Winjer
10-11-14, 16:03
PowerVR Series7 GPU unveiled: Bringing desktop-class graphics to the mobile masses (http://www.extremetech.com/computing/193868-powervr-series7-gpu-unveiled-bringing-desktop-class-graphics-to-the-mobile-masses)
Imagination Technologies has detailed its next-gen PowerVR Series7 GPUs — and boy are they beastly. Imagination is claiming a performance boost of 60% over its Series6 GPUs, clock for clock, which for a single 12-month generational leap is pretty darn impressive. The top-end Series7XT GT7900 will feature 16 clusters (somewhat equivalent to a GPU “core”), for a total of 512 ALUs and peak theoretical performance above 1 teraflop — or around the same as an Nvidia GTX 750. There’s also full support for the Android Extension Pack, a new feature in Android 5.0 Lollipop (for supported GPUs) that brings hardware tessellation, ASTC texture compression, and other neat “desktop-class” graphics features.
Imagination Technologies, much like Qualcomm, ARM Holdings, and other players in the mobile space, have had an utterly crazy few years. For the longest time, these companies were the designers of obscure, low-power chips for embedded solutions and other non-sexy products — and then, with the popularization of the smartphone in 2007, everything changed very quickly. Now, almost every cellular device in the world has a Qualcomm chip inside it — and almost every mobile device uses a CPU licensed from ARM Holdings. Imagination Technologies, a fabless chip designer based in England, is mostly known for its PowerVR brand of GPUs, which have powered Apple’s SoCs since the A4 (the iPhone 4/original iPad). As of the A8X SoC (http://www.extremetech.com/computing/192628-the-ipad-air-2-with-a-tri-core-cpu-is-almost-as-fast-as-a-modern-pc) (iPad Air 2), the six-cluster PowerVR GX6650 is just about neck-and-neck with the Kepler-based GPU inside Nvidia’s Tegra K1 SoC.
The Series7 GPU is essentially an upgraded and reworked version of the Series6 Rogue GPU. There’s hardware support for new features — tessellation, virtualization, optional DirectX/Direct3D 11 support, an optional FP64 ALU for supercomputing applications — and lots of performance and efficiency tweaks over Series6. A rejigged cluster layout and cache design improvements apparently result in a 3x boost in parallel processing performance. Overall, Imagination is reporting a 60% speed-up over Series6 GPUs, clock-for-clock, cluster-for-cluster in “industry standard benchmarks.”
Read: PowerVR Wizard GPU combines 3D and ray tracing hardware for some amazing results (http://www.extremetech.com/computing/178743-powervr-series6-wizard-gpu-is-the-first-to-combine-3d-and-ray-tracing-hardware-in-a-single-chip)
Looking at the block diagrams, you can see that the Series7XT and Series6XT — except for low-level tweaks — are basically identical.
http://www.extremetech.com/wp-content/uploads/2014/11/series7xt-architecture-640x400.gif (http://www.extremetech.com/wp-content/uploads/2014/11/series7xt-architecture.gif)
PowerVR Series7XT block diagram
http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series6XT-GPU-640x365.jpg (http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series6XT-GPU.jpg)
PowerVR Series6XT block diagram – very similar to the Series7
The layout of each Unified Shading Cluster (USC) is also identical, with the exception of an optional FP64 ALU — the main difference being that Series7 now allows for up to 16 USCs, where Series6XT maxed out at six.
http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series7-Series7XT_USC-640x440.png (http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series7-Series7XT_USC.png)
PowerVR Series7XT GPU cluster layout
I’m not entirely sure how Imagination Technologies squeezed another 60% out of much the same USC design; I guess we’ll have to wait until we actually get our hands on some hardware to see how the performance claims stand up. (At this point, all of Imagination’s claims are based on software-based emulation of the design — which is how every chip designer tests and verifies a new design before going through the very expensive and time-consuming tape-out process.)
