PowerVR? Why the excitment?

[Quitch]

I see quite a few mentions on this board of PowerVR. I remember when they were competing with the first Voodoo... having to make the choice, reading that one was best on lower machines because of its memory, while the PowerVR would take off on a 166 and above.

Anyway, I never really knew much then (what was the truth back then?). I did wait on their second card for the PC... but it never came, and they made graphics for something else, yes? Didn't they also then put out the Kyro?

Anyway, I see people get excited about PowerVR? Why? I'm new to keeping up with this industry (only really started paying attention with the FX). What exactly is so good about them?

 

[Mariner]

Anyway, I see people get excited about PowerVR? Why? I'm new to keeping up with this industry (only really started paying attention with the FX). What exactly is so good about them?

You'd have to ask Kristof and Simon F about that, I expect.

Didn't Beyond3D originally start out as Kristof's PowerVR fan-page?

 

[Ailuros]

One of the primary reasons should be that they are the last ones remaining that are basing their architectures on Tile based deferred rendering. There's quite a mythos surrounding the strengths or even weaknesses, which has been a never ending topic for arguments for years now one way or another.

They claim to be working on a PC standalone graphics board that will aim for the high end market, which is a first time ever since the advent of 3D and probably a good chance to proove the claimed superiority and/or efficiency of their approach.

 

[Bouncing Zabaglione Bros.]

Anyway, I see people get excited about PowerVR? Why? I'm new to keeping up with this industry (only really started paying attention with the FX). What exactly is so good about them?

Simple answer: PowerVR are the only people doing Tile Based Deferred Rendering, one of the future trick rendering technologies. Where more traditional architechtures have to draw a lot of stuff that you never see or gets thrown away, TBDR never draws it in the first place. This give PowerVR a big boost because of this "free speed" of not having much overdraw.

Since PowerVR came out though, other companies have made strides forward with things like wider memory busses and early culling of stuff they need don't need to draw, but PowerVR are still the only people who have shipped, and are continuing to seriously persue TBDR architechtures.

Another interesting note is that in the last few years, all the main companies (Nvidia, ATI and even 3DFX) all bought small and expensive companies in order to gain access to their TBDR technology and patents. At some point in the past, today's main manufacturers were seriously looking at TBDR, and probably have the ability to implement it in the future if they need to.

 

[Captain Chickenpants]

I beleive (but may be wrong) that we are also the only ones who have actually produced and shipped a TBDR graphics card.
TBDR also has benefits for mobile applications as it is easily scalable thus enableing easier power/performancs tradeoffs, and can generally perform better with lower memory and clock speeds.

CC

 

[Gubbi]

Never mind the low power stuff. When are you coming out with high end competition busting part ?

I loved my K2

Cheers
Gubbi

 

[Captain Chickenpants]

I ain't sayin nuffin!


CC

 

[MfA]

PowerVR implementations were products which could have been a hit if only they didnt take so long to get to market (by the time they got to market it was easy to see from the transistor count and memory used that they were designed for a bygone era). I think that if they can get a product in the market which uses a good semiconductor-process/clock-rate/feature-set/memory-speed for its intended niche instead of using a combination of those which would have been optimal 6-9 months earlier they will provide incredible value for money.

I dont think it matters which niche that is as far as performance goes, I think they can compete in all, but as XGI said ... the bottom up approach isnt a good way to create brand recognition. Also low end products make execution even more important, high end migrates into low end. Low end just migrates into oblivion.

 

[David G.]

Because we really need a third competitor to bring prices back in the 299$ range @ product launch and 129$ in mainstream/performance .

XGI is probably doing some serious business there . SIS invested a lot of money . They bothered to create a new company .

They seem serious as they've probably noticed that their DX9 core wasn't doing good enough even if it had 8 pipelines so they started this whole new Dual thing just to prove that they want to bring some performance to their solutions .

BUT , I think that only on XGI's second generation of products we will be able to look @ them as a serious competitor for ATi and nVIDIA .

S3 ... I don't know ... VIA clearly tends to stick to low/mid end . That's what they do with their CPU's too ...

Matrox will not get a new VPU out next year ... I don't think that they will ever atempt to get some real 3D on their cards after the P512 fiasco . They's probably to their best thing : 2D & multi.

PowerVR are probably the only guys serious out there with a proven solution that worked and competed against nVIDIA and ATi solutions .

 

[Entropy]

PowerVR are probably the only guys serious out there with a proven solution that worked and competed against nVIDIA and ATi solutions .

3Dlabs, albeit in a different market segment.

Entropy

 

[David G.]

Yes ... I had great hope for the entry of 3DLabs into the gaming market under the guidance of Creative but it seems that's over and done now .

 

[gkar1]

My excitement is greatly dependent on the intentions of IMGtech. Do they still plan on being an IP company selling chip desings or have they decided to go the extra step and produce the chips themselves(or with a foundry partner) and then market them to OEM and AIB vendors ala Nvidia?

If the latter is true then who's their foundry partner and do they have any desing wins or any prospective buyers?

 

[Loek Frederiks]

I have seen an interview somewhere on the web clearly stating they are still alive and are appearently still working on a core for the PC market. I still think their strong point is what they call the MBX core, a graphics
core dedicated for the embedded market.

It's a pity STBs and mobile phones do not really require 3D yet. The MBX
is as also stated by someone else perfect for bandwidth limited (and power consumption) applications.

 

[Gubbi]

I have seen an interview somewhere on the web clearly stating they are still alive and are appearently still working on a core for the PC market. I still think their strong point is what they call the MBX core, a graphics
core dedicated for the embedded market.

I think PowerVR's (or is it IMGtec, I can never figure it out) way of building graphics processors lends itself better to going for multiple segments of the market (with the same technology) than more traditional immidiate mode renderers.

The traditional way of boosting performance in IMRs has been to add pixel pipelines. PowerVR did the same with the Kyro line of accelerators (2 pipelines). And they could indeed go the same way as IMRs with more and more pipelines.

But since the workload (scene) is divided into tiles by the frontend another possibility is to build a one pixel core (with DX9/10 capabilities), and then just add more of these processors onto a single die (like an über Voodoo 2 SLI on-a-chip). Caches and FIFOs would be scaled accordingly.

Each processor would then operate on it's own tile. Pros would be simple design and verification/test for new products with more cores. Cons are lower spatial locality (ie. youd need something like 8 x temporal storage for textures for a 8 core system)

Edit: This would also make it ideal for licensing the core to chipset vendors

Cheers
Gubbi

 

[Kristof]

Gubbi, IMHO, such a design would be a bad idea since a lot of functionality would be duplicated which should be shared, so your silicon utilisation becomes quite poor due to all the redundant logic... cost is still a key issue especially in integrated or embedded solutions.

K-

 

[Gubbi]

Well. I'm just speculating

I was under the impression that shaders and buffers made up the bulk of the Si.

Cheers
Gubbi