A man with a gun stands above the crowd at GDC, the barrel levelled at the words "Next Gen Raytracing Performance & Capabilities." Alas, he is not here to shoot the technojargon out of existence, nor to ballistically punish the unforgivably inappropriate use of title case. He himself is a ray-traced gamething, here to illustrate at least three of the many hardware-based tricks and tools that Microsoft are hyping up for Project Helix: the next, apparently PC game-running Xbox console.
Xbox devs! Here’s Microsoft’s pitch for Project Helix straight outta GDC:
God, just... the state of that slide. I know it’s a GDC talk and thus the intended audience is comprised of qualified gamesmiths, but despite certain efforts, most developers are still human beings. It may have profited Project Helix if these tech upgrades were communicated in something resembling a human language. Luckily, I speak fluent Marketing, so here’s what each line of the Project Helix wordblast actually means – and what, in practical terms, it’s promising to do.
"Powered by Custom AMD SoC"
A SoC, or System on a Chip, is essentially a very compact computer. It squeezes the system’s CPU, graphics processor, memory, and input/output doodads all onto a single chip, rather than spreading them out across the motherboard like most desktop PCs do. AMD, who have experience making similarly combined CPU+GPU chips (themselves called APUs, or Accelerated Processing Units) for previous Xboxes, the Steam Deck, and the upcoming Steam Machine, have been called in once again to make a super-special SoC for Project Helix.
"Codesigned for Next Generation of DirectX"
"Codesigned" is the most Bachelor of Science way of saying "designed for" that it’s possible to conceive. DirectX, as you probably already know if you’re reading a PC gaming site, is software that allows your games to talk to your PC’s hardware. "Run me", they’ll say. "Okay we’ll try", the hardware will say back. Very wholesome.
Newer versions, or generations, of DirectX tend to unlock general performance optimisations, as well as more modern and advanced features like ray tracing. We’ve been on DirectX version 12 for over a decade now, so to be fair to Microsoft, compatibility with the next one is probably something you’d want. And if you don’t know what ray tracing is...
"Next Gen Raytracing Performance & Capabilities"
Ray tracing (it has a space, sorry) has been around for a while now, but is probably still best known as "that setting wot fucks my framerates". What it actually does is make games’ lighting and shadow effects more convincing and/or reactive by simulating, in real time, how light rays would realistically bounce around the scene. The results are often prettier, albeit much heavier on your graphics card than ‘rasterised’ (i.e. non-ray traced) effects.
It’s not entirely clear "next gen ray tracing" means, unless it’s just the same concept but executed better. Or it’s a bit of sneaky wordplay that simply amounts to saying that because Project Helix is a next-gen Xbox, its ray tracing capabilities are inherently next-gen. Which, in turn, means it’s probably about on par with a decent 2026-spec PC.
"GPU Directed Work Graph Execution"
Urgh, I nodded off just typing that. Work Graphs are a DirectX feature that essentially offload previously CPU-based rendering work to the GPU, where it can be dealt with faster, removing the CPU as a bottleneck and generally making the whole process more efficient. Imagine a restaurant where the maître d' is finally told "Y’know what mate, it probably doesn’t make sense for you to be chopping potatoes, let’s have the chef do it himself", thereby producing an increase in Dauphinoise-per-second. Work Graphs were added to DirectX for PCs in 2024, so this line is just saying that the next Xbox will get the feature too.
"AMD FSR Next"
FSR Next (or FSR Diamond, as it’s called elsewhere) is AMD’s collection of upscaling and frame generation toys, and seemingly the successor to FSR Redstone. FSR, which stands for FidelityFX Super Resolution, primarily seeks to boost game performance by having each frame rendered at a lower resolution (thus reducing PC load) and then digitally redrawing them to match your monitor’s actual resolution. Successive FSR versions have each improved post-upscaling image quality, to the point where Redstone is hard to tell apart from the historically superior Nvidia DLSS.
According to AMD computing and graphics exec Jack Huynh, FSR Diamond is being developed with Project Helix in mind, though that doesn’t necessarily mean it will look nicer and run smoother than Redstone does on conventional PCs – just that it’s the first Xbox-bespoke version of FSR, previous versions only having console support worked in later.