Overall, it sounds like the Series7XT is mostly just a refined version of the Series6XT, with the addition of some juicy new hardware features (tessellation, virtualization, FP64 forsupercomputing (http://www.extremetech.com/tag/supercomputing)). This makes sense, given how it’s only been less than a year since Series6XT was announced, and the mobile market — like the PC market — is now very much about evolutionary updates, rather than revolutionary changes that generally upset developers and break compatibility. The Series6XT GPU inside the iPad Air 2 performs surprisingly well, keeping up with Nvidia’s much-lauded Tegra K1 (http://www.extremetech%2Ccom/tag/tegra-k1) — assuming the Series7XT delivers as promised, it should put PowerVR licensees like Apple in very good stead.
http://www.extremetech.com/wp-content/uploads/2014/11/powervr-series7xt-animated-gif.gif (http://www.extremetech.com/wp-content/uploads/2014/11/powervr-series7xt-animated-gif.gif)
PowerVR Series7XT sample scenes. (Well, possibly rendered on a Series7XT – it’s hard to say, and Imagination Technologies isn’t giving much in the way of details right now.)
We should see the first products with a Series7XT GPU (http://blog.imgtec.com/powervr/powervr-series7xt-gpus-push-graphics-and-compute-performance) in the next 12-18 months, which should line up nicely with Nvidia’s next-gen Tegra with a Maxwell-based GPU. Presumably the Apple A9 or A10 SoC will feature a Series7XT GPU. There’s no word on when we might see Qualcomm’s 500-series Adreno GPU, or perhaps a beefier 400-series GPU, but presumably it’ll debut alongside a 64-bit Snapdragon SoC at some point in 2015.
Imagination Technologies also announced the Series7XE (http://blog.imgtec.com/powervr/new-powervr-series7xe-family-targets-the-next-billion-mobile-and-embedded-gpus), a smaller, cheaper GPU for mid-range devices that still has hardware tessellation, Android Extension Pack support, and all the other goodies — it just has a much smaller cluster count (as low as half a USC/16 ALUs, if you so wish).
Concorrência para a nVidia e os Tegra?
Imagination Technologies has detailed its next-gen PowerVR Series7 GPUs — and boy are they beastly. Imagination is claiming a performance boost of 60% over its Series6 GPUs, clock for clock, which for a single 12-month generational leap is pretty darn impressive. The top-end Series7XT GT7900 will feature 16 clusters (somewhat equivalent to a GPU “core”), for a total of 512 ALUs and peak theoretical performance above 1 teraflop — or around the same as an Nvidia GTX 750. There’s also full support for the Android Extension Pack, a new feature in Android 5.0 Lollipop (for supported GPUs) that brings hardware tessellation, ASTC texture compression, and other neat “desktop-class” graphics features.
Imagination Technologies, much like Qualcomm, ARM Holdings, and other players in the mobile space, have had an utterly crazy few years. For the longest time, these companies were the designers of obscure, low-power chips for embedded solutions and other non-sexy products — and then, with the popularization of the smartphone in 2007, everything changed very quickly. Now, almost every cellular device in the world has a Qualcomm chip inside it — and almost every mobile device uses a CPU licensed from ARM Holdings. Imagination Technologies, a fabless chip designer based in England, is mostly known for its PowerVR brand of GPUs, which have powered Apple’s SoCs since the A4 (the iPhone 4/original iPad). As of the A8X SoC (http://www.extremetech.com/computing/192628-the-ipad-air-2-with-a-tri-core-cpu-is-almost-as-fast-as-a-modern-pc) (iPad Air 2), the six-cluster PowerVR GX6650 is just about neck-and-neck with the Kepler-based GPU inside Nvidia’s Tegra K1 SoC.
