[B3D News]

Qualcomm has announced new chips that integrate their proprietary Scorpion CPU (ala Cortex-A8) and ATI's OpenGL ES 2.0 GPU (which they now own). They're on TSMC's 45nm process, but unlike some of their other chips they do not integrate RF. That is, if the other ones did in the first place.

Read the full news item

[Wishmaster]

It would seem that with all those announcements recently qualcomm tries to regain the lost trust. Which is a good thing. Maybe they will open more their solutions so that community could use their hardware. Of course if it's not just a pile of PR BS
What they announce is very impressive and I think competitive with TI and nvidia's offering. In terms of CPU performance it is the best the market has to offer (not considering cortex A9) although it might not be the most power efficient. Multimedia experience on the paper seems nice (1080p playback/record, 5.1 sound support only integrated HDMI is missing) but will it support more sophisticated codecs like VC-1 and h.264@high profile?

In the press release they announced that planned sampling of the lowest variant will be mid-2009. Does that mean that they will have working version of the chip then and it will be available to ODM or does that mean that it will be available in devices by mid-2009?

I remember that year ago (11th February 2008) they announced new chips and I think they were later used in msm7201 and first devices using msm7201 were unveiled mid-2009 like HTC Touch Diamond. That is why I'm confused what they mean by sampling...

Now that's a long comment I hope that with time I get answers to my questions. Maybe some will come from MWC. For now I hope that we will see more chips like that from more manufacturers

[Ailuros]

TI doesn't have yet any 45nm cores and nothing from the SGX520 family of cores yet. The latter is also a single TMU/shader pipe core and IMG has estimated it at 65nm to be around 2.6mm2 rated at 7.0 MTris/s @200MHz with <50% shader load.

The current 65nm OMAP3xx0 SoCs contains a 65nm SGX530 which at 110MHz (I think) supposedly delivers 10 MTris under same conditionals as above.

[roninja]

Ailuros - what about TI's 36xx series...just google for OMAP3640 and tell me what you can see...the TI website had OMAP's 3610 to 3640 listed for a short time as 45nm but these were then taken down...

[roninja]

Did TI licence SGX520? My guesses were 530 for OMAP3xxx series and 540 for OMAP4xxx but thats based on zero clues so far..

[Lazy8s]

Well over a year is usually taken to go from the sampling of the chip to the shipping of devices based upon it.

[wishiknew]

Originally Posted by roninja

Did TI licence SGX520? My guesses were 530 for OMAP3xxx series and 540 for OMAP4xxx but thats based on zero clues so far..

OMAP4 announced with the 540 today. But god how many years do we have to wait before it gets into phones. EDIT Oh powervr got a 543 now.

[Lazy8s]

The 45nm OMAP36xxs should help fill the gap by at least a few months.

[Ailuros]

Originally Posted by roninja

Did TI licence SGX520? My guesses were 530 for OMAP3xxx series and 540 for OMAP4xxx but thats based on zero clues so far..

No idea. If however TI is planning a smaller than OMAP3xxx SoC it would be more than reasonable to use an about half as small core as 520 (compared to 530) for that one.

Originally Posted by wishiknew

OMAP4 announced with the 540 today. But god how many years do we have to wait before it gets into phones. EDIT Oh powervr got a 543 now.

540 is IMHLO for high end, 530 for mainstream. SGX543 doesn't sound like it'll appear sooner than about 2 years from now in any devices and I don't necessarily expect to see a core with roughly twice the throughput (except texel fillrates) than a 540 in mobile phones.

[roninja]

http://focus.ti.com/general/docs/wtb...ontentId=53243

website says SGX540 in the recently announced OMAP4's

[Ailuros]

Does anyone know what Sony Ericsson's Idou contains exactly? It sounds rather a multimedia only applications processor, but has anyone further details on it?

[roninja]

My "guess" would be it packs the EMP U380 (Ericsson baseband + OMAP3430)...

[Ailuros]

Originally Posted by roninja

My "guess" would be it packs the EMP U380 (Ericsson baseband + OMAP3430)...

It's weird that there's not a single hint anywhere (unless I've missed something) about 3D.

[Arun]

Idou is the greatest paper launch ever AFAIK, it think it also was the device announced at MWC that had the least amount of public showcasing... (just presentations done by SE staff at the booth) - so while interesting in theory, I wouldn't bother speculating about specs too much just yet.

[Wishmaster]

Arun do you know anything about openMAX being used in any upcoming SoC?
It seems to be a good way of unifying video acceleration similar to what openGL did for graphics. After all that was probably the purpose for it in the first place.

I've recently found job offer on qualcomm's site for HD video encoder and it said that knowledge of openMAX is a plus (it wasn't the only offer in which openMAX was mentioned). Could it mean that upcoming qcom chips will use openMAX and maybe finally we wouldn't have to worry about not being able to use hardware acceleration and that we could write our own more efficient drivers for DSP?

TI's DSP seems already openMAX compatible and ready but what about nvidia and their tegra platform?

[Arun]

NV was actually one of the first (if not the first?) to implement OpenKODE and OpenMAX, and all multimedia functions on Tegra go through it AFAICT. The reason they've been so aggressive at that is because they have a few high-up people at Khronos (especially Neil Trevett), so they tend to be a pretty big supporter of everything they do.

I'm not sure how OpenMAX would allow you to write your own DSP drivers though?

[Wishmaster]

Originally Posted by Arun

I'm not sure how OpenMAX would allow you to write your own DSP drivers though?

My bad.
What I wanted to say was that thanks to openMAX we could write codec (driver, set of instructions, et al.) for our own video player using e.g. NEON. It would be good for all those situations where we have underpowered video player cause we can't access h/w acceleration but we can use neon and improve the performance. AFAIK that's how can works right?