"Built for Next Generation of Neural Rendering"
Credit for not using "codesigned" again, I suppose. Neural rendering is the practice of combining bread-and-butter GPU rendering techniques with more AI-tinged methods. Microsoft are obviously bigging up the modernity of the latter on Project Helix, but these will surely be based on neural rendering tricks we’re already familiar with. FSR Redstone, for instance, uses forms of neural rendering: its upscaling component is based on machine-learned algorithms, presumably from an AMD supercomputer that’s spent the past few years looking at nothing but up-rezzed images.
"Next Generation ML Upscaling"
Whoops, I sort of defined this in the previous entry. But yeah, “ML" is just "machine learning", so FSR Diamond’s upscaler will have been taught to upscale extra-nicely by a purpose-built computer. Again, much like DLSS has been since its inception.
"New ML Multi Frame Generation"
Frame generation, as variously described and complained about in previous RPS articles, is the practice of a GPU creating new frames outside of its usual rendering pipeline, using data from surrounding frames to approximate the expected image. Generated frames are slotted in-between the 'real' ones, producing the impression of a faster, smoother framerate, while the addition of machine learning can make the system generally smarter about how it should draw the image and deal with potential motion. Multi frame generation means the system can produce several ‘fake’ frames for every real one.
One twist is that the PC editions of FSR, including Redstone, have never attempted multi frame generation before – despite competitors Nvidia having previously introduced it with DLSS 4. Be wary, though, of console/PC hybrids bearing gifts of AI-swollen framerates. Frame gen can work nicely in games that are already running well, but in those that are barely pushing 30fps through standard rendering techniques, it looks crap and adds an unpalatable amount of input lag.
"Next Gen Ray Regeneration for RT and Path Tracing"
Starting at the back, path tracing can be thought of as souped-up ray tracing: whereas most games’ ray tracing puts a cap on the number of simulated rays it can follow at a time, path tracing opens the taps, allowing multiple rays to share a single pixel of scene space and letting them bounce around randomly until they meet a light source, thus creating the final effect. It’s potentially more realistic-looking, but definitely more hardware-intensive. Current-gen consoles have never attempted it, so for better or worse, Project Helix will be the first.
Ray regeneration, meanwhile, is another neural rendering move, whereby GPUs do some additional AI-powered number crunching to work out how and where the simulated rays should bounce around. Nvidia call their version Ray Reconstruction, dontcha know. At its best, ray regeneration can remove visual errors and even slightly reduce the performance penalty for enabling path tracing in the first place. As with "Next Gen Raytracing Performance & Capabilities", it’s not clear what makes improved ray regeneration "next gen" as opposed to simply "better than the previous version"; expect implementation and performance to continue varying on a game-by-game basis, as is the current case on PC.
"Deep Texture Compression"
I think a Thai lady once did this to me by elbowing my calf muscles into dust. Alternatively, it sounds like merely a machine-learning-aided version of texture compression techniques that games have used for ages: in short, the images that make up in-game textures are shrunk down to reduce RAM usage. This might, therefore, help keep performance up in max-quality games with very high-rez textures, though the underlying concept is nothing new.
"Neural Texture Compression"
See above.
"DirectStorage + Zstd"
DirectStorage is another familiar name, and unlike most of the jargon here, was confusing Xbox owners long before baffling the PC crowd. It’s all about cutting load times: typically, accessing game data involves touring most of the system, the data first moving from the SSD or hard drive to the system RAM, then to the CPU for decompression, then back to the RAM, then onto the GPU for rendering. DirectStorage cuts out the CPU from the process, allowing the data to flow straight from the RAM to the GPU, where it can be decompressed even faster than on the CPU. Between this and Work Graphs, then, we know that our CPU are in fact feckless losers.
Zstd, or Zstandard, is a compression algorithm developed by Facebook (of all people). It’s just been added to DirectStorage recently, and is supposedly a particularly efficient compressor of game assets, so may help cut load times and help prevent texture pop-in.
While I don’t doubt that Microsoft are making Project Helix – and dumping slides of tech-blah upon unsuspecting developers – for the benefit of their console business far more than their PC one, I look at these features and only see the ongoing PC-ification of the Xbox brand. No wonder it’s playing our games, when it’s borrowing so much of our tech. FSR, ray tracing, path tracing, ray regeneration/reconstruction, AI upscaling, frame gen, even the Zstd that it’s contracted, they all began on PC. And you don’t need a flair for corporate languages to understand the culturally appropriate response: "You're", followed by "welcome".

1 hour ago
4








English (US) ·