The Series7 GPU is essentially an upgraded and reworked version of the Series6 Rogue GPU. There’s hardware support for new features — tessellation, virtualization, optional DirectX/Direct3D 11 support, an optional FP64 ALU for supercomputing applications — and lots of performance and efficiency tweaks over Series6. A rejigged cluster layout and cache design improvements apparently result in a 3x boost in parallel processing performance. Overall, Imagination is reporting a 60% speed-up over Series6 GPUs, clock-for-clock, cluster-for-cluster in “industry standard benchmarks.”
Read: PowerVR Wizard GPU combines 3D and ray tracing hardware for some amazing results (http://www.extremetech.com/computing/178743-powervr-series6-wizard-gpu-is-the-first-to-combine-3d-and-ray-tracing-hardware-in-a-single-chip)
Looking at the block diagrams, you can see that the Series7XT and Series6XT — except for low-level tweaks — are basically identical.
http://www.extremetech.com/wp-content/uploads/2014/11/series7xt-architecture-640x400.gif (http://www.extremetech.com/wp-content/uploads/2014/11/series7xt-architecture.gif)
PowerVR Series7XT block diagram
http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series6XT-GPU-640x365.jpg (http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series6XT-GPU.jpg)
PowerVR Series6XT block diagram – very similar to the Series7
The layout of each Unified Shading Cluster (USC) is also identical, with the exception of an optional FP64 ALU — the main difference being that Series7 now allows for up to 16 USCs, where Series6XT maxed out at six.
http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series7-Series7XT_USC-640x440.png (http://www.extremetech.com/wp-content/uploads/2014/11/PowerVR-Series7-Series7XT_USC.png)
PowerVR Series7XT GPU cluster layout
I’m not entirely sure how Imagination Technologies squeezed another 60% out of much the same USC design; I guess we’ll have to wait until we actually get our hands on some hardware to see how the performance claims stand up. (At this point, all of Imagination’s claims are based on software-based emulation of the design — which is how every chip designer tests and verifies a new design before going through the very expensive and time-consuming tape-out process.)
Overall, it sounds like the Series7XT is mostly just a refined version of the Series6XT, with the addition of some juicy new hardware features (tessellation, virtualization, FP64 forsupercomputing (http://www.extremetech.com/tag/supercomputing)). This makes sense, given how it’s only been less than a year since Series6XT was announced, and the mobile market — like the PC market — is now very much about evolutionary updates, rather than revolutionary changes that generally upset developers and break compatibility. The Series6XT GPU inside the iPad Air 2 performs surprisingly well, keeping up with Nvidia’s much-lauded Tegra K1 (http://www.extremetech%2Ccom/tag/tegra-k1) — assuming the Series7XT delivers as promised, it should put PowerVR licensees like Apple in very good stead.
http://www.extremetech.com/wp-content/uploads/2014/11/powervr-series7xt-animated-gif.gif (http://www.extremetech.com/wp-content/uploads/2014/11/powervr-series7xt-animated-gif.gif)
PowerVR Series7XT sample scenes. (Well, possibly rendered on a Series7XT – it’s hard to say, and Imagination Technologies isn’t giving much in the way of details right now.)
We should see the first products with a Series7XT GPU (http://blog.imgtec.com/powervr/powervr-series7xt-gpus-push-graphics-and-compute-performance) in the next 12-18 months, which should line up nicely with Nvidia’s next-gen Tegra with a Maxwell-based GPU. Presumably the Apple A9 or A10 SoC will feature a Series7XT GPU. There’s no word on when we might see Qualcomm’s 500-series Adreno GPU, or perhaps a beefier 400-series GPU, but presumably it’ll debut alongside a 64-bit Snapdragon SoC at some point in 2015.
Imagination Technologies also announced the Series7XE (http://blog.imgtec.com/powervr/new-powervr-series7xe-family-targets-the-next-billion-mobile-and-embedded-gpus), a smaller, cheaper GPU for mid-range devices that still has hardware tessellation, Android Extension Pack support, and all the other goodies — it just has a much smaller cluster count (as low as half a USC/16 ALUs, if you so wish).
Concorrência para a nVidia e os Tegra?