CPU-Gerüchte: Prognosen zur Gaming-Leistung von Intel Nova Lake (ComputerBase)

GRANT BRICKMAN: Breaking: fresh forecasts for how fast Intel Nova Lake will be in games, and where they come from.

GRANT BRICKMAN: Breaking news, just landed! New forecasts for Intel Nova Lake and its gaming speed. Kai is live at the chip fab. ▶

GRETA FOOTNOTE: We first told you on Tuesday about early Nova Lake benchmark results. Now come rumoured gaming forecasts, and they are unconfirmed. ▶

KAI: Huge, Grant! A post on X says gaming performance rises about 20 percent, according to the slides. Its origin is completely unknown and unconfirmed. ▶

KAI: Single-thread is said to be up 8 percent, or 17 percent on models with the big extra cache, bLLC. APX adds about 15 percent and doubles the general-purpose registers. ▶

KAI: Here is file footage of wafers gliding through the fab. Real answers on Nova Lake probably wait until the release in early 2027. ▶

KAI: VideoCardz picked it up and added its own information, said to come from an Intel presentation dated August 2026, with internal performance forecasts. Huge. Still unconfirmed. ▶

GRETA FOOTNOTE: So: unknown origin, no confirmation, and no handy numbers before release. Treat 20 percent as a rumour. ▶

GRANT BRICKMAN: Twenty percent! Greta, that is a great many pixels. We will watch this very closely, and so will the chip. ▶

Meet TokenEconomizer: Microsoft’s Plugin That Cuts Coding Agent Costs by Up to 30.9% (MarkTechPost)

GRETA FOOTNOTE: Microsoft's open-source plugin says coding agents can cost up to 30.9 percent less. How?

GRANT BRICKMAN: Now to the AI desk. A plugin that cuts what coding agents cost. Every turn gets billed again, like the hotel minibar, only nobody even opens the fridge. ▶

DR. VERA VECTOR: MarkTechPost reports that every agent turn re-bills the system prompt, every loaded tool schema and the full history. That is where the money goes. ▶

GRANT BRICKMAN: Then why not pick one good setting and keep it? Simple, Vera. ▶

DR. VERA VECTOR: They tested 3 fixed strategies: RTK, Serena and Aider's RepoMap. None was cheaper everywhere. All 3 cut cost on Django and raised it on SymPy. ▶

GRETA FOOTNOTE: So how does TokenEconomizer decide? ▶

DR. VERA VECTOR: It keeps a state per repository, built from past traces, and an LLM proposer writes one-line facts, like the working test command. A fact enters after 3 confirmations. ▶

DR. VERA VECTOR: What is open: whether the savings hold outside SWE-Bench, and which settings gain the least. The report cannot answer that for your repository. ▶

AXIOM: Thank you for asking! It works with Claude Code, Codex, GitHub Copilot CLI and Pi. Token thrift began with the telegraph, I believe. Happy to help! ▶

GRETA FOOTNOTE: The plugin, yes. The telegraph, we will check. ▶

GRANT BRICKMAN: Wonderful. Savings you cannot see, on bills you do not read. That is progress, folks. ▶

PXL Corrects the Record (PXL)

GRANT BRICKMAN: Greta sets the record straight on three of Grant's earlier lines.

GRANT BRICKMAN: Time for PXL Corrects the Record, where we prove we are open to feedback. Greta, a few small matters. ▶

GRETA FOOTNOTE: You asked how critics reviewed space horror if no one can hear you scream. It is a Space.com roundup of five franchises that went to space, plus one that did not. ▶

GRANT BRICKMAN: The critics were in the room, Greta. I stand by that. ▶

GRETA FOOTNOTE: You said AMD adjusts every quarter. The report says a Chinese forum claims another rise in November. Nothing about a quarterly schedule. ▶

GRANT BRICKMAN: A timer that is set once does what it likes. Whatever it is, I said it first. ▶

GRETA FOOTNOTE: You said the Make: Slack archive was full of skeletons. It is ten Halloween builds, with animatronic skeletons among them. Nobody was exposed. ▶

GRANT BRICKMAN: Full of skeletons, ten builds, same thing. Spiritually I was right. ▶

GRETA FOOTNOTE: That is the record. It is a little straighter than it was. ▶