GPU Developments in 2017

Posted: 03.14.18

Jon Peddie Research (JPR), the industry's research and consulting firm for graphics and multimedia has just released in annual review of GPU developments for 2017.

In spite of the overall slow decline of the PC market, PC-based GPUs sales, which include workstations, have been increasing. In the mobile market, integrated GPUs have risen at the same rate as mobile devices and the SoC in them. The same is true for the console market where integrated graphics are in every console and they too have increased in sales over the year.

Figure 1: Mobiles devices use more GPUs than all other platforms combined.


We didn’t count the GPUs used in the data center since they don’t drive a display (so the “G” in data-center GPUs is irrelevant). Even if we had included them they would not have made any noticeable amount.

Nearly 28% of the world’s population bought a GPU device in 2017, and that’s in addition to the systems already in use. And yet, probably less than half of them even know the term GPU, what it stands for, or what it does. To them the technology is invisible, and that means it’s working—they don’t have to know about it.

The ubiquity and invisibility of GPUs speaks to their success not just in marketing terms but in terms of real contribution to quality of life, safety, entertainment, and the advancement of science.

The GPU has evolved since its introduction in the late 1990s from a simple programmable geometry processor to an elaborate sea of 32-bit floating point processors running at multiple gigahertz speeds. The software supporting and exploiting the GPU, the programming tools, APIs, drivers, applications, and operating systems have also evolved in function, efficiency, and unavoidably, complexity.

The market for, and use of, GPUs stretches from supercomputers and medical instrumentation to gaming machines, mobile devices, automobiles, and wearables. Just about everyone in the industrialized world has at least a half dozen GPUs, and technophiles can easily count a dozen or more.

The manufacturing of GPUs approaches science fiction with features that will move below 10 nm next year and have a glide-path to 3 nm, and some think even 1 nm—Moore’s law is far from dead, but is getting trickier to tease out of the Genie’s bottle as we drive into subatomic realms that can only be modeled and not seen.

We acquire 90% of the information we digest through our eyes. Naturally, we want and need abundant sources of information-generating devices to supply our unquenchable appetite for yet more information. And the machines we build to service and support us have a similar demand for information, albeit maybe not visible, although in some case such as in robots and autonomous vehicles that’s exactly what they need. The GPU can not only generate pixels, but it can also process photons captured by sensors.

Over the past 12 months we have seen a few new, and some clever adaptations of GPUs that show the path for future developments, and subsequent applications.

2017 was an amazing year for GPU development driven by games, eSports, AI, crypto currency mining, and simulations. Autonomous vehicles started to become a reality, as did augmented reality. The over hyped consumer-based PC VR market explosion didn’t happen, and had little to no impact on GPU developments or sales. Most of the participants in VR already had a high-end system and the HMD was just another display to them.

Mobile GPUs, exemplified by Qualcomm, ARM, and Imagination Technologies’ customers introduced some amazing devices with long batter life, screens at or approaching 4K, and in 2017 we started talking about and showing HDR. 

The calendar is an artificial event marker and the developments in GPUs doesn’t pause for holidays, or annual clock ticks. Many is not all the developments we will see in 2018 were started as early as 2015, and that three to four-year lead time will continue, if not get longer as we learn how to deal with chips constructed with billions of transistor manufactured at feature sizes smaller than X-rays.

2018 is going to be an even more amazing year, with AI being the leading applications that will permeate every sector of our lives.

About the JPR GPU Developments in 2017 Report
The report is free to all subscribers of JPR. Individual copies of the report can be purchased for $100.

For information about purchasing the JPR GPU Developments in 2017 Report, please call 415- 435-9368 or visit Jon Peddie Research at  Based in Tiburon, California, Jon Peddie Research provides consulting, research, and other specialized services to technology companies, including graphics development, multimedia for professional applications and consumer electronics, high-end computing, and Internet-access product development.


