PowerVR - Eurasia

[Ailuros]

Considering that die space would still be a headache, what on God's green earth would you theoretically need on a TBDR extra ram for, when it already has an on-chip framebuffer and the external memory fetches are minimal, especially with FSAA or even FSAA + float HDR combinations?

 

[PC-Engine]

Considering that die space would still be a headache, what on God's green earth would you theoretically need on a TBDR extra ram for, when it already has an on-chip framebuffer and the external memory fetches are minimal, especially with FSAA or even FSAA + float HDR combinations?

Huh? 1T-SRAM comes in two flavors, standard and embedded. I was talking about the standard variety.

 

[SiBoy]

random thoughts:

*does ATI have anything on the horizon that can compete with Eurasia in the power-sensitive handheld market?

Does Eurasia compete in the "power-sensitive handheld market"?

And forget about Eurasia, what's the MBX core's real power playing a game? All I have seen is marketing numbers - and Nvidia's marketing numbers, for example, are as low as PowerVR's marketing numbers. Regardless of the fact that one is a TBDR and one isn't.

I'd love to see an apples-apples comparison of a TBDR vs. IMR to see if the claimed power advantage is real.

Likewise everyone is talking about shaders on mobiles, and no one ever mentions the power. Programmability has to come at a price.

 

[Lazy8s]

SiBoy:

All I have seen is marketing numbers - and Nvidia's marketing numbers, for example, are as low as PowerVR's marketing numbers.

MBX is more capable, though.

 

[archie4oz]

Could you elaborate a bit more...

 

[Lazy8s]

Similarity between GoForce and MBX power consumption doesn't discredit the power saving claims of tile accelerated display list rendering considering the graphics performance advantages favor MBX -- fillrates, programmability, feature set, image quality, etc.

 

[Ailuros]

Similarity between GoForce and MBX power consumption doesn't discredit the power saving claims of tile accelerated display list rendering considering the graphics performance advantages favor MBX -- fillrates, programmability, feature set, image quality, etc.

Assuming power consumption is on a similar level between sollution X and Y and functionalities and performance is times higher on Y, then power consumption should obviously be considered with those conditionals added to the comparison.

In the given case it would be Intel 2700G (MBX Lite/noVGP) vs. GoForce, Imageon etc.

All I have seen is marketing numbers - and Nvidia's marketing numbers, for example, are as low as PowerVR's marketing numbers. Regardless of the fact that one is a TBDR and one isn't.

Polygon rates from GoForce devboards I've heard this far are miles apart from what I'd call encouraging.

Likewise everyone is talking about shaders on mobiles, and no one ever mentions the power. Programmability has to come at a price.

I doubt any serious contender for the second generation of PDA/mobile chips will be able to skip programmability, whereby programmability should be always considered for that particular market and the obvious limitations.

 

[archie4oz]

Similarity between GoForce and MBX power consumption doesn't discredit the power saving claims of tile accelerated display list rendering considering the graphics performance advantages favor MBX -- fillrates, programmability, feature set, image quality, etc.

GoForce or GoForce3D? The latter is a far more impressive part than the former... I wouldn't even consider the former in the same category as the MBX.

 

[Lazy8s]

There isn't any renderer of a more conventional variety, anything from nVidia or anyone else, that casts doubt on the effectiveness of PowerVR's power conservation. Falanx's hybrid-based Mali architecture reinforces such cost saving notions from the tile accelerated side of the approach.

 

[SiBoy]

There isn't any renderer of a more conventional variety, anything from nVidia or anyone else, that casts doubt on the effectiveness of PowerVR's power conservation. Falanx's hybrid-based Mali architecture reinforces such cost saving notions from the tile accelerated side of the approach.

Oh please. Let's see some numbers.

 

[Ailuros]

Similarity between GoForce and MBX power consumption doesn't discredit the power saving claims of tile accelerated display list rendering considering the graphics performance advantages favor MBX -- fillrates, programmability, feature set, image quality, etc.

