PowerVR GX6650: redefining performance in mobile with 192 cores

It’s been a few weeks since we introduced the ground-breaking PowerVR Series6XT GPU family and already we’ve seen amazing interest in the new line-up of GPUs from our customers and ecosystem. On our blog, our most avid readers have been eager for us to reveal how competitive our newly-released GPUs are compared to other mobile graphics parts announced at CES 2014.

All Rogue GPUs contain anywhere from half a cluster to eight; this scalability enables our customers to target the requirements of a growing range of demanding markets from entry level mobile to the highest performance embedded graphics including premium smartphones, tablets, console, automotive and home entertainment.

This article is meant to provide an overview of PowerVR GX6650, the shining star of the PowerVR Series6XT family and the most powerful GPU IP core available today.

You want cores and GFLOPS? PowerVR GX6650 has plenty.

PowerVR GX6650 has six Unified Shading Clusters (USCs) and 192 cores, making it our fastest GPU to date. If you want to understand how we define a core and get an idea of how the USC works, please read this blog article from Rys.

PowerVR GPU_PowerVR GX6650

A block diagram of the PowerVR GX6650 GPU shows the six clusters and 192 cores

Additionally, PowerVR GX6650 gives developers the flexibility to choose the most appropriate precision path for their application, making it the undisputed leader for low-power performance. For example, GX6650 includes a low-power, high-performance FP16 mode which can provide up to twice the performance of FP32 within a constrained power budget. We’ve improved the performance of this FP16 mode because FP32 is not always the best choice for optimum performance with power efficiency.

PowerVR GX6650 offers the best pixel throughput available

Another major advantage of our PowerVR GX6650 GPU is its massive pixel throughput rate. It is able to process 12 pixels per clock, which is up to triple the number of our competitors.

Pixel throughput is a metric which tends to be overlooked in favor of GFLOPS but is equally important when thinking about high resolution displays. Since PowerVR GX6650 is aimed at processors targeting high-end, high-resolution tablets or 4K smart TVs, pixel throughput becomes a hot topic, especially when you have FPS-heavy gaming or complex user interfaces.

Add to that the fact that PowerVR GX6650 has an up to 50% higher texture rate compared to its direct competitors, and you quickly begin to understand that the front-end of PowerVR Series6XT GPUs is a beast.

PowerGearing G6XT and PVR3C keep power consumption in check

Now that we’ve answered your performance-related questions, let’s look more closely at probably the biggest factor which influences mobile graphics: power consumption.

When it comes to power efficiency, PowerVR is and has always been ahead of the pack. PowerVR GX6650 includes all the latest power-saving technologies from Imagination, including PVR3C and PowerGearing G6XT.

PowerGearing G6XT is a set of power management mechanisms that lead to sustained performance within fixed power and thermal envelopes. PowerGearing for PowerVR GX6650 enables us to control all GPU resources, allowing dynamic and demand-based scheduling of the six shading clusters and other processing blocks.

PowerVR GX6650

PowerGearing G6XT offers advanced power management for best in class efficiency

PVR3C is a suite of triple compression formats that includes PVRTC and ASTC for texture compression, PVRIC for frame buffer compression and PVRGC geometry compression. Whereas competing solutions have only started implementing these concepts, PowerVR GX6650 implements our second generation PVR3C solutions, with optimized algorithms for all of the use cases that matter in mobile (gaming, web browsing, UIs and many more).

PVRIC

PVRIC works particularly well on high-resolution home screens found in modern mobile operating systems, yet still finds a great compression rate in more traditional dynamic gaming situations

Beyond the core count lies the best feature set available for mobile

The mobile industry has little in common with the desktop market which explains why desktop-class, brute-force GPUs have always been, at best, a niche solution. The same consideration applies to proprietary APIs which can be popular as niche solutions on desktops but are not the way forward in mobile, which has thrived on standards like OpenGL ES and OpenCL EP.

There is a reason why Khronos has created desktop and embedded profiles for both graphics and compute; it allows GPU companies to create solutions that are optimized for those specific markets and their specific requirements. It is true that you can offer various extensions through which you can enable certain unique features – this is something we’ve done very successfully over the years; but, fundamentally, if you are targeting mobile devices, you want to make sure your GPU is optimized for the right use cases and fully supports mobile API standards.

SoftKitty OpenGL ES demo - PowerVR Rogue (1) SoftKitty OpenGL ES demo - PowerVR Rogue (2)

Our new Soft Kitty OpenGL ES 3.0 demo highlights the performance of high-end PowerVR Rogue GPUs

Success in mobile is not just a question of attaching the biggest number to one metric of a GPU: it requires a balanced approach which delivers a blend of performance, efficiency and features. PowerVR GX6650 offers uncompromised support for OpenGL ES 3.0, OpenCL 1.2 EP, Direct3D 11 Levels 9_3/10_0, OpenGL 3.x, OpenCL, and Renderscript, plus future-proofing for upcoming mobile APIs but without the burden of advanced desktop features which cannot be used because no mobile API exists to expose them to developers.