GPU Developments Table of Contents
Executive Summary...7
Scope of report...9
First quarter 2017...9
Vega10 architecture and technology...10
High Bandwidth Cache (HBC) and HBM2 technology...11
New NCU...13
Enhanced geometry pipeline...15
Next generation pixel engine...16
FreeSync 2 display technology...18
Radeon Crimson Software...22
What do we think?...22
AMD Embedded claimed 75% of discrete graphics market share in global casino gaming...23
AMD accused companies of infringing patents...25
Imagination Technologies...26
New PowerVR Series8XE Plus GPUs for mid-range markets...26
Inside Series8XE Plus GPUs...29
Software tools and availability...30
What do we think?...31
MediaTek got some serious graphics...31
What do we think?...32
PowerVR Furian GPU architecture...32
Furian architecture highlights...33
What do we think?...35
Intel Xeon E3-1200...38
GeForce GTX 1050 Ti and 1050 for laptops...38
Nvidia GTX 1080 Ti...39
What do we think?...40
Adjacent developments...44
The API path...45
Adoption and capabilities grew for Vulkan...47
Vulkan extension sets...48
One size fits all?...48
What do we think?...48
I’m seeing more than I want to, or maybe not enough...48
Closing comments...51
Micron’s 3D Xpoint memory structure...51
Recipient of foundation graphics patents made claims...52
What do we think?...59
Second quarter 2016...59
Mali got an ISP module...60
What do we think?...62
ARM’s Mali-Cetus display processor...62
What do we think?...65
Imagination Technologies...65
The Apple of my eye turned black and blue...66
Going it alone...66
More compute, less pixel pushing...66
Collateral damage...67
Adjacent developments...68
Third quarter 2017...68
The road to GPU control: the challenges of vertical integration...69
External GPU...72
Imagination Technologies...73
PowerVR Series9XE and Series9XM GPUs...73
What do we think?...75
Launched 8th generation Core processors...75
Adjacent developments...77
Multiple displays can increase productivity by 42%...77
To MXM or not MXM, is that the question?...78
AMD offered Radeon ProRender for Blender and Solidworks...79
What do we think?...81
Fourth quarter 2017...82
AMD APUs roll out...82
What do we think?...84
Imagination Technologies...84
Imagination breakup and sale...84
Who’s GPU?...86
What do we think?...86
Intel and AMD collaborate on technology...87
Back to the gossipy stuff...89
Snapdragon 845...93
Kryo 385 CPU...94
Adreno 630 GPU...96
Spectra 280 ISP...97
Fabulous fab...98
What do we think and final thoughts...98
Adjacent developments...99
PCIe next gen is here—for.oh...99
Summary and Conclusion...102
No catching up...103

