Can we get Intel Arc GPUs working on a Pi before Nvidia kills them?
Jeff Geerling 1:44:37
52,218 views · 1,853 likes Watch on YouTube ↗
It's been five years since I started working towards full eGPU support on the Raspberry Pi.
And we're finally here. Check out all the graphics cards I'm testing on: https://pipci.jeffgeerling.com
And PLEASE consider donating to the UOAA, who has helped me and millions of ostomates over the years, to live a better life after surgery. You can use the donation option on YouTube, or support the UOAA directly: https://www.ostomy.org/donate/
You can also donate on my 5K Run for Resilience fundraising page—any option supports the UOAA directly! https://runsignup.com/jeff-geerling
Other things I mentioned in the video:
- Run LLMs with llama.cpp on GPU on Pi: https://www.jeffgeerling.com/blog/2024/llms-accelerated-egpu-on-raspberry-pi-5
- Full Pi eGPU acceleration with tiny patch: https://www.jeffgeerling.com/blog/2025/full-egpu-acceleration-on-pi-500-15-line-patch
- JMT eGPU Dock + M.2 adapter: https://amzn.to/4mLcnNF
- OCuLink cable: https://amzn.to/4gRWBiO
- ThirdReality Smart Zigbee Outlets (for power monitoring): https://amzn.to/3IItDW6
Support me on Patreon: https://www.patreon.com/geerlingguy
Sponsor me on GitHub: https://github.com/sponsors/geerlingguy
Merch: https://www.redshirtjeff.com
2nd Channel: https://www.youtube.com/@GeerlingEngineering
3rd Channel: https://www.youtube.com/@Level2Jeff
Contents:
00:00:00 - Intro and UOAA fundraiser info
00:03:12 - The setup - Pi 500+ and an eGPU Dock
00:04:55 - The 23-line Linux kernel patch and setup
00:07:49 - Doom 3 on 7900 XT
00:08:57 - Can it run Crysis? (And Steam?)
00:12:31 - You can actually *see* it run Crysis
00:13:46 - Doom Eternal on 7900 XT (RTX ON)
00:19:00 - Portal 2 (almost works)
00:19:28 - Ultimate Epic Battle Simulator II (CPU-crusher)
00:21:47 - Power consumption measurements
00:22:54 - SSH from Windows 11 to check on UEBS II
00:24:38 - Minecraft, but faster
00:27:12 - Practicality and AI/LLMs
00:28:10 - Open WebUI and llama.cpp for AI models (llama and gpt-oss)
00:31:00 - Special guests!
00:32:58 - gpt-oss 20B on the 7900 XT (and AI lies!)
00:35:34 - Keeping your Sony cams cool (featuring audio insanity)
00:35:50 - Chat questions - AI TOPS and Bookworm vs Trixie
00:37:08 - Pi-apps and Steam install (and more chat questions)
00:40:45 - GPU swap, AMD Radeon RX 9070 XT edition
00:44:47 - What happens when Mesa's too old (a little more kernel talk)
00:48:27 - RX 9070 XT works (with newer Mesa), plus gaming
00:52:25 - But Steam doesn't (with newer Mesa)
00:53:35 - llama.cpp AI on the 9070 XT (and power consumption)
00:58:18 - Let's get to INTEL (B580, A750)
01:00:45 - What happens on Intel?
01:01:05 - Minisforum DEG1 eGPU dock doesn't work as well
01:02:03 - Intel, seriously? Copy protection??
01:02:52 - The curse of the blinking cursor
01:04:42 - Wayland a no-go, but what about X
01:07:08 - Intel B580 is blue
01:09:27 - Switching to Trixie, recompiled Mesa
01:12:17 - Jeff is so yellow
01:12:55 - B580 experience and new Trixie features
01:17:55 - Debugging - dmesg, lspci, firmware, PCIe, Resizeable BAR
01:20:30 - Disabling iGPU, enabling PCIe Gen 3
01:24:38 - A brief field trip!
01:26:42 - Chat corrects Jeff on PCIe link speed reporting (ft nvtop)
01:29:00 - B580 power draw, current status, and CO2 levels
01:33:24 - Old school value with the RX 460
01:38:20 - Chat questions - early bootloader not displayed
01:38:54 - AI on an old GPU? Is it useful?
