panajev2001a]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

Well, but if that operation in particular is hard-coded we would not gain more than 4 FP ops/s performance when doing Matrix vs Vertex or dot products and other similar operations.

 

[Simon F]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

Well, but if that operation in particular is hard-coded we would not gain more than 4 FP ops/s performance when doing Matrix vs Vertex or dot products and other similar operations.

Yes... that's what I just said. The spec quoted above is somewhat conservative.

[Edit] AH! Just re-read what you actually wrote. I assume you actually meant to write; "but if that operation in particular is hard-coded wouldn't we gain more?"

 

[Ailuros]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

Even better. No idea if you can answer that question yet I'll still try:

Are the specifications between VGP and VGP Lite identical as being illustrated above?

 

[Simon F]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

Even better. No idea if you can answer that question yet I'll still try:

Are the specifications between VGP and VGP Lite identical as being illustrated above?

Well, obviously, there are some differences or else it wouldn't be suffixed with "lite" (Ughh. I hate that spelling).

I've only programmed the MBX VGP but I think the only only difference is that performance is lower... but don't quote me

 

[Panajev2001a]

It actually can do higher than 4FLOPs/cycle .. the 4x4 matrix mul in 4cycles should be evidence of that.

Well, but if that operation in particular is hard-coded we would not gain more than 4 FP ops/s performance when doing Matrix vs Vertex or dot products and other similar operations.

Yes... that's what I just said. The spec quoted above is somewhat conservative.

[Edit] AH! Just re-read what you actually wrote. I assume you actually meant to write; "but if that operation in particular is hard-coded wouldn't we gain more?"

Yes, but in the case of that particular operation being quite hard-coded then in all other kinds of work-loads (not Matrix * Matrix operations) we would not gain more (unless what you hard-coded in the 4x4 Matrix multiplication can help multiplying a Matrix with a Vertex).

 

[Simon F]

Yes, but in the case of that particular operation being quite hard-coded then in all other kinds of work-loads (not Matrix * Matrix operations) we would not gain more (unless what you hard-coded in the 4x4 Matrix multiplication can help multiplying a Matrix with a Vertex).

I am referring to DP4.

 

[Ailuros]

So do we know of any PDAs or any other devices that may be using these chips? I'd sure like one.

Strike 1: Intel Carbonado

Intel demonstrated a new PDA running 3D games using an Intel 2700G multimedia accelerator at their technology showcase. The new Intel 2700G multimedia accelerator is a low-power graphics accelerator for mobile PDA devices. It supports 3D graphics with the ability to process nearly 1 million polygons per second - ideal for handheld 3D games. It also supports dual-display which makes it very flexible for presentation on large screen monitors. Besides that, it also support MPEG-2, MPEG-4 and WMV9 formats.

http://www.hardwarezone.com/articles/view.php?id=1240&cid=18&pg=6

 

[Simon F]

So do we know of any PDAs or any other devices that may be using these chips? I'd sure like one.

Strike 1: Intel Carbonado

FWIW, the 3D core is an MBX-Lite.

 

[Ailuros]

I think you'll get the hang out of that spelling sooner or later

 

[Nite_Hawk]

Kristof/Simon:

Are there/Will there be OpenGL drivers ready for the eventual linux ports for PDAs that use this gfx chip?

Nite_Hawk

 

[Rodéric]

Kristof/Simon:

Are there/Will there be OpenGL drivers ready for the eventual linux ports for PDAs that use this gfx chip?

Nite_Hawk

OpenGL|ES likely, since PowerVR is member of the Khronos group.