GPU 2025: Market and Products

Description

The GPU 2025: Market and Products report is our annual summary of GPU and associated developments for the year. Jon Peddie Research’s annual GPU Market and Products report consolidates GPU developments for 2025. It includes announcements by the leading GPU suppliers, market size and share, and benchmark results.

The global PC GPU market exceeded 251 million units in 2024, with a 6% year-over-year increase. More GPUs (~24%) are shipped each year than CPUs.

Table of Contents

• INTRODUCTION
  • How do modern GPUs work?
  • Where do GPUs come from?
  • Where do GPUs go?
    • Integrated vs. discrete performance
    • Personal computers
  • Platforms
  • In what form are GPUs supplied?
• HIGHLIGHTS OF THE YEAR
• THE FIRST QUARTER OF 2025
  • Snapdragon X series expands
  • VESA extends DisplayPort 2.1 cable reach
  • Nvidia 5000 series announced at CES
  • AMD went to Las Vegas
  • Is Imagination Technologies China’s backdoor to AI tech?
  • Nvidia GPUs double performance in 3D, video, generative AI
  • Intel’s B570 shows excellent scaling
  • Nvidia DLSS 4 introduces Blackwell GPU
  • Nvidia RTX 5080, the replacement for RTX Ti
  • AMD introduces the RX 9070 XT and RX 9070
  • AIBs undergo rigorous assessment in JPR MTTL testing
  • Bolting out of the shadows
  • Apple’s M3 Ultra is the most powerful graphics offering to date
  • AMD vs. everything Nvidia—1,424 tests
  • Nvidia’s midrange RTX 5070 AIB tested
• THE SECOND QUARTER OF 2025
  • Intel’s Arc GPUs’ overhead associated with older CPU chips
  • India planning a GPU
  • Big week for U.S. chips as export rules get rolled back
  • Nvidia RTX Pro 6000 Blackwell workstation performance
  • AMD to launch Radeon RX 9060 XT at end of Computex
  • AMD introduces neural ray tracing
  • Maxsum stacks two Intel Pro B60s
  • Nvidia GeForce RTX 5060
  • Intel unlikely to enter the mobile workstation dGPU market
  • Testing the new entry-level AMD Radeon 9060
  • Shader execution reordering, opacity microamps added to DirectX
  • Game over?
  • How does FSR4 compare to FSR3 and DLSS 4?
  • Intel texture compression relieves memory crunch
  • Steam’s in-game overlay reports GPU performance
  • AMD’s next-generation GPU architecture: UDNA and RDNA 5
  • Intel’s 18A process node
  • HDMI 2.2—it’s the wire, dummy
• THE THIRD QUARTER OF 2025
  • Arm’s CPU-first AI push gets teeth with Lumex
  • Nvidia Rubin CPX and disaggregated long-context inference
  • China loves Nvidia
  • The problem with UMA is the M
  • Meta swallows Rivos—now what?
  • Chinese GPUs and nm
  • Lisuan G100 GPU does well on OpenCL
  • Moore Threads’ MUSA GPU tested
  • Chinese GPU IPOs
  • Sparkle joins the dual-GPU club
  • Life can be a game, so go play
  • AMD shows split ends with TressFX 5.0
  • Bolt Graphics’ Zeus GPU technical overview
  • Intel XeSS 2 supports AI-driven frame generation on all GPUs
  • The B50 is not a new bomber, it’s Intel’s workstation AIB
  • Loongson 9A1000 GPU development
  • Oxmiq Labs enters the GPU market, softly
• THE FOURTH QUARTER OF 2025
  • Super-duper computer—AMD does it again
  • Super-duper Nvidia
  • Intel sort of clarifies its GPU roadmap
  • First made-in-USA Blackwell rolls out
  • Jensen delivers a GPU to SpaceX
  • Panther Lake’s Xe3: Intel’s biggest iGPU yet
  • Imagination’s open-source driver grows up
  • Apple’s A19 and M5 chips
  • Intel unveils Crescent Island GPU for the AI inference era
  • What started as a pixel pusher became the heart of AI
  • Why, with 146 competitors already in the field, would OpenAI go custom?
  • AMD AIPs are going to have a great 2H’26
  • GPUs vs. ASICs—the threat from within
  • Xiang Di Xian introduced its Fuxi A0 GPU
  • AMD gets closer to revealing the MI430X GPU
  • Nvidia’s Game of Thrones
  • AMD will pay Trump’s protection money
  • Moore Threads files IPO, raises $1.1 billion
  • RISC-V GPUs
• INDEX