ARM’s New GPUs to Step up Mobile War With NVIDIA, Imagination Tech.

ARM Holdings, Plc. (LON:ARM) is making waves in the crowded mobile graphics market, airing a new, more powerful next-generation design [1][2].  

The chipmaker, best known for its licensed reduced instruction set computer (RISC) processors, is also pushing licensable mobile graphics processing unit (GPUs) technology, including the new design — dubbed the Mali-T658.  The Mali-T658 smartphone-aimed multi-core GPU promises smoother HD video playback and improved polygon pushing in games.

I. Mali’s Competitors are Mali’s Partners

ARM Holdings (or just ARM, for short) makes a GPU solution that competes with Qualcomm, Inc.’s (QCOM) Adreno GPU, NVIDIA Corp.’s (NVDA) mobile GeForce derivatives, and Imagination Technologies Group Plc.’s (LON:IMG) PowerVR designs.  

Qualcomm and NVIDIA also happen to be the two of ARM Holdings biggest clients, licensing its CPU design, used in their respective Snapdragon and Tegra lines.  

In other words, ARM is competing against its own biggest clients.  The awkward relationship brings to mind Google Inc.’s (GOOG) dual-role as a smartphone OEM competitor via recently acquired Motorola Mobility and its dual-role as a smartphone OS maker, gifting its Android OS on its OEM-level competitors like South Korea’s Samsung Electronics Comp., Ltd. (KS:005930).  Coincidentally Samsung licenses ARM’s CPU and GPU designs, and other Android handset makers like Google almost exclusively use ARM-licensed CPUs.

The GPU wars have moved from the PC onto smartphones and tablets! [Source: Flickr]
The odd web of ARM CPU-makers, GPU-makers, and Android OEMs may be messy, but thus far it’s spread a whole lot of revenue and prosperity around.  In other words, cooperation has been good, and the competition hasn’t been bad either.

II. Eight Cores of Power

Setting aside the sticky web of licensing and GPU entrants, the new T658 GPU represents a major technical stem forward for mobile GPUs.

Like most GPU architectures, ARM’s GPU design consists of shader cores each with arithmetic pipelines capable of doing parallel math operations.  Modern graphics is largely about intense floating point (e.g. scientific notation decimal numbers) math, except when it comes to colors.

Both the T604 and T658 are Mali Midgard GPUs.  The Midgard GPU uses an arithmetic, load/store and texturing “tri-pipe” design.  Figures indicate a double ALU per core design is used.  In other words, each core can do three or four simultaneous kinds of operations on two ALUs.

A VERY high level view of the Mali Midgard’s operation. [Source: ARM Holdings]
It’s important to note that this is somewhat different from NVIDIA’s Geforce-derived design in Tegra 3, which employs shaders with a single ALU per core.  As a second point of comparison, PowerVR offers a very similar dual ALU layout to what ARM Holdings is doing here.

ARM has enhanced its GPU in two ways from the previous Mali-T604 design. 

First, it’s doubled the arithmetic pipelines in each shader core.  Second it’s doubled the number of (fragment) shader cores from four to eight.  The new octacore GPU is on paper four times faster, as the result of these two major changes.  These changes allow it to essentially match rival Imagination Technologies’ Power VR SGX544 design (which also has 8 cores x 2 ALUs per core).
A (slightly) lower level view of the Mali T658 chip. [Source: ARM Holdings]
The Midgard family looks promising, not only for smartphone chip designs (the primary target), but tablets, netbooks, and laptops, as well.  Specifically it packs support for a host of GPU computing and graphics APIs, making it among the most multifunctional chips on paper.  

Comments ARM Media manager Steve Steele, “I should mention that Midgard supports all your favourite graphics APIs (Khronos™ OpenGL® ES, OpenVG™, Microsoft® DirectX® 11) and Compute APIs (Khronos OpenCL™, Google® RenderScript™ compute and Microsoft DirectCompute®). And not just any OpenCL I might add – Full Profile OpenCL is supported thanks to the IEEE 754-2008 floating point capability built into the GPU hardware.”