In conclusion, PowerVR GX6650 shows that our TBDR graphics architecture is still the leader and is set to remain in that position for the long term.

What do you think about this high-end GPU? Let us know in the comment box below. Make sure to follow us on Twitter (@ImaginationPR, @PowerVRInsider, @GPUCompute) for the latest news and updates from Imagination.

, , , , , , , , , ,

  • Luca

    Woow Alex this gpu design seems an absolutely beast!! I never thought this could be compared with the gpu design of the Tegra K1! Good stuff!!
    Luca

  • https://plus.google.com/u/0/115605168398121887465/about Anzhel Bozhinov

    Tegra K1 supports OpenGL 4.4, so GeForce 5 ULP>PowerVR GX6650

  • roninja

    Rogue supports anything between half and eight clusters. Will we ever see the latter in G68xx configuration?

  • http://withimagination.imgtec.com/index.php/author/alexvoica Alexandru Voica

    And why exactly would you assume that one GPU is superior to another only based on support for one API?

  • http://withimagination.imgtec.com/index.php/author/alexvoica Alexandru Voica

    Thanks, I’ve spent quite some time thinking about how I’d approach this article. I’m glad you’ve enjoyed it.

  • http://withimagination.imgtec.com/index.php/author/alexvoica Alexandru Voica

    Yes, you are right – the Rogue architecture can scale up to 8 clusters. The publicly-available roadmap includes two six cluster parts: PowerVR G6630 and GX6650.

  • mdriftmeyer

    In a world where OpenGL ES 3.0 is the target, you’re discussing OpenGL 4.4 that will never be utilized by the Tegra K1 in any OS.

    Desktop hardware accelerated Environments throughout have OS X at OpenGL 4.1 and OpenCL 1.2.

    Windows: DirectX

    Linux Games: Some utilize OpenGL 4.x partial.

    Linux DE: OpenGL 2.1 in KDE 4.12 and GNOME 3.8 max.

    OpenGL 3.x is 9-12 months away for Linux.

    In short, OpenGL 4.4 won’t arrive until say OS X 11.0, at the earliest.

    These PowerVR beasts within an Apple A-Series future SoC will scream efficiency, scalability and energy lean.

  • Pingback: Imagination Technologies anuncia su GPU PowerVR GX6650 “Rogue 2.0” | Blogdesktop

  • https://plus.google.com/u/0/115605168398121887465/about Anzhel Bozhinov

    you’re discussing OpenGL 4.4 that will never be utilized by the Tegra K1 in any OS

    So, we don’t need OpenGL 4.4, right? Khronos did something wrong. OK. [/irony]

    In a world where OpenGL ES 3.0 is the target

    Are you sure that hypothetical Android 4.5 will not support OpenGL 4.4 for smartbooks (like HP Slatebook x2), game consoles (like OUYA or SHIELD) and similar stuff?

    Linux Games: Some utilize OpenGL 4.x partial.

    This! Hardware must give choice to developer – use OpenGL 4.4 or maybe only OpenGL 1.0.

  • Pingback: Imagination Technologies anuncia su GPU PowerVR GX6650 “Rogue 2.0” | SINETEC

  • darkich

    This is an unprecedented level of transparency and openness for such a company.
    Amazing

  • Pingback: Imagination Technologies presenta Rogue 6XT, posible GPU del iPhone 6

  • L

    Ouya or Shield do support OpenGL 4.4? Nice to know that. I am not sure Nvidia knows that, maybe you may want to tell that to Nvidia.
    But in meantime the question would be who support Ouya or Shield, tell me do these device still exist?

  • Pingback: Chip wars heat up at Mobile World Congress – ZDNet | Top Computers Outlet

  • Pingback: Chip wars heat up at Mobile World Congress | FIND THE BEST DEALS FAST

  • Pingback: Chip wars heat up at Mobile World Congress – ZDNet | You Buy Computers

  • Pingback: Chip wars heat up at Mobile World Congress : World Latest News

  • https://plus.google.com/u/0/115605168398121887465/about Anzhel Bozhinov

    Keyword is like
    Nvidia will use Tegra K1 in devices like SHIELD or OUYA.

  • Pingback: PowerVR GR6500: новый уровень графики в мобильных устройствах | Gadgets News

  • http://withimagination.imgtec.com/index.php/author/alexvoica Alexandru Voica

    Thanks, we’ve put a lot of effort into making this more information more accessible to people.