PS4 Pro DDR3 Read Speed Raised From 10.6 MB/s to 504 MB/s via DMA Patch

·14 views

A developer working on Linux for the PS4 Pro found that reads from its DDR3 memory were capped near 10.6 MB/s because the CPU fetched data in tiny 8-byte chunks through a narrow PCI window. The chip's southbridge contains a DMA engine that could handle the transfers, but no Linux driver existed for it, and Sony's gameOS driver is closed-source. The developer reverse-engineered the kernel binary to recover the chip's address mapping and the DMA command sequence. A patch was then built that binds the engine as a PCI device and routes transfers through a coherent buffer allocated at startup. Reported results show reads improving to about 504 MB/s and writes to about 425 MB/s, with the engine capable of roughly 1.4 GB/s.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Cross-Border Search Needs More Than Query Translation

Translating a search query once does not make retrieval work across languages, because tokenization, morphology, character normalization and word order all change what documents get matched. Languages such as Chinese, Japanese and Thai are written without spaces, so whitespace splitting turns whole sentences into single unmatched terms. Inflected languages like German, Russian and Arabic need stemming, or plural and case forms will never match each other. The article proposes a three-layer architecture of language-specific analysis, vector alignment and localized ranking, noting that omitting any layer leaves recall gaps in some languages. It also recommends keeping filters such as currency, dates and compliance rules separate from relevance scoring and measuring effectiveness per language rather than in blended averages.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

UUIDv7 Offers Database Speed but Its Embedded Timestamp Is Not a Reliable Clock

UUIDv7 identifiers are gaining favor because they keep database indexes efficient while staying unique without central coordination. Each ID embeds a millisecond timestamp, which means anyone who sees an ID can roughly infer when it was created. A reproduction described in an October 19, 2026 post showed that the embedded time can drift from the real clock, running about an hour ahead after the system clock was set back. The RFC allows implementations to adjust the timestamp, so it should be treated as a hint rather than an accurate record. Developers debate whether exposing this metadata matters, with the author recommending UUIDv7 internally while avoiding exposing it in sensitive public-facing contexts.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Pre-commit hook fixed to check staged translation files, not working tree

The author's Locale Guard tool, a GitHub Action and CLI for translation files, added a pre-commit hook in version 1.3.0. The first version of that hook checked files on disk, so it approved a commit that held an English key but lacked matching entries in other language files. The fix makes the check read staged content directly from git, so incomplete commits are now blocked. A pre-push hook and the pull request check were also added to catch anything skipped with --no-verify. Demo repositories for Husky and Lefthook are available for testing.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

What spec-driven development is and why AI coding tools are splitting into competing camps

The article explains spec-driven development (SDD), where requirements are written as a clear document that AI agents work from rather than from chat memory. It describes three traits of SDD: specs are version-controlled files, humans approve them before code is written, and every implementation step can be traced back to the spec. The author notes that the real differences between SDD tools lie in how specs relate to verification, not in feature counts. Three maturity levels are outlined: spec-first, spec-anchored, and spec-as-source, each answering differently who judges whether code is correct. The piece is the first part of a series comparing these tools.

0 comments Read more at DEV Community