Mali200 from Falanx - new challenger for desktop markets or pure utopia?

[Kaotik]

http://www.hardocp.com/article.html?art=ODAy

Basicly, for what I understand, Falanx thinks of their Mali200 as possible new challenger even for current desktop cards, in form of motherboard integrated graphics chip coming on second half of 2006 (of course current cards aren't the top tier H2/06, but still it sounds impressive)

Interestigin tidbit picked from the article:

So how many pixel pipelines does this Mali200 technology have? Prepare to once again think outside of the box, as our answer is 1. Instead of building redundant shader pipelines as we see in current GPU technology, think of Mali200 as a stackable IP. Need more graphics power? Just add more Mail200 units. Its modular nature allows expanding the powers of Mali200 as you need them.

How should this be understood - that you can have anything from 1 to x^n pipelines, what ever you just want to build?
Can this approach actually work at all to build anything even close to current high end cards?

They also claim 2 AA modes - 4xAA for free, and 16xAA as alternative.

The feature set is said to exceed SM3.0

 

[neliz]

The first thing that springs to mind flipping over the article "looks good on the outside, S3 on the inside"

It just sounds a bit too good to be true, many times smaller die than current graphics offering, power consumption is a hundred times less but still it finds ways to output images on a better level than anything before.

It will support just about anything out there but still NOT use third party licenses.

I saw this nice guetimate at hardocp:

A very rough estimate but take the Mali200 vs 7800GTX pixel performance (No idea of shader performance on either, or geometry performance on 7800GTX... so this is all with a grain of salt.)

Mali200 = 300Mpix @ 200mhz
7800GTX = 10.32Bpix @ 430mhz

Bump the speed by 2.15 so the Mali is on part MHZ wise with 7800 (Mali 645Mpix @ 430)
10320Mpix / 645Mpix = 16 cores. Completely rough estimate... and I seriousally doubt it would only take that many, but I couldn't imaging 50 to 80 cores or pipelines on one chip.

Course, we could always quote Bill Gates too, so anything is possible.

 

[Arun]

Hopefully a certain engineer here will dare to comment on this
As for that comparaison, it misses the point completely neliz. It's partly deferred, and it's extremely scalable. I would guess that it doesn't NEED to have more fillrate considering the deferred approach, but I could be very wrong on this. Anyhow, whoever you quoted is quite clueless about this.

it will most certainly be an interesting competitor to the SGX however; I'd love to see actual comparaisons one day or another, when both are actually buyable somehow I would still tend to believe the SGX might be more interesting for integrated parts, however, because it's fully deferred and has an unified pipeline. The coolest part of the Mali200 - and that's BIG for mobile phones imo - is the free 4x AA. But people who buy integrated solutions might sadly not notice these kind of things as much.

Still, it could performance way better per-transistor than the SGX disregarding AA, or it could perform worse. We just don't know, and the architectural details we have on both parts aren't exactly heavy. Ah well, hopefully [insert name] isn't on a strict NDA about this.


Uttar
P.S: Regarding pipelines, there never is more than ONE *shading* pipeline on Mali200. I don't know about ROPs etc. though, sadly. Obviously the integrated solution would still only have a single pipeline, but it'd also have a LOT more functional elements than one that is supposed to work with just a few mW, rather than Watts.

 

[neliz]

The product sheets focus heavily on the mobile market so maybe we should compare this product with the goforces and imageons of this world?

Goforce3d, 100Mpixels/second 1,5Million triangles at at 130nm but at which clocks?

 

[Arun]

Stop missing the point - GoForce 3D 4800 and Imageon have been announced a while ago, it's basically a generation older than Mali200. The only part that's roughly comparable is the PowerVR SGX since it was announced in the same timeframe imo. This is not a market where newly announced products are released overnight.

And yes, the released specs are to be compared with the such parts, just newer ones. But the idea is that the chip would (supposedly) be sufficiently scalable for a *modified* version to be ideal for high-performance integrated graphics too; I've got my doubts about this, but then again I'm not competent to judge such details. It does seem plausible to me, though.


Uttar

 

[Ailuros]

The coolest part of the Mali200 - and that's BIG for mobile phones imo - is the free 4x AA.

