NVIDIA's GeForce On stream goes live today at 8 AM PT, three hours ahead of Gamescom Opening Night Live and a day before the Cologne show floor opens, and the company has said exactly nothing about what it is announcing. On the other side of the stack, AMD's Zen 6 kernel work is running well ahead of its silicon.
A confirmed booth, an unconfirmed announcement
The broadcast premieres on NVIDIA's YouTube and Twitch channels and covers, in the company's own wording, the latest GeForce features and upcoming PC games. That is as specific as it has been willing to get. NVIDIA's own Gamescom page is far more forthcoming about the show floor, naming the RTX demo lineup: Aniimo, CONTROL Resonant, Gears of War: E-Day, Marvel Rivals, METRO 2039, Silver Palace, The Witcher 3: Wild Hunt Songs of the Past, Tides of Annihilation, Where Winds Meet, and Zenless Zone Zero, with ASUS, GIGABYTE, HP, PNY, and ZOTAC running RTX 50 hardware in adjacent halls.
The obvious candidate for the stream is DLSS 4.5 Ray Reconstruction, which VideoCardz expects to land alongside it. That pairing is inference from a release window, not an NVIDIA statement, so treat the timing as reported rather than confirmed. What NVIDIA has committed to is the month: its Ray Reconstruction announcement puts the release in August, running on every GeForce RTX generation, with 27 titles supporting it at launch and an override path through the NVIDIA app for the rest.
The mechanism matters more than the schedule. Ray Reconstruction replaces a game's hand-tuned denoiser and folds denoising into the same model that handles super resolution, so a ray-traced or path-traced scene is reconstructed once instead of being denoised and then upscaled. The second-generation transformer NVIDIA describes carries roughly 20 percent more parameters at similar performance and about 35 percent more compute than the model it replaces. Those are vendor figures on vendor terms. What a reader actually feels is ghosting behind moving light sources and boiling in reflections, and that is settled by frame comparisons in the games people play, not by a parameter count.
Nothing on the GeForce side forces a GPU decision this week. Ray Reconstruction runs on every RTX generation, which makes it an argument for keeping a card rather than replacing one, and the honest way to judge it is a side-by-side in a path-traced title once it ships.
AMD posts Zen 6 client patches for Linux, months before the chips exist
An AMD engineer posted patches to the Linux kernel mailing list in late July extending the HSMP (Host System Management Port) driver, previously reserved for EPYC server chips, to Zen 6 client silicon: the Medusa mobile APUs and the Olympic Ridge desktop parts. Phoronix covered the patch set, which defines a separate client message interface the patches themselves describe against Ryzen Master, AMD's Windows-only overclocking and monitoring tool. The message IDs cover setting the TjMax threshold, enabling or disabling overclocking, per-core OC frequency and OC voltage, and dynamic FCLK control, the same categories Ryzen Master exposes on Windows today.
None of it does anything yet. The patches are freshly posted, not merged, and no user-space tool has been announced to put a Linux desktop in front of these controls even once they land. It is also a separate effort from the Zen 5/Zen 6 identification fix that shipped in Linux 7.2-rc6 earlier this month: that patch corrected which CPU model IDs the kernel reads as Zen 5 versus Zen 6, while this one is new client-specific plumbing for chips that have not launched. Taken together they say AMD's kernel work on Zen 6 is running well ahead of the silicon, which is normal for a platform this far out, but it makes Linux-side overclocking support a fact about groundwork right now, not a feature anyone can use.
For Zen 6, the patches are worth knowing about as a preview of what Ryzen Master parity on Linux might eventually look like, not a reason to expect it soon.