Imagination Technologies - Faster GPUs for SoCs

I met with Imagination Technologies, the makers of the GPU in everything from the iPhone 3GS to the Motorola Droid. They were showing off a 1.1GHz Samsung Cortex A8 SoC with an integrated PowerVR SGX 540:

The demo isn't anything impressive, but the GPU should be in the Tegra 2 class of performance. Definitely higher than what's in the iPhone 3GS and Palm Pre today. The PowerVR SGX 540 is what is being used in TI's Cortex A9 SoCs, which appear to be the direct competitor to NVIDIA's Tegra 2. I spoke with Samsung about their 1.1GHz A8 SoC (now running at 1.2GHz actually), but I wasn't able to get a firm date on when we can expect A9 based silicon from them.

Our friends at Imagination put together a tech demo of a 3D user interface for TVs running on Intel's CE4100 SoC. The CE4100 was announced back at IDF and integrates an Atom core with PowerVR SGX graphics. Imagation believes that TVs or set-top boxes with chips like the CE4100 will use the GPU in a similar capacity to smartphones: for fancy UIs.

MIPS, another SoC vendor that does a large amount of business in the set-top box market showed off a number of Android devices for the home. Many of these were set-top boxes running Android that could do everything from video playback to watching over-the-air TV:

Unfortunately most cable providers in the US aren't as receptive to devices that can enable content consumption from sources other than cable TV, and thus I wouldn't expect to see much support for technology like this in the US just yet. MIPS has found the best success in Asia for these types of STBs so far.

Index NVIDIA - Fermi Up and Running, Tegra 2
Comments Locked

40 Comments

View All Comments

  • Zool - Tuesday, January 12, 2010 - link

    I think that gf100 is different from the tesla die and that they have two gpu dies this time. The tesla die with maxed DP calculations and a similar GF100 die without the DP functionality. This site http://www.semiconductor.net/article/438968-Nvidia...">http://www.semiconductor.net/article/43...en_Calls... says 3.2 bilion transistors is the chip and that they have problems with the defect rates at TSMC with those sizes.

    Two 1.5 bilion transistor chips conected on one package would have much lover defect rates than gf100. Nvidia can dream about giant chips with zero defects but i think the future is in smaler dual, quad core gpus that can be efectivly manufactured. Similar road than the cpu-s had taken years ago with one difference. Gpu-s are super parallel so performance could be near linear with added cores all the time.
  • Calin - Monday, January 11, 2010 - link

    There are some issues here:
    1. NVidia wants Fermi to be faster overall, but it might be possible that it won't be faster than 5870 in DX11 for now. Drivers will help in the long way, and game engines optimised for Fermi will help also (maybe a lot).
    2. NVidia wants Fermi to succeed on the desktop, but their big hopes are for Fermi in the supercomputing arena. They might have chosen a poorer performing design for the desktop in order to crush everything in the race for teraflops.
  • Calin - Monday, January 11, 2010 - link

    There are some issues here:
    1. NVidia wants Fermi to be faster overall, but it might be possible that it won't be faster than 5870 in DX11 for now. Drivers will help in the long way, and game engines optimised for Fermi will help also (maybe a lot).
    2. NVidia wants Fermi to succeed on the desktop, but their big hopes are for Fermi in the supercomputing arena. They might have chosen a poorer performing design for the desktop in order to crush everything in the race for teraflops.
  • lopri - Saturday, January 9, 2010 - link

    That GMA is finally on par with equivalent offerings from AMD/NV (about which I still have my doubt) is the most exciting news of the CES 2010 for you?
  • Someguyperson - Monday, January 11, 2010 - link

    As a side note, the integrated graphics being able to handle AutoCAD isn't very impressive, as my High school computers with integrated GMA 950 graphics could handle AutoCAD and AutoCAD Inventor, however they would BSOD if you made ~10,000 lines and tried to zoom in and out really quickly. It was pretty fun.
  • Calin - Monday, January 11, 2010 - link

    Intel is fighting in the graphics market by price alone (pretty solid integrated hardware and drivers for it at very low quality and price). AMD and NVidia fight in this market by not-so-low prices but better quality.
    This will bring at least two advantages:
    1. Most of the computers will come with better integrated graphics (Intel owns more than half the integrated market graphics)
    2. NVidia and AMD are forced to either price-war with Intel (which they won't do in the integrated graphics department) or have better products.

    I for one am expecting the better products :)
  • - Sunday, January 10, 2010 - link

    he's been drinking to much of the Intel Kool aid - it's now a reflex
    asH
  • lopri - Saturday, January 9, 2010 - link

    It's a question, btw. :) I thought you had covered it prior to CES.
  • formulav8 - Sunday, January 10, 2010 - link

    Like some others, i'm not to impressed by Intels new graghics in itself.

    The graphics has direct access to the memory controller and STILL can't beat 2 year old solutions except in a few areas. To me that is pretty pathetic.

    Obviously though, others are quite impressed, like anand but I would expect much more from the gpu having ondie memory controller access, even Intels typical trash.



    Jason
  • JarredWalton - Sunday, January 10, 2010 - link

    I don't think anyone caught the two links Anand posted. The two Intel items that Anand was most impressed with are the wireless HD transmitter and the Moorestown smartphone demonstration. Arrandale and the new Intel IGP isn't one of the most impressive items from CES.

    Personally, I'm moderately impressed that Intel has finally put some real effort into improving their IGP. This is evidenced by the fact that a 20% increase in the number of shader cores/pipelines resulted in up to a 100% increase in performance.

    The GMA 900/950 series was horrible in comparison to the NVIDIA and ATI IGPs of the time (Radeon Xpress 1200/1250 and GeForce 6100). The next generation GMA 3000 stuff was still less than 1/4 the performance of the newer NVIDIA/ATI IGPs (Xpress 3200 and 7100 series). Then we had a lot of stagnation and the 8100 and 9300/9400 from NVIDIA and the HD 3200/4200 really didn't do much more compared to their last IGPs. The GMA 4500MHD has much of the HD decode support and with appropriate drivers is only about half the performance of ATI/NVIDIA IGPs. Now they have closed the gap, and I'll be surprised if ATI and NVIDIA do much more than about a 10-20% performance boost on IGPs (though naturally ATI will have a DX11 IGP).

    When you get right down to it, though, all IGPs still pretty much suck for 3D performance. They can handle minimum detail settings at 800x600 most of the time, and on less demanding titles you can even get 1366x768 with medium detail. Any $100 discrete GPU ends up being at least twice as fast, and with hybrid graphics really kicking off now that Win7 is out, if you want 3D performance there's really no reason not to find a laptop with a discrete GPU.

Log in

Don't have an account? Sign up now