Featured image: Photo via Pexels
Sony has not said a single official word about a handheld successor to the PS5, but that has not stopped the leaks from getting specific. On September 20, Russian hardware outlet GameGPU published a detailed breakdown of a device it calls Project Canis, nicknaming it the “PSP 3,” and the numbers inside are strange enough to explain on their own why the story spread so fast. The headline claim: games would render internally at just 540p, then get stretched up to the device’s native 1080p screen by a new upscaler called PSSR 3.0.
That is a low bar for a console launching in 2027. It is also, according to the leak, exactly the point.
The short version
- What leaked: Specs for Project Canis, a rumored PlayStation handheld, via hardware outlet GameGPU
- The chip: A custom 3nm AMD SoC with six Zen 6 cores and 16 RDNA 5 compute units, paired with 24GB of LPDDR5X memory
- The catch: Native rendering resolution is only 540p, upscaled to 1080p by PSSR 3.0
- The payoff: Effective performance is claimed at 10 to 11 TFLOPS handheld and roughly 14 TFLOPS docked, closing in on the PS5 Pro’s 16.7 TFLOPS
- Timeline: GameGPU points to a late 2027 launch, which may or may not line up with the mainline PS6
None of this is confirmed by Sony, and it is worth sitting with that caveat before going further. But the leak lines up closely enough with earlier reporting from separate sources that it is worth taking seriously as a picture of where Sony’s engineers are pointed, even if some of the exact numbers shift before anything ships.
What GameGPU actually claims
Strip away the framing and the leak describes a fairly conventional handheld APU for 2027, built around AMD’s next mobile architecture rather than anything exotic. The custom Canis processor reportedly uses six Zen 6 CPU cores, split between four standard Zen 6c cores and two power sipping Zen 6 LP cores, a configuration built for battery life first and burst performance second. Memory is 24GB of shared LPDDR5X on a 192-bit bus, and the whole chip is designed to operate inside a strict 15 to 25 watt envelope, roughly in line with what a Steam Deck or ROG Ally already targets.
The graphics side is where things get more interesting. Sony’s leaked GPU uses 16 compute units built on AMD’s RDNA 5 architecture with Dual-Issue FP32 support, clocked at 1.2GHz in handheld mode for about 4.91 TFLOPS, and up to 1.65GHz when docked for roughly 6.75 TFLOPS. On paper, that is modest, well behind even a base PS5. On its own, none of this would be a story.
How the specs stack up against what already exists
Numbers on their own do not mean much without something to measure them against. Here is how the leaked Project Canis compares with the PS5 Pro, Nintendo’s current handheld king, and the cheapest current-gen console on the market.
| Device | GPU compute | Raw TFLOPS | Native resolution target |
|---|---|---|---|
| Project Canis (handheld) | 16 RDNA 5 CUs @ 1.2GHz | 4.91 | 540p, upscaled to 1080p |
| Project Canis (docked) | 16 RDNA 5 CUs @ 1.65GHz | 6.75 | Upscaled output, target unclear |
| PlayStation 5 Pro | 60 CUs, custom RDNA 2/3 hybrid | 16.7 | 1080p to 1440p, upscaled to 4K |
| Nintendo Switch 2 (docked) | Nvidia Ampere, 1536 CUDA cores | 3.07 | Varies by title, often upscaled to 4K |
| Xbox Series S | 20 CUs, RDNA 2 | 4.0 | 1080p, upscaled to 1440p |
Read the raw TFLOPS column and Project Canis looks unremarkable, sitting close to an Xbox Series S. That is reportedly the whole point of routing everything through PSSR 3.0 before it reaches the screen.
The 540p bet, and why Sony thinks it works
According to GameGPU, once PSSR 3.0 is factored in, Project Canis effectively performs like a 10 to 11 TFLOPS device in handheld mode and close to 14 TFLOPS when docked, within striking distance of the PS5 Pro’s 16.7 TFLOPS despite running a fraction of the compute units at a fraction of the power draw. The trick is an aggressive upscale, from a 540p internal render up to the display’s native 1080p, combined with what Sony architect Mark Cerny has previously teased as proper frame generation support finally arriving on PlayStation hardware.
The leak is not appearing in isolation. Hardware leaker Kepler_L2 has separately claimed that Project Canis beats the Xbox Series S in rasterization and pulls well ahead of it in ray tracing and path tracing workloads, and has gone as far as suggesting PSSR 3.0’s image quality could surpass Nvidia’s DLSS 4.5. That is a bold claim for an upscaler that has not shipped on any device yet, and it is exactly the kind of claim that deserves skepticism until independent hands-on testing exists. Sony’s own history with checkerboard rendering on the PS4 Pro and PSSR on the PS5 Pro suggests the company is comfortable leaning hard on reconstruction techniques rather than brute silicon, and this would simply be the most aggressive version of that strategy yet.
Reasons to keep expectations in check
This is a single outlet’s report on leaked internal documentation, not a Sony announcement, and specs this early in a console’s development routinely change before launch. GameGPU’s own framing admits the mainline PS6 could slip well past 2027, which would leave Project Canis’ release date, and possibly its final specs, hostage to a console it may or may not need to match. Treat the TFLOPS math as a best case scenario Sony’s marketing would love to hit, not a guaranteed outcome.
Why a Sony hybrid handheld makes sense right now
Whatever the exact numbers turn out to be, the timing of a Sony push into hybrid handhelds is not hard to explain. Nintendo’s Switch 2 has sold briskly by pairing modest docked power with software that scales gracefully, Valve’s Steam Deck normalized the idea of a pocketable PC gamers will happily pay a premium for, and Microsoft has already priced its own Xbox handheld for the same market. Sony sitting out that category a second time, after the PS Vita’s commercial struggles, would mean ceding an entire hardware segment to competitors while its own back catalog of PS5 exclusives sits unplayed on the go.
PSSR 3.0 is also not appearing out of nowhere. It follows the same reconstruction lineage as the technology geekblog covered when Nvidia’s rival approach arrived, where DLSS 5 pushed upscaling toward something closer to a real-time generative filter than a simple sharpening pass. If Sony’s next-generation upscaler is genuinely competitive with that approach, betting a handheld’s entire value proposition on it stops looking reckless and starts looking like the only realistic way to fit meaningful graphical fidelity inside a 15 to 25 watt battery budget.
When could this actually show up
GameGPU’s report points to a late 2027 launch window for Project Canis, but the outlet is careful to note that this timeline may not be tied to the mainline PlayStation 6, which some rumors suggest could land considerably later than 2027. That would not be unprecedented. Sony has shipped handheld and home hardware on staggered schedules before, and a PSSR 3.0 debut on a cheaper handheld ahead of the flagship console would give the company a full product cycle to iron out the upscaler before it needs to carry a $500 console’s marketing pitch.
Bottom line
Take the specific TFLOPS figures with the grain of salt any pre-announcement leak deserves, but the shape of the story is coherent enough to matter. Sony appears to be designing a handheld that treats raw compute as a secondary concern and treats its upscaler as the actual product, a bet that a 540p image reconstructed well can beat a higher native resolution rendered on weaker reconstruction. Whether PSSR 3.0 can deliver on that promise will not be knowable until someone outside Sony gets to hold a working unit, but the direction the leak describes tracks with everything else the handheld market has been doing for the past three years. The raw numbers are modest. The bet is that nobody will notice once the upscaler is done with them.