By contrast the SGX554 is a bit more limited in its standards support, though this doesn’t exactly affect most users on the smartphones end — as most smartphone apps (e.g. browsers and image editors) are still in the early stages of GPU computing optimization.

Both Tegra 3 and the Mali T658 are capable of supporting up to 1080p.  Tegra 3 promises 1280×1024 pixel resolution — OS allowing.  By contrast, details on Mali’s LCD driving capabilities weren’t revealed.

It’s rumored that one advantage that Mali may have over rival designs like Geforce and PowerVR is cache coherence — i.e. in layperson’s terms, it can share its fast local memory between the GPU and CPU.

In the limited available information some important details are unknown, namely:

  1. Did the vertex shader count change?
  2. Is the ALU count per fragment shader still two (as early diagrams suggested)?
  3. How wide is the ALU/are the ALUs? (See AnandTech‘s review for a guess at the width from the last gen. design.)
  4. What does the pipeline look like/how deep is it?
  5. What’s the clock speed?
  6. What resolutions are supported by the built-in LCD driver?

We’re reaching out to ARM for answers.  But don’t necessarily expect a response — the company is notoriously cagy on details about its implementation.

III. What Phones/Tablets Will This Land in, And How Good Will They Be?

The GPU is “out now” from ARM, but that means someone still has to go and put the little bugger on a finished chip design.

Samsung’s processor line recently made an important switch from Imagination Technologies GPUs to ARM Holding GPUs.  The older Hummingbird chips (still in use) pack PowerVR GPUs, but the new Exynos chips pack Mali GPUs.  This a huge win for ARM Holdings as it means that its GPUs are in the latest and greatest products from Samsung, including the best-selling Android smartphone, the Galaxy S II.  

The Galaxy S II used the Mali T400 design — the last major design before Midgard.  This two-generations-old design was capable of the best overall graphics performance of any smartphone in OpenGL benchmarks (though trailing higher-clocked Imagination Technologies’ SGX540 in pure triangle output) in September, but has since been dethroned by the SGX543 MP2 (onboard the iPhone 4S) in benchmarks.

It’s hard to say when the Mali-T658 products will land, but Q3 2012 seems a fair guess.  That’s just after Imagination Technologies is deploying its octacore-ready SGX554 architecture (Q2 2012).

The latest Samsung roadmap shows the Exynos 5250 as the first chip to sport the Mali T604.  That’s due out in 2012.  Samsung commonly releases two system-on-a-chip designs per year, so a second half 2012 launch sounds about right.  

With the next-generation not-yet-named Exynos Mali T658-equipped chip and the Exynos 5250, ARM Holdings is poised to likely spring into second or third, behind Imagination Technologies (and potentially NVIDIA) in the smartphone/tablet market.  PowerVR remains entrenched on devices from Apple, Inc. (AAPL) and Motorola Mobility.   

The major unknown is Tegra.  By 2012, Tegra 4 (codename Wayne) chips will be dropping with a host of improvements, including a GPU design which can have up to 64 cores (of course, comparing this to ARM Holdings and Imagination Technologies’ octa-core GPUs is a bit questionable as the architectures are apparently very different).  If Tegra 4 can come out ahead of the SGX554 at the Mali-T658 in performance, it just might continue its climb in market share.  Tegra is increasingly being picked up by Motorola, LG Electronics, Inc. (KS:066570), and traditional PC makers turned tablet market entrants like Acer Inc. (TPE:2353).

A final player worth mentioning is Qualcomm, whose Adreno design comes in the Snapdragon processors heavily used by HTC Corp. (TPE:2498).  Qualcomm will surely be looking to bring its A game too, so it’s too soon to count it out.

A final word:

Who offers the best power efficiency, the best performance, and the best combination of the two is clearly only part of the picture.  A major aspect of the success or failure of these various entrants — including the Mali-T658 — boils down to what kind of deals with OEMs are reached, and how those OEMs fare in the market.