[Captain Chickenpants]

Originally Posted by Wishmaster

My bad.
What I wanted to say was that thanks to openMAX we could write codec (driver, set of instructions, et al.) for our own video player using e.g. NEON. It would be good for all those situations where we have underpowered video player cause we can't access h/w acceleration but we can use neon and improve the performance. AFAIK that's how can works right?

Pretty much, there are several layers at which you could accelerate things under OpenMax.
You could write an OpenMax IL component which is pretty much writing a full encoder that accelerates (or decelerates )the whole encode process.

Or you could write OpenMax DL primatives which accelerate specific parts of the encode/decode process e.g. DCT / Motion Estimation. It is this that you would probably look at doing with a DSP or a specialised processor instruction set.

CC

[Wishmaster]

Originally Posted by Captain Chickenpants

Pretty much, there are several layers at which you could accelerate things under OpenMax.
You could write an OpenMax IL component which is pretty much writing a full encoder that accelerates (or decelerates )the whole encode process.

Or you could write OpenMax DL primatives which accelerate specific parts of the encode/decode process e.g. DCT / Motion Estimation. It is this that you would probably look at doing with a DSP or a specialised processor instruction set.

CC

Ok thanks for your thorough answer

[Laurent06]

Originally Posted by Captain Chickenpants

Or you could write OpenMax DL primatives which accelerate specific parts of the encode/decode process e.g. DCT / Motion Estimation. It is this that you would probably look at doing with a DSP or a specialised processor instruction set.

BTW ARM provides several implementations of OpenMAX DL:

1. a C one
2. an ARMv6 one
3. an ARMv7 one using NEON

More info here: http://www.arm.com/products/multimed...max/index.html

[Wishmaster]

Arun I've got a question for you
How hard can it be to make scorpion CPU have OoOE just like cortex A9?

Probably it would be easier to use A9 and don't try to change scorpion but maybe it would benefit from OoOE and have some advantage over A9...

[silent_guy]

Originally Posted by Wishmaster

Arun I've got a question for you
How hard can it be to make scorpion CPU have OoOE just like cortex A9?

Probably it would be easier to use A9 and don't try to change scorpion but maybe it would benefit from OoOE and have some advantage over A9...

Pretty much impossible. OoOE determines the whole architecture of a CPU. It's not something that you can just bolt on.

[Arun]

As silent_guy said, it'd pretty much be an entirely new CPU generation; OoOE is one of the most fundamental changes you can make to a CPU design... I have no idea whether they plan to invest in a new generation, obviously the back-end guys were focusing on 45nm but I have no idea what the architectures ones might be working on for the 28nm timeframe (if anything).

BTW, it feels pretty good being spot on (in the original news piece) - now if only someone else didn't have the scoop straight from Qualcomm even before I speculated about it, bah... http://www.linleygroup.com/Newslette.../lm090129.html

Talking of the A8/A9/Scorpion, I stumbled upon this case of a pretty epic fail with NEON: http://infocenter.arm.com/help/index...h16s05s02.html - while not a big deal for some things, it seems to me that it'd make it harder to benefit much from NEON in a game engine. Apparently will be improved in the A9, I wonder how bad it is on Scorpion...

And finally one quick off-topic note: looks like I was wrong in one of my posts about OMAP3, the U380 SoC from Ericsson Mobile Platforms (which might or might not have been canned following the merger with ST-NXP) was on 65nm, not 45nm. So what I heard there was clearly wrong. That sadly gives me a hunch I might have been way too optimistic about the Motorola SoC which I speculated was therefore also 45nm; who knows though. BTW, the STn8820 was also 100% canned to focus on the U8500, and the U500 also seems to have been canceled.

[Wishmaster]

Originally Posted by Arun

BTW, it feels pretty good being spot on (in the original news piece) - now if only someone else didn't have the scoop straight from Qualcomm even before I speculated about it, bah... http://www.linleygroup.com/Newslette.../lm090129.html

So I was right after all that qcom will try to put scorpion almost everywhere so that they could benefit from the hard work they done on designing it

Talking of the A8/A9/Scorpion, I stumbled upon this case of a pretty epic fail with NEON: http://infocenter.arm.com/help/index...h16s05s02.html - while not a big deal for some things, it seems to me that it'd make it harder to benefit much from NEON in a game engine. Apparently will be improved in the A9, I wonder how bad it is on Scorpion...

Maybe when they were designing scorpion they figured out some way to overcome this 'flaw' but even if they didn't it will still be better than current A8 implementation.

[Arun]

Indeed, definitely looks like they want to amortize Scorpion in a lot of designs. They're also being a lot more aggressive with video & 3D hardware on 45nm. All good news for the industry...

But well, you know, I'm a tad subjective when it comes to NEON because I 300% agree with the point of view of a certain third-party I won't name that think it's as much of a joke as I think I am, and will stick to VFP. These companies have three zillion redundant units capable of doing the same thing, then they act all surprised when their die size is higher than the competition. Oh gosh, hoocoodanode?

Just look at Snapdragon: you could do video decoding on NEON. Or you could do it on the advanced DSP. Or you could do it in the dedicated accelerators! Errr, what exactly is the point, guys? And how many application developers do you really think are going to use SIMD intrinsics on an open platform where 90%+ of the user base doesn't have it available? The same thing is true to differing extents for other architectures, including OMAP4. It's as if all these guys were living in a world of their own where there's no competition and higher die size is merely a benefit because it allows you to justify a higher selling price... Oh well, I really should stop being so bitter!