SShortSingh.
Back to feed

Single RTX 5090 Replicates 12-GPU Cluster Proof, Using 41% Less Search Work

0
·1 views

A software developer rebuilt a 2023 NeurIPS game-solving algorithm and ran it on a single Nvidia RTX 5090 to benchmark a decade of GPU progress against a 12-card GTX 1080Ti cluster. The experiment used an identical solver, problem, and settings — only the hardware differed — targeting a mathematical proof for a 7x7 Go opening called JA. The single card completed the proof in 41.4 hours versus the cluster's 8.9 hours, but operated with just 1/16 the parallel workers. Despite the slower wall-clock time, the RTX 5090 required 41% less total search work, largely because its co-located trainer updated the neural network guidance roughly every 37 seconds compared to every 2.5 minutes on the cluster. The results suggest that more frequent model updates during online fine-tuning meaningfully reduce search effort, though the author notes worker count and update frequency remain statistically confounded in this single comparison.

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)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Why macOS Developers Are Moving Toward Unified SSH, RDP, and VNC Workflows

Developers using macOS as their primary workstation often juggle multiple separate tools to handle SSH, RDP, VNC, Telnet, and serial connections, creating a fragmented daily workflow. The growing complexity of managing many remote environments has prompted interest in unified remote access clients that organize connections by device rather than by protocol. Such tools aim to reduce context-switching by consolidating different connection types, authentication methods, and file transfer capabilities in one place. The author draws on experience building DartShell to highlight that creating a consistent cross-protocol user experience is a greater engineering challenge than simply supporting multiple protocols. For developers managing only a few servers, dedicated tools may suffice, but those overseeing larger, mixed environments stand to gain the most from a unified approach.

0
ProgrammingDEV Community ·

Why TypeScript Is Worth the Extra Effort, Even for Solo Developers

TypeScript is a superset of JavaScript that adds static typing and improved developer tooling, compiling down to plain JavaScript at the end. A developer initially dismissed it as unnecessary complexity but changed their view after building a SaaS project, where TypeScript flagged ignored bugs during deployment before they could reach users. While the added type definitions and compiler errors can feel like extra overhead, especially on small or prototype projects, the trade-off pays off in fewer runtime bugs and more maintainable code. For larger, long-term applications, TypeScript has become an industry standard largely because it enforces consistency across codebases and catches mistakes early in development.

0
ProgrammingDEV Community ·

Xserver's fail2ban blocks SSH after repeated auth attempts, no manual unban available

Xserver officially confirmed that its fail2ban security mechanism temporarily blocks source IPs that exceed a certain authentication attempt threshold, though the exact parameters are kept confidential. The issue commonly affects WordPress developers whose SSH clients or tools like paramiko try multiple keys per connection, rapidly accumulating failed attempts across parallel sessions. Blocked users experience two escalating symptoms: an initial 'Connection closed' response followed by a full 'Connection refused' if retries continue. Xserver provides no manual unban channel, meaning affected users must either wait roughly half a day to a full day for automatic release or switch to a different source IP to regain immediate access. Minimizing the number of keys offered per connection — by disabling agent forwarding and key discovery in tools like paramiko — can help prevent triggering the block in the first place.

0
ProgrammingDEV Community ·

Developer builds open-source tool to give AI read-only access to Microsoft Loop safely

A developer has released an open-source tool called loop-reader-mcp that allows AI assistants to read Microsoft Loop pages without bypassing user permissions. The challenge arose because Microsoft Loop lacks a content API and stores data across SharePoint Embedded and OneDrive, while SharePoint Embedded does not accept standard per-user authentication tokens for content retrieval. The solution splits the process into two steps: content discovery runs under the user's identity via Microsoft Graph Search, which automatically trims results to only pages that user can access, while actual retrieval uses an app-level identity gated strictly to items the user personally discovered. This design ensures that authorization decisions are enforced by Microsoft rather than the developer's own code, preventing any user from accessing Loop pages they do not have permission to view. The project is available on GitHub under an MIT license and exposes three read-only tools through a remote Model Context Protocol server.

Single RTX 5090 Replicates 12-GPU Cluster Proof, Using 41% Less Search Work · ShortSingh