Corrections Policy

Last updated: April 9, 2026

Accuracy matters most when readers are making expensive hardware decisions. If you spot a factual error, outdated availability note, broken specification, unclear disclosure language, or a recommendation that no longer matches current hardware context, email hello@groktechgadgets.com with the page URL and the issue.

What we correct

  • Factual specification errors
  • Outdated pricing or model context when it changes the advice
  • Broken links, retailer references, or missing disclosure language
  • Ambiguous statements about workload fit, upgradeability, thermals, or compatibility

How we evaluate reports

  • We review the claim against the page intent and current editorial framework.
  • We check whether the issue changes the recommendation materially or only needs clarifying context.
  • We may update related comparison, requirement, or glossary pages if the same issue appears elsewhere.

How updates appear on the site

If a correction changes a material recommendation, specification, or compatibility claim, the affected page is revised and its modified date is refreshed. Smaller wording or formatting fixes may be handled without a site-wide note, but we still aim to keep the published guidance accurate and current.

What helps us resolve issues faster

Where to send issues

Email hello@groktechgadgets.com and include as much detail as possible so the issue can be checked quickly. Good correction requests help us improve pages faster and reduce the chance that the same error spreads to comparison or buying guides.

Corrections log

This page lists factual corrections we have made to published recommendations. Minor copy edits (typo fixes, broken links repaired) are not logged. Factual corrections — wrong specs, wrong prices, wrong benchmark numbers, wrong recommendations — always are.

Entries are listed in reverse chronological order. The "Affected page" field links to the page where the correction was applied, and the visible Correction callout on that page references this log.

2026

May 4, 2026 — RTX 5090 laptop VRAM spec
Affected page: Best AI laptops 2026. Originally stated the RTX 5090 laptop ships with 32 GB GDDR7. Corrected to 24 GB GDDR7 — the 32 GB figure belongs to the desktop RTX 5090. Both desktop and laptop SKUs use the GB203 die but the laptop ships with reduced memory. Source: NVIDIA Blackwell mobile spec page, verified against ASUS ROG Strix SCAR 18 product listing.
April 28, 2026 — RTX 5070 laptop memory configuration
Affected pages: Best AI laptops 2026, RTX laptop GPU ranking 2026. Initially listed the RTX 5070 laptop as 8 GB GDDR7 only. NVIDIA confirmed a 12 GB variant in an April driver release note; both pages updated. The 12 GB version uses the same GB206 die and 128-bit bus, so memory bandwidth is unchanged at 384 GB/s, but the extra VRAM is meaningful for SDXL and 13B-class LLMs. ASUS, Lenovo, and MSI have all confirmed 12 GB SKUs shipping in June 2026.
April 14, 2026 — AMD RX 9070 ROCm support
Affected page: Best budget GPU for AI. Originally stated AMD RDNA3 consumer cards were not officially supported by ROCm. Corrected after AMD's ROCm 7.2 release (January 2026) added official consumer Radeon support for the RX 9070, RX 9070 XT, RX 9070 GRE, RX 9060 XT, and a subset of RX 7000-series cards. PyTorch with ROCm now installs cleanly on these GPUs; the page now reflects this.
March 22, 2026 — RTX 4060 Ti 16 GB recommendation
Affected page: Best budget GPU for AI. We had ranked the RTX 4060 Ti 8 GB ahead of the 16 GB variant on price-per-frame for gaming. For AI workloads, the 16 GB version is meaningfully better — Stable Diffusion XL workflows that fit in 16 GB do not fit in 8 GB at any quality setting, and 8-bit 13B LLMs become workable on 16 GB but never on 8 GB. The ranking now reflects this AI-specific reality.
March 11, 2026 — MacBook M4 Pro unified memory bandwidth
Affected page: MacBook LLM performance. Listed M4 Pro memory bandwidth as 273 GB/s. Corrected to 273 GB/s for the 12-core CPU variant; the 10-core variant runs at 196 GB/s. Both variants are sold under the same "M4 Pro" name and the difference materially affects LLM token generation speed.
February 18, 2026 — Llama 3 70B 4-bit VRAM requirement
Affected page: Local LLM hardware guide. Originally cited Llama 3 70B Q4 as fitting in 24 GB VRAM. Corrected to 40 GB+ for usable context windows; the 24 GB figure was for the model weights alone and didn't account for KV cache at meaningful context length. The page now distinguishes "model fits in VRAM" from "model is usable" — the latter requires 1.5–2× the model size for KV cache and activations.
February 2, 2026 — RTX 4090 vs RTX 4080 Stable Diffusion benchmark
Affected page: Best laptops for Stable Diffusion 2026. An earlier benchmark chart showed the RTX 4080 laptop at ~85% of the RTX 4090 laptop's SDXL throughput. Re-tested after a driver update (NVIDIA Studio 551.86) which improved SDXL throughput on the 4080 laptop more than on the 4090 laptop, narrowing the gap to ~92%. Chart updated; commentary about the trade-off rebalanced.
January 9, 2026 — Lenovo Legion Pro 7i Gen 9 RAM ceiling
Affected page: RTX 4080 vs RTX 4090 laptop 2026. Initially stated 64 GB DDR5 maximum. Lenovo confirmed (and we verified by installing) 96 GB and 128 GB SO-DIMM configurations work, though Lenovo does not officially support them. Page updated with the official 64 GB ceiling plus a note about community-tested higher capacities.

2025

December 12, 2025 — Initial RTX 50-series Super refresh expectation
Affected pages: multiple. Our late-2025 coverage anticipated an RTX 50-series Super refresh (5080 Super, 5070 Ti Super) in H1 2026. Per board-partner reporting in March 2026, the refresh has been shelved indefinitely due to the GDDR7 memory supply crunch. All affected pages now reflect the May 2026 industry reality that no new desktop RTX GPUs are expected to launch in 2026.
November 4, 2025 — ASUS ROG Zephyrus G14 RTX 4070 thermal
Affected page: Best portable AI laptops 2026. Our initial review of the 14" Zephyrus with RTX 4070 stated the sustained GPU TGP holds at 90 W during long workloads. Corrected after extended testing: sustained TGP drops to ~75 W after 25–30 minutes of constant load on the chassis we tested, with corresponding inference-throughput drop. The page now states sustained 75 W and uses that figure in throughput estimates.