New ISA & ALUs: An Extremely Wide Architecture

As mentioned, the ALU architecture as well as ISA of the new A-Series is fundamentally different to past Imagination GPUs, and in fact is very different from any other publicly disclosed design.

The key characteristic of the new ALU design is the fact that it’s now significantly wider than what was employed on the Rogue and Furian architectures, going to up a width of 128 execution units per cluster.

For context, the Rogue architecture used 32 thread wide wavefronts, but a single SIMD was only 16 slots wide. As a result, Rogue required two cycles to completely execute a 32-wide wavefront. This was physically widened to 32-wide SIMDs in the 8XT Furian series, executing a 32-wide wavefront in a single cycle, and was again increased to 40-wide SIMDs in the 9XTP series.

In terms of competing architectures, NVIDIA’s desktop GPUs have been 32-wide for several generations now, while AMD more recently moved from a 4x16 ALU configuration with a 64-wide wavefront to native 32-wide SIMDs and waves (with the backwards compatibility option to cluster together two ALU clusters/CUs for a 64-wide wavefront).

More relevant to Imagination’s mobile market, Arm’s recent GPU releases also have increased the width of their SIMDs, with the data paths increasing from 4 units in the G72, to 2x4 units in the G76 (8-wide wave / warp), to finally a bigger more contemporary 16-wide design with matching wavefront in the upcoming Mali-G77.

So immediately one might see Imagination’s new A-Series GPU significantly standing out from the crowd in terms of its core ALU architecture, having the widest SIMD design that we know of.

All of that said, we're a bit surprised to see Imagination use such a wide design. The problem with very wide SIMD designs is that you have to bundle together a very large number of threads in order to keep all of the hardware's execution units busy. To solve this conundrum, a key design change of the A-Series is the vast simplification of the ISA and the ALUs themselves.

Compared to the Rogue architecture as depicted in the slides, the new A-Series simplifies a execution unit from two Multiply-Add (MADD) units to only a single MADD unit. This change was actually effected in the Series-8 and Series-9 Furian architectures, however those designs still kept a secondary MUL unit alongside the MADD, which the A-Series now also does without.

The slide’s depiction of three arrows going into the MADD unit represents the three register sources for an operation, two for the multiply, and one for the addition. This is a change and an additional multiply register source compared to the Furian architecture’s MADD unit ISA.

In essence, Imagination has doubled-down on the transition from an Instruction Level Parallelism (ILP) oriented design to maximizing Thread Level Parallelism(TLP). In this respect it's quite similar to what AMD did with their GCN architecture early this decade, where they went from an ILP-heavy design to an architecture almost entirely bound by TLP.

The shift to “massive” TLP along with the much higher ALU utilization due to the simplified instructions is said to have enormously improved the density of the individual ALUs, with “massive” increases in performance/mm². Naturally, reduced area as well as elimination of redundant transistors also brings with itself an increase in power efficiency.

The next graphic describes the data and execution flow in the shader processor.

Things start off with a data master which kicks off work based on command queues in the memory. The 3D data master here also handles other fixed-function pre-processing, which will trigger execution of per-tile hidden surface removal and workload generation for the shader programs. The GPU here has a notion of triangle merging which groups them together into tasks in order to get better utilization of the ALUs and able to fill the 128 slots of the wavefront.

The PDS (Programmable Data Sequencer) is an allocator for resources and manager. It reserves register space for workloads and manages tasks as they’re being allocated to thread slots. The PDS is able to prefetch/preload data to local memory for upcoming threads, upon availability of the data of a thread, this becomes an active slot and is dispatched and decoded to the execution units by the instruction scheduler and decoder.

Besides the primary ALU pipeline we described earlier, there’s a secondary ALU as well. First off, a clarification on the primary ALUs is that we also find a separate execution unit for integer and bitwise operations. These units, while separate in their execution, do share the same data paths with the floating-point units, so it’s only ever possible to use one or the other. These integer units are what enable the A-Series to have high AI compute capabilities, having quad-rate INT8 throughput. In a sense, this is very similar to Arm’s NN abilities on the G76 and G77 for integer dot-product instructions, although Imagination doesn’t go into much detail on what exactly is possible.

The secondary pipeline runs at quarter rate speed, thus executing 32 threads per cycle in parallel. Here we find the more complex instructions which are more optimally executed on dedicated units, such as transcendentals, varying operations and iterators, data conversions, data moving ops as well as atomic operations.

