https://poulsback.in/blog/ Notes from bringing SGX535 back to life: driver design, real hardware and the measurements behind each step. en-us Mon, 09 Nov 2026 20:55:00 +0900 https://poulsback.in/blog/poulsback-needs-you/ https://poulsback.in/blog/poulsback-needs-you/ Mon, 09 Nov 2026 20:55:00 +0900 We develop poulsback on two machines, and there were dozens of GMA 500 designs. A new live USB stick collects a device report in one boot, with nothing to type and nothing installed. Here is how to make one and send it. https://poulsback.in/blog/same-gpu-a-two-ghz-atom/ https://poulsback.in/blog/same-gpu-a-two-ghz-atom/ Sun, 08 Nov 2026 16:35:00 +0900 A 5.6-inch Fujitsu LOOX U/C50N pairs the same Poulsbo SGX535 with a 2.00 GHz Atom Z550. Separating the display aperture from the SGX 2D window brought it up cleanly, and matched benchmarks show where a faster CPU moves the frame rate. https://poulsback.in/blog/teaching-sgx535-to-build-its-own-mipmaps/ https://poulsback.in/blog/teaching-sgx535-to-build-its-own-mipmaps/ Thu, 05 Nov 2026 10:15:02 +0900 The smallest mip holds four bytes of pixel data. Reaching it on SGX535 meant understanding tile writes, ordering six GPU passes and checking application draw state and queries. https://poulsback.in/blog/a-tablet-a-familiar-core-a-different-machine/ https://poulsback.in/blog/a-tablet-a-familiar-core-a-different-machine/ Wed, 04 Nov 2026 11:30:07 +0900 A working Linux console, a deliberate driver refusal and an unexpected x87 test failure: the Q550 turns a second machine into a useful set of questions. https://poulsback.in/blog/a-cube-that-can-measure-the-driver/ https://poulsback.in/blog/a-cube-that-can-measure-the-driver/ Tue, 03 Nov 2026 14:33:00 +0900 A benchmark we can understand gives the neon cube another job: connect what the GPU draws with the time and CPU work needed to draw it. https://poulsback.in/blog/when-the-gears-finally-turned/ https://poulsback.in/blog/when-the-gears-finally-turned/ Tue, 03 Nov 2026 14:33:00 +0900 Getting glxgears onto the Dell’s panel joined Mesa, the kernel driver and X11. Moving its vertex work onto SGX535 was a later milestone. https://poulsback.in/blog/from-the-first-cube-to-the-logo/ https://poulsback.in/blog/from-the-first-cube-to-the-logo/ Tue, 03 Nov 2026 14:33:00 +0900 The first successful cube gave poulsback a recognisable shape. It also taught us to ask exactly which parts of a rendered frame were working. https://poulsback.in/blog/a-second-life-for-a-five-hundred-yen-dell/ https://poulsback.in/blog/a-second-life-for-a-five-hundred-yen-dell/ Tue, 03 Nov 2026 14:33:00 +0900 Around July, a non-working Dell from HARD OFF in Japan became the next poulsback machine. Repairing it gave the project a reference it could keep testing. https://poulsback.in/blog/how-poulsback-began/ https://poulsback.in/blog/how-poulsback-began/ Tue, 03 Nov 2026 14:33:00 +0900 I started on February 1, 2026, because I wanted Linux to work better on my VAIO P. A broken motherboard and a ¥500 Dell changed the plan. https://poulsback.in/blog/giving-the-gpu-room-to-work/ https://poulsback.in/blog/giving-the-gpu-room-to-work/ Tue, 03 Nov 2026 00:00:00 +0900 Moving vertex work onto SGX535 was the first step. A frame trace showed how much time the GPU still spent waiting for its own engines.