GoForce or GoForce3D? The latter is a far more impressive part than the former... I wouldn't even consider the former in the same category as the MBX.

http://www.nvidia.com/page/handheld.html

I personally meant with GoForce in my former post AR10, which is GoForce3D actually.

GoForce3D vs. MBX Lite/sans VGP then.

 

[Lazy8s]

Efficiency of power usage is just the amount of performance for the amount of power consumed. MBX gets extra performance from the 66% savings on consumptive pixel drawing that display list rendering yields over early Z on average and the similar savings in consumptive bandwidth that tile accelerated rendering delivers over conventional methods.

 

[SiBoy]

Efficiency of power usage is just the amount of performance for the amount of power consumed. MBX gets extra performance from the 66% savings on consumptive pixel drawing that display list rendering yields over early Z on average and the similar savings in consumptive bandwidth that tile accelerated rendering delivers over conventional methods.

That's nice marketing-speak and all, but I'd like to see measured power on the same content for an SoC using MBX vs. an AR10, etc.

 

[Lazy8s]

It's demonstrated by the 4xAA, 100% fillrate effectiveness, internal precision for color blending and floating-point depth sorting, lighting included with its transform set-up, and SM 1.1 vertex shading that MBX applications have.

 

[Ailuros]

It's demonstrated by the 4xAA, 100% fillrate effectiveness, internal precision for color blending and floating-point depth sorting, lighting included with its transform set-up, and SM 1.1 vertex shading that MBX applications have.

Power consumption between MBX Lite (sans VGP) could be on the same level as AR10. The VGP obviously will raise gate count and in extension power consumption.

Assuming power consumption is in the above case on the same level, then it comes down to what exactly comes out at the other end; something tells me that one of the two will "look like a**"(tm) and you don't even need to factor in AA, since it's limited to only SSAA on the Lite model and should come at a sizeable fill-rate cost, especially with 4x samples.

 

[Lazy8s]

GoForce3D advertises 50-mW consumption on a 130-nm process at its most demanding output. The effective fillrate with 4xAA would still be good at only an advertised 75-mW for a 50-MHz MBX HR-S with VGP on 130-nm.

 

[Ailuros]

GoForce3D advertises 50-mW consumption on a 130-nm process at its most demanding output. The effective fillrate with 4xAA would still be good at only an advertised 75-mW for a 50-MHz MBX HR-S with VGP on 130-nm.

Forget advertised numbers (most of the times estimates); it's the real numbers that count on finalized products and I haven't seen any for either/or yet.

 

[TEXAN]

What's an AR10?

 

[Lazy8s]

AR10 refers to a first generation nVidia GoForce3D processor, the GoForce3D 4500 used in the Gizmondo.

The only real battery life figures documented for GoForce3D and MBX seem to be those for the GoForce3D 4500 in the Gizmondo and the MBX Lite in the Dell Axim x50v. The two devices differ quite a bit in implementation, so the numbers aren't directly comparable. With an 1100-mAh battery, the Gizmondo incorporates its graphics core at about 75-MHz with a 400-MHz ARM9 CPU and 240x320 TFT screen to rate at around 3 hours of continuous play at full stress. With an 1100-mAh battery, the Dell Axim x50v runs its graphics core at around 67-MHz (maybe up to 75-MHz) with a 624-MHz Intel ARM XScale CPU and a 640x480 display to rate at about 4.5 hours of continuous play at full stress.

 

[Loewe]

With an 1100-mAh battery, the Dell Axim x50v runs its graphics core at around 67-MHz (maybe up to 75-MHz) with a 624-MHz Intel ARM XScale CPU and a 640x480 display to rate at about 4.5 hours of continuous play at full stress.

No way.
The Axim x50v run with the normal 1100 mAh battery at full stress 2 hours. I have tried it yesterday evening with full speed running my earthbench in a new not published version.