Falanx claimed even for their first cores single cycle 4xRGMS and I've no reason not to believe them. 16xRGMS obviously cost quite a bit in fill-rate since it'll require 4 cycles.

In any case usually on a TBDR Multisampling is in a relative sense fill-rate and bandwidth free.

Every time I hear about free AA from any company my head starts spininng. While I do realize that these are power critical small embedded devices and there are obvious limitations, antialiasing doesn't start and end at polygon edges/intersections. Either there is a fast adaptive AF algorithm present to complement MSAA (yes even 2x samples would be enough as a starter) or give me Supersampling instead.

Any of those companies that have managed to complement their Multisampling algorithms with a fast performing adaptive AF algorithm will have the obvious winner.

SGX claims programmable multisampling for free (insert whatever description you want for "programmable" since I've no idea what it stands for).

 

[Ailuros]

The product sheets focus heavily on the mobile market so maybe we should compare this product with the goforces and imageons of this world?

Goforce3d, 100Mpixels/second 1,5Million triangles at at 130nm but at which clocks?

1.5 M Tris? Just were for heaven's sake? LOL

 

[Arun]

SGX claims programmable multisampling for free (insert whatever description you want for "programmable" since I've no idea what it stands for).

I would assume that to be roughly similar to what ATI is doing: sampling patterns are configurable on a per-frame basis (this feature is controlled by the driver, at least on ATI GPUs). This idea was introduced in the R300, and it's what allowed things like TSAA to work in the X800 Series.

Uttar

 

[neliz]

The point? so far the Mali200 is the only point.. a very discrete product for mobile "gaming" which in the future might be able to be set-up in a multi core setup.
Right now we should be realistic and look at a product which they WANT to launch in H2 06 and that is a single core, low power chip which does feature an awfully big set of specifications.
If they do plan on designing and building a multi core setup, what would be the timeframe for that? H2 2007?, 2008 before we see a marketable product launch?

 

[Ailuros]

Albeit product cycles for high end mobile phones are relatively long (about 2 years or so), the smaller IP companies are, the more they have to struggle to execute on time.

The first Mali cores where beyond doubt impressive and with quite promising features/capabilities on paper. What I'd like to know is if any of those have been integrated yet in SoCs.

Oh by the way Uttar, it's arjan de lumens if my memory serves me well

 

[Arun]

Ailuros: The official NVIDIA PR claims 1.5 million triangles sustained for the GoForce 3D 4800 and up. Link: http://www.nvidia.com/object/IO_18437.html (search for "triangle" in Acrobat Reader)
Neliz:

If they do plan on designing and building a multi core setup, what would be the timeframe for that? H2 2007?, 2008 before we see a marketable product launch?

Reread what I said. It is not a multicore design that would be used for that. Please, have pity on my soul... Also, they don't sell Chips. They sell IP. Other companies integrate that into SoCs, along with other IP from other companies. And if you licensed the Mali200, I would assume you have the right to use as many functional blocks as you see fit, in which case it might be suitable for integrated graphics, unless HardOCP misunderstood something, which is still possible at this point.
It's not Falanx that would be designing a more powerful Mali200-based design, it's whoever bought the IP from them for it (if anyone), and they would only receive "assistance" from Falanx for it, I would assume.

Uttar
Ailuros: I know, but I don't like summoning engineers by name Plus I figured that if he couldn't comment on this, he wouldn't want multiple people asking for it.

 

[Ailuros]

I know what they claim, as of course do I know why I have every reason to doubt it; just like it's claimed pixel shader or geometry processor.

 

[Ailuros]

I would assume you have the right to use as many functional blocks as you see fit, in which case it might be suitable for integrated graphics, unless HardOCP misunderstood something, which is still possible at this point.
It's not Falanx that would be designing a more powerful Mali200-based design, it's whoever bought the IP from them for it (if anyone), and they would only receive "assistance" from Falanx for it, I would assume.

Yes of course would Falanx or any other company provide the full design. That's what repeat licenses are for.

Mali200 has an optional VS/Geometry processor. Could be the source of the misconception here.

