Qualcomm Roadmap (2011-2012)

These days, phones aren't that easy to pocket either.

If you know you're going to someplace interesting and you'll be taking pictures, you can lug the gear. Phones are great for impromptu or unexpected photo opportunities.

Then there are people who go to these scenic places, at some expense, and take pictures with their phones. Maybe they can't afford or are uninterested in better cameras. Or have unrealistic expectations.

 

I go to exotic locations for the experience of the moment.

I once bought a DSLR but discovered that it didn't improve my picture taking skills (surprise) and that, unprocessed, they came out with less pop than a small Casio. The latter obviously oversaturated etc., but guess which pictures got the most wows?

Then when you do major travels to far away exotic countries with a backpack and that DSLR becomes a serious weight burden. And at the end of the day, the only pictures that you really keep coming back to are not the ones with a pretty sunset but those with other people.

DSLR went to eBay, pocket camera is gathering dust. iPhone camera is totally usable for 95% of my needs. The remaining 5% will remain underserved. I'll live.

 

Perfectly valid choice not to prioritize the quality of the pictures and be content with "good enough" phone cameras.

It's the ones who expect that phone cameras can be so good that they build a product differentiation or selection strategy around it, to make it one of the top features.

Company like Nokia is more reliant on the camera for differentiation but may be raising some unrealistic expectations.

 

Wrt to new features: I use the HDR function on the iPhone quite a bit. The 1-shot only HDR more on Tegra 4 is pretty interesting. Only a matter of time before everybody else has it. (Please, Apple, please!)

 

I look at the pictures I took with my N8 and usually see my expectations being surpassed.

 

Not sure if this is the right place to post it and it is already being discussed in other parts of this forum, but I thought I'd mention it here anyway.

http://www.anandtech.com/show/6804/opencl-drivers-discovered-on-nexus-4-and-nexus-10-devices

Pretty interesting to see a first demonstration of OpenCL compatibility with a shipping Snapdragon device (and Exynos, but this is a Qualcomm thread). It's not 'official', so app developers shouldn't start depending on it, but it sure is interesting to see progress being made on this front.

 

Interesting find..what kind of things could be used for on a mobile device? Stitching panoramic photos together? Video editing? Upscaling? In game effects? Er...

 

Anything that would make sense for GPGPU; yes stuff like image processing/adjusting and for games physics for stuff like cloth simulation etc.

 

Seeing open cl be switched on for developers to use throughout all mobile OS will be a nice jump IMO.

The question I have is...is there enough processing power in mobile gpus to make open cl usable?....is sgx 543 mp2 capable of gpgpu compute?....if so perhaps when apple turns this feature on in ios and mayne when games a built for a minimum of sgx 543 mp2/adreno 320/mali t604 performance target...we may well see a nice bump up in gaming capability.

 

Here is a nice demo is you're interested: http://www.engadget.com/2013/01/04/opencl-mod-for-the-kindle-fire-hd/
It was done on a TI OMAP4470 with an IMG SGX544, but they say they can do the same on an OMAP4460 with an IMG SGX540.

 

One word wow.

It didnt realise gpgpu compute could make such a difference on a low powered device...that demo was awesome...

It just proves what I was saying recently about current mobile aps and games are no where near taking full advantage of the hardware....I feel future os api updates will unleash unified shader mobile gpus.

 

See also: http://www.imgtec.com/demo_room/viewdemo.asp?DemoID=83&DemoTech=PowerVR%20Graphics&DemoDev=Imagination&#ViewPort

For stuff like image processing and some in game physics for example on a mobile device definitely. The more recent a mobile GPU is the higher the OpenCL API version they support. Here's a demo IMG an on the SGX540 last year for simple image adjustments. The SGX543 (as well as 540, 531, 535, 530) should be OCL1.0, SGX544/554 OCL1.1. Mali T604 and 624 I quickly checked on ARM's page are also 1.1. Adreno320 is according to Anand OCL1.2.