01:41:58 - Even an old GPU is way faster than Pi's CPU
01:43:06 - RX 460 AI power draw
And we're finally here. Check out all the graphics cards I'm testing on: https://pipci.jeffgeerling.com
And PLEASE consider donating to the UOAA, who has helped me and millions of ostomates over the years, to live a better life after surgery. You can use the donation option on YouTube, or support the UOAA directly: https://www.ostomy.org/donate/
You can also donate on my 5K Run for Resilience fundraising page—any option supports the UOAA directly! https://runsignup.com/jeff-geerling
Other things I mentioned in the video:
- Run LLMs with llama.cpp on GPU on Pi: https://www.jeffgeerling.com/blog/2024/llms-accelerated-egpu-on-raspberry-pi-5
- Full Pi eGPU acceleration with tiny patch: https://www.jeffgeerling.com/blog/2025/full-egpu-acceleration-on-pi-500-15-line-patch
- JMT eGPU Dock + M.2 adapter: https://amzn.to/4mLcnNF
- OCuLink cable: https://amzn.to/4gRWBiO
- ThirdReality Smart Zigbee Outlets (for power monitoring): https://amzn.to/3IItDW6
Support me on Patreon: https://www.patreon.com/geerlingguy
Sponsor me on GitHub: https://github.com/sponsors/geerlingguy
Merch: https://www.redshirtjeff.com
2nd Channel: https://www.youtube.com/@GeerlingEngineering
3rd Channel: https://www.youtube.com/@Level2Jeff
Contents:
00:00:00 - Intro and UOAA fundraiser info
00:03:12 - The setup - Pi 500+ and an eGPU Dock
00:04:55 - The 23-line Linux kernel patch and setup
00:07:49 - Doom 3 on 7900 XT
00:08:57 - Can it run Crysis? (And Steam?)
00:12:31 - You can actually *see* it run Crysis
00:13:46 - Doom Eternal on 7900 XT (RTX ON)
00:19:00 - Portal 2 (almost works)
00:19:28 - Ultimate Epic Battle Simulator II (CPU-crusher)
00:21:47 - Power consumption measurements
00:22:54 - SSH from Windows 11 to check on UEBS II
00:24:38 - Minecraft, but faster
00:27:12 - Practicality and AI/LLMs
00:28:10 - Open WebUI and llama.cpp for AI models (llama and gpt-oss)
00:31:00 - Special guests!
00:32:58 - gpt-oss 20B on the 7900 XT (and AI lies!)
00:35:34 - Keeping your Sony cams cool (featuring audio insanity)
00:35:50 - Chat questions - AI TOPS and Bookworm vs Trixie
00:37:08 - Pi-apps and Steam install (and more chat questions)
00:40:45 - GPU swap, AMD Radeon RX 9070 XT edition
00:44:47 - What happens when Mesa's too old (a little more kernel talk)
00:48:27 - RX 9070 XT works (with newer Mesa), plus gaming
00:52:25 - But Steam doesn't (with newer Mesa)
00:53:35 - llama.cpp AI on the 9070 XT (and power consumption)
00:58:18 - Let's get to INTEL (B580, A750)
01:00:45 - What happens on Intel?
01:01:05 - Minisforum DEG1 eGPU dock doesn't work as well
01:02:03 - Intel, seriously? Copy protection??
01:02:52 - The curse of the blinking cursor
01:04:42 - Wayland a no-go, but what about X
01:07:08 - Intel B580 is blue
01:09:27 - Switching to Trixie, recompiled Mesa
01:12:17 - Jeff is so yellow
01:12:55 - B580 experience and new Trixie features
01:17:55 - Debugging - dmesg, lspci, firmware, PCIe, Resizeable BAR
01:20:30 - Disabling iGPU, enabling PCIe Gen 3
01:24:38 - A brief field trip!
01:26:42 - Chat corrects Jeff on PCIe link speed reporting (ft nvtop)
01:29:00 - B580 power draw, current status, and CO2 levels
01:33:24 - Old school value with the RX 460
01:38:20 - Chat questions - early bootloader not displayed
01:38:54 - AI on an old GPU? Is it useful?
01:41:58 - Even an old GPU is way faster than Pi's CPU
01:43:06 - RX 460 AI power draw
Playback is via YouTube's official embedded player. Data from YouTube; Exumo is not affiliated with YouTube.