 

[neliz]

When questioned about Mail200 performance we were told that Mali200 could be offering much more efficient integrated graphics by the second half of 2006 that would be "on par" with the current add-in board graphics processing units of today.

The Mali200â„¢ architecture allows for multiple cores to be parallelized in larger footprint SoCs for desktop and integrated graphics processing applications.

So, in eight months, we'll see a product that is compareable with Integrated x300's and fx5200's , where some soc has to spend a lot of time getting it to work.. and work on the software.

http://www.hardocp.com/image.html?image=MTEyMzAwNzYxNVB0Y0tNT3JyM05fMV83X2wuanBn

Besides the really really 3rd part IP box (can't they draw a straight line??) what's up with "Falanx Software Vertex Shaders" they obviously haven't got everything integrated in the mali200 as they need the arm11 for hardware VS.

 

[Ailuros]

So, in eight months, we'll see a product that is compareable with Integrated x300's and fx5200's , where some soc has to spend a lot of time getting it to work.. and work on the software.

It rather claims "on par" with add in boards. That would be more like low end GPUs of today.

Besides the really really 3rd part IP box (can't they draw a straight line??) what's up with "Falanx Software Vertex Shaders" they obviously haven't got everything integrated in the mali200 as they need the arm11 for hardware VS.

Two different core proposals for different functionalities. Read the text underneath; left is lower end since it's for feature phones and right is the higher end sollution since it's for high end mobile phones.

 

[Nappe1]

Oh by the way Uttar, it's arjan de lumens if my memory serves me well

Your memory serves you quite well.
This forum has also member that has first hand experiences from Bitboys G-series, but as long as he does not want to introduce himself on that light, I am not the one revealing who he is.

anyways, back to the topic...
Mali200 seems to be quite flexible indeed, though FalanX has taken complete different approach than PowerVR to next gen. (PowerVR has clear 3 basic models to be customized for customer, FalanX has bunch of units that can be combined several ways to reach the target.) It is still too early to say which one takes the cake here, but as I said on the SGX thread, the others haven't been just sitting fingers crossed.

For hand held console manufacturers this is like day in the candy shop as a kid: no problem with available designs to choose from.

 

[TEXAN]

The MBX could do 4X FSAA.

http://www.khronos.org/devu/library/sf_2005/platforms/TI_-_OMAP_and_Linux.ppt

 

[Ailuros]

The MBX could do 4X FSAA.

http://www.khronos.org/devu/library/sf_2005/platforms/TI_-_OMAP_and_Linux.ppt

2x horizontal MSAA + 2x vertical SSAA afaik. 2xMSAA comes for free.

 

[psurge]

Is there any information floating around on what exactly Falanxes "partially deferred" rendering scheme entails? IIRC, PowerVR supports something which may be similar
(reserving a fixed size buffer for tile lists and rendering them on overflow).

SGX and Mali200 appear to be worthy contenders for eachother - kudos to PVR and Falanx for designs that are much more interesting/innovative than what we're seeing in the highend desktop segment (now if only we could get some more info on them ).

 

[TEXAN]

Your memory serves you quite well.
This forum has also member that has first hand experiences from Bitboys G-series, but as long as he does not want to introduce himself on that light, I am not the one revealing who he is.

anyways, back to the topic...
Mali200 seems to be quite flexible indeed, though FalanX has taken complete different approach than PowerVR to next gen. (PowerVR has clear 3 basic models to be customized for customer, FalanX has bunch of units that can be combined several ways to reach the target.) It is still too early to say which one takes the cake here, but as I said on the SGX thread, the others haven't been just sitting fingers crossed.

For hand held console manufacturers this is like day in the candy shop as a kid: no problem with available designs to choose from.

As long as PowerVR is the only one who is doing TBDR they shall remain untouchable in this mobile realm.

Anybody can do TBR, anybody can do it via developing their own methods by not touching img patents, but you have to realize TBDR is where the real magic happens, this is where PowerVR is set apart and make no mistake this has been succesfully patented worldwide.

This reason and this reason alone is why nobody shall be able to touch Img and it's PowerVR line from total domination of the entire scene.