A New Architecture To Bring IMG's Return? Fixed Function Changes & Scalability
POST A COMMENT

143 Comments

View All Comments

  • ET - Tuesday, December 3, 2019 - link

    > I fear that this will be a very niche product unless it absolutely dominates all other solutions.

    At least from the description in the article, it seems to dominate Mali. Even next gen Mali. I don't expect Apple or Qualcomm to move. Samsung I think would be flexible. It's impossible to say how well RDNA fits all price points or when it will arrive, so ImgTech could find a place there. And with the A series supposedly much better than Mali in performance per silicon, I don't think that HiSilicon using it is totally out of the question.
    Reply
  • Spunjji - Tuesday, December 3, 2019 - link

    No reason HiSilicon can't change their minds if there's a compelling reason. PPA advantages directly translate into cost savings, which is very compelling indeed.

    MediaTek are probably going to be the biggest customer, though.
    Reply
  • mode_13h - Wednesday, December 4, 2019 - link

    Chinese want nothing to do with ARM, any more. Got it?

    So, anyone who's using Mali, or any Chinese phone makers who are using Qualcomm are potential customers.
    Reply
  • vladx - Wednesday, December 4, 2019 - link

    If you bothered to read the article, you would've found the answer but I guess Americans can't be bothered to read. Reply
  • s.yu - Wednesday, December 4, 2019 - link

    "Americans can't be bothered to read"
    Wow, calling others haters, and look at you.
    Reply
  • Etain05 - Tuesday, December 3, 2019 - link

    I know that they don’t actually compete, since Apple will never offer its design for licensing, but I still think it’s interesting to compare them.

    Let’s take the numbers from the Huawei Mate 30 Pro review and compare making some assumptions.

    Andrei says: “The comparison implementation here would be an AXT-16-512 implementation running at slightly lower than nominal clock and voltage (in order to match the performance).”

    Let’s assume the AXT-16-512 is underclocked by 10% to get to the same performance as the Exynos 9820 and Snapdragon 855. Let’s also assume that an AXT-32-1024 is exactly double the performance of the AXT-16-512.

    So, a nominally clocked AXT-16-512 would have 110% the performance of the Snapdragon 855 and Exynos 9820. Double that, and you get 220% the performance, for the AXT-32-1024.

    Looking at the Huawei review, here are the numbers:

    GFXBench Aztec Ruins High

    Exynos 9820 and Snapdragon 855: ~16fps —> AXT-32-1024: 16fps + 120% = 35,2fps
    Apple A13: 34fps

    GFXBench Aztec Ruins Normal

    Exynos 9820 and Snapdragon 855: ~40fps —> AXT-32-1024: 40fps + 120% = 88fps
    Apple A13: 91fps

    GFXBench Manhattan 3.1

    Exynos 9820 and Snapdragon 855: ~69,5fps —> AXT-32-1024: 69,5fps + 120% = 153fps
    Apple A13: 123,5fps

    GFXBench T-Rex

    Exynos 9820 and Snapdragon 855: ~167fps —> AXT-32-1024: 167fps + 120% = 367fps
    Apple A13: 329fps

    It seems that at least on performance (with generous assumptions), if the new architecture fulfils all promises, it would be competitive, even slightly better than the Apple A13. The problem is that it won’t compete with the A13, but the A14...

    How did we get to Apple dominating GPUs too, so fast?
    Reply
  • drexnx - Tuesday, December 3, 2019 - link

    they're totally unafraid to spend as much die space as they need to get their performance scaling. look at a history of Ax die sizes and you'll see they're all over the place Reply
  • Spunjji - Tuesday, December 3, 2019 - link

    Agreed. It's their vertical integration at work - they're the only company prepared to spend that much die area on performance because they're the only company besides Samsung that can guarantee to sell all every chip they make in a high-end, high-margin device. Reply
  • Andrei Frumusanu - Tuesday, December 3, 2019 - link

    Apple's GPUs are the second smallest in the space - only Qualcomm uses less die area. Reply
  • vladx - Wednesday, December 4, 2019 - link

    When you extort your customers like Apple does, you can afford to design more expensive SoCs while still keeping huge profits.

    Apple is the biggest example of what a toxic system capitalism can become.
    Reply

Log in

Don't have an account? Sign up now