GPGPU = general purpose GPU; while some things can be used for in the future in mobile games heterogeneous computing is a different animal. That one might help also: http://withimagination.imgtec.com/i...aginations-gpu-compute-special-edition-part-1

 

Game physics has been the projected use case for GPGPU in gaming for a long time and still isn't standard even for high-end PC games. In many cases, there just isn't enough spare GPU power to do it since the GPU is already taxed doing graphics work. There's probably also inefficiencies having the GPU switch back and forth between graphics and compute and transferring compute results between the GPU and CPU. In comparison multi-core CPUs are generally under-utilized and are free for physics work if more focus was put on multithreading and taking full advantage of SIMD instruction sets as PhysX was criticized for a while ago. Where compute is currently used in games seems to be to make certain graphics algorithms more efficient rather than taking on additional work. I think this will be even more the case on mobile. I don't see the GPU even in top of the line devices being underutilized in Infinity Blade 2, Modern Combat 4, Real Racing 3 or other high-end games. In comparison, there's probably a core or two available in quad core CPU SoCs as developers improve multithreading.

I suppose GPGPU physics could be useful for graphically simply physics games, which admittedly are common on mobile. The advantage there is probably power efficiency more than enhanced performance.

http://forum.beyond3d.com/showpost.php?p=1680788&postcount=139

I asked about the OCL support for various PowerVR cores last time their marketing department put out graphics indicating that only SGX544 and up are OCL1.1 capable. Rys responded that all USSE cores are OCL1.1 Embedded capable. I wish they would revise those graphics.

I wonder if OCL1.2 Embedded support is also possible at least on USSE2?

 

Thanks
..yea we await these advanced apis to be included. ..can uou imagine what an A6x or s600 would be capable of if all features were taken advantage of??

Edit some good news regarding the zte grand memo..there will be 2-3 versions. ..chinese version carrying s4 pro...international versions carrying snapdragon s800@1.7ghz...and a 1080p screen...2gb ram (hopefully lpddr3) 13mp sensor (likely sony exmor) 3200 mah battery, dolby surround sound,andoid 4.1.2 and some 3d skinning (yuk!)..

If that comes at s reasonable price I might go for it.

 

It always comes down to what developers want to do; it's not that for cloth simulation you'd need at any price GPGPU. One of the past IMG demos dealt with cloth simulation through VS3.0 vertex texturing if memory serves well; as long as the effects are kept on a reasonable level ie small parts of the screen it's not going to be a problem either way.

Now for desktop when you have a game like Batman and in order to showcase physics that even a blind observer would notice Batman's cape will gracefully swing in the wind and when he runs through a big hall there are going to swing a gazillion of flags over him. It's always like that when new features get introduced; many developers get carried aways and just overdo it with new effects. If you use various stuff in reasonable portions and where it actually is a necessity and makes sense it's typically never a problem.

It always was and always will be a smart combination of effects combined with artistic talent that'll give you a great looking game.

http://withimagination.imgtec.com/index.php/powervr/powervr-g6100-small-is-powerful

2nd slide is where my assumption came from; hadn't seen Rys' comment though.

 

http://withimagination.imgtec.com/index.php/powervr/powervr-g6630-go-fast-or-go-home

When I originally asked about it a few months ago, I was looking at Fig 1 from the G6630 launch blog. I'm betting for the G6100 launch the PR people are continuing to expand incorrect slides rather than Rys being wrong originally.

 

I'd put my money on Rys compared to marketing But it's eventually picking at straws really. IMG themselves from public comments don't seem to expect GPGPU taking off before 2014; and I don't think it's because vendors don't want that early to give too much control to the GPU. It's my understanding (and according to the whispers that arrive from those 3rd parties that have touched relevant hw) that Rogue will be far more efficient with GPGPU and no its no surprise with a completely new architecture.

By the way I don't even have a clue what the difference in requirements between OCL1.0, 1.1 or 1.2 are

 

Nothing except atomics. The changes are cosmetic compared to hw changes we have seen since it first came out. Every API is years behind hw, except CUDA, which is not surprising.

 

Besides atomics, OCL1.1 raised the local memory minimum requirement from 16KB to 32KB, which is what kept the nVidia 8000/9000/200 series from fully supporting OpenCL 1.1. That even the original USSE supports OCL1.1, I'm surprised Imagination in the mobile space ended up being more forward thinking than nVidia in the desktop space for GPU compute.

 

And on mobile space they are well behind...however they have the potential to go straight to number one as soon as the first cuda part comes to tegra.