Table of Figures
Figure 1: Mobiles devices use more GPUs than all other platforms combined...7
Figure 2: Former AMD’s Raja Koduri introduced the company’s flagship GPU for 2017: Vega. (Source: AMD)...10
Figure 3: A high-level look at Vega’s key microarchitecture components. (Source: AMD)...11
Figure 4: The primary trio of markets for high-performance discrete GPUs. (Source: AMD)...12
Figure 5: Cross-sectional look at first-generation HBM memory, as implemented with AMD’s previous-generation Fiji GPU. (Source: AMD)...13
Figure 6: The current (pre-Vega) heart of the GCN architecture: the Polaris-generation Compute Unit (CU). (Source: AMD)...14
Figure 7: Vega’s next generation CU, the “NCU,” boosts IPC by boosting both parallel processing and faster clocks. (Source: AMD)...15
Figure 8: Vega’s Intelligent Workgroup Distributor claims to more effectively allocate incoming geometry to available geometry, compute, and pixel engine resources: (Source: AMD)...16
Figure 9: Vega’s DrawStream Binning Rasterizer should boost performance, especially for applications that can defer rendering. (Source: AMD)...17
Figure 10: Vega’s render back-ends were then direct clients of the L2 cache. (Source: AMD)...18
Figure 11: FreeSync 2 enabled better-than-sRGB displays without costly monitor-side HDR tone- mapping. (Source: AMD)...20
Figure 12: FreeSync 2 allowed the system—application, GPU, and display—to select at the optimal color gamut, not quite HDR10 but significantly beyond sRGB. (Source: AMD)...21
Figure 13: AMD is major supplier in slot machine market. (Source: AMD)...24
Figure 14: Arm’s Dynamiq CPU structure. (Souce: Arm)...26
Figure 15: Full HD is the dominant resolution in smartphones. (Source: Imagination)...27
Figure 16: Full HD is also still significant in TV through 2020. (Source: Imagination)...27
Figure 17: New PowerVR GPU roadmap introduced the Series8XE Plus. (Source: Imagination) 28 Figure 18: Imagination’s new PowerVR 8XCE Plus GPU design...29
Figure 19: Imagination’s PowerVR 8XE—area/cost saving versus competition. (Source: Imagination)...30
Figure 20: Imagination Technologies GPU development history. (Source: Imagination)...33
Figure 21: Imagination’s Furian overall architecture. (Source: Imagination)...34
Figure 22: Imagination’s Furian PowerVR Furian ALU cluster overview. (Source: Imagination) 34 Figure 23: Intel’s 7th  gen CPU with integrated graphics. (Source: Intel)...37
Figure 24: Nvidia’s new GTX1080TI. (Source: Nvidia)...39
Figure 25: Qualcomm component block diagram Snapdragon 835...42
Figure 26: The history of graphics APIs...46
Figure 27: The screen door effect when seen up close...49
Figure 28: Log of sinusoidal Contrast sensitivity (Campbell, F. W. and Robson, J. G. (1968), Application of Fourier Analysis to the Visibility of Gratings)...50
Figure 29: Micron’s 3D point memory structure. (Source: Micron)...52
Figure 30: Historical path of graphics patents...55
Figure 31: Arm's Mali-C71 ISP supported multiple inputs and outputs. (Source: Arm)...60
Figure 32: VPUs are at the center of advanced image processing, CNNs, and augmented reality. 61 Figure 33: Arm’s Mali-Cetus manages Android windows and composition. (Source: Arm)...63
Figure 34: Arm’s Mali-Cetus AFBC DMA Unit. (Source: Arm)...64
Figure 35: Display output unit. (Source: Arm)...65
Figure 36: Coprocessor interface and HDR composition. (Source: Arm)...65
Figure 37: Multi-monitor deployment has steadily increased over time...77
Figure 38: AMD’s Ryzen 7 2700U with 4 cores, 8 threads, 10 CUs running at 2.2 GHz, with boost up to 3.8 GHz. (Source: AMD)...82
Figure 39: Ryzen processor with Radeon graphics, Zen meets Vega...83
Figure 40: Intel’s embedded multi-die interconnect bridge uses multiple routing layer to reduce z- height. (Source: Intel)...88
Figure 41: Intel introduces a new product in the 8th Gen Intel Core processor family that combines high-performance CPU with discrete graphics for a thin, sleek design. (Source: Intel)...89
Figure 42: Nvidia’s Holodeck promises fully immersive VR collaboration in the workplace. (Source: Nvidia)...91
Figure 43: The Pegasus AI computer is roughly the size of a license plate and can be installed in a vehicle to equip it for autonomous driving. (Source: Nvidia)...92
Figure 44: Stylized Sanpdragon 845 block diagram...93
Figure 45: Qualcomm’s Kryo 385 CPU. (Source: Qualcomm)...95
Figure 46: Snapdragon 845’s system cache. (Source: Qualcomm)...95
Figure 47: Qualcomm’s Sanpdragon Adreno 630 fully supports Rec.2020 HDR. (Source: Qualcomm)...96
Figure 48: Bus bandwidth has gone up exponentially for the past 43 years and shows no sign of slowing down...100
Figure 49: Improvement in GFLOPS of GPUs over time...103
Figure 50: Relative die size of GPUs as the feature set gets smaller. Smaller die with more transistors...104

Table of Tables
Table 1: Imagination’s new 8XE GPUs were 4–8 times faster than the 8XE. (Source: Imagination)....28
Table 2: Intel Codenames...36
Table 3: Comparison: GTX108TI to GTX1080 and Titan X...40
Table 4: AMD’s Ryzen APU family...84
Table 5: Comparison of Qualcomm’s Snapdragon 835 to Snapdragon 845...94
Table 6: Computer buses characteristics...100

Login to download