When Silksong finally arrived in September 2025, six and a half years after its announcement, the conversation for the first fortnight was about one thing: it was too hard. Hornet drops rosaries on death, most late enemies hit for two masks, healing costs a full spool of silk, and the early bench prices felt like an insult after a difficult stretch. Team Cherry's mid-September patch cut those prices, softened a handful of early bosses and trimmed a few of the more punishing enemy placements. Then the game went quiet, in the way good games do, and people got on with playing it.

A year later, we've played it again from the start on Switch 2, finished it a second time on PC, and taken it through a couple of hundred hours between the three of us. This is our proper verdict, written with the benefit of not being angry at a bench.

The movement is the argument

Hollow Knight was a game about a small, heavy figure in a large, sad place. Silksong is about a fast, tall one in a place that is trying to kill her on purpose. Hornet's dash, her sprint, the way her needle strike carries momentum through the air: the whole kit is built for a player who wants to be in motion. The first game rewarded patience. This one rewards commitment, and punishes the half-step.

That's the design decision underneath the difficulty complaints. Hornet heals three masks at once by spending a full spool of silk, and she generates silk by hitting things. Standing back and waiting for a safe moment doesn't work because there isn't one. You get healthy by being aggressive. Once that clicks, usually somewhere around the second major boss, the game stops feeling unfair and starts feeling like a conversation you're finally holding up your end of.

Pharloom, upward

Hallownest went down. Pharloom goes up, and the game uses that constantly: you're climbing toward the Citadel, and every new region is visible from the last one, usually as a silhouette above you. The map is bigger than Hollow Knight's by a comfortable margin and denser with distinct places. The Marrow, Greymoor, Bilewater, the Choral Chambers: they don't just look different, they demand different tools. The crest system, where Hornet's basic moveset changes depending on which crest she's wearing, gives the exploration a second layer. The Reaper crest turns her into a slow heavy-hitter; the Wanderer crest is closer to Hollow Knight's Knight. Swapping between them at a bench is the closest the genre has come to letting you choose a class without breaking the map design.

Side quests are handled through a wish board in each settlement. Most are fetch or hunt tasks, and a few are tedious. But the way they gate new areas and NPCs behind quiet acts of decency means the towns feel populated in a way the first game's did not. Bellhart, once you've dealt with its problem, is one of the best hub towns in a game like this.

About that difficulty

The patches helped. Bench and tool prices are sane now, Moorwing and Sister Splinter no longer wall players in the first few hours, and the enemy that used to farm you at the entrance to the Marrow has been moved. What the patches didn't do is change the design, and they shouldn't have. Silksong's bosses are long, fast and mostly fair. Lace, the recurring rival, is a fencing match. The Cogwork Dancers are a lesson in reading two rhythms at once. The Last Judge is a wall, and it's meant to be.

Where the game is still legitimately harsh is in its economy of failure. Losing your rosaries to a death, then dying again on the way back to your cocoon, still stings a year on. Some of the platforming gauntlets in the third act ask for near-perfect input over long stretches with no checkpoint. If you bounced off Hollow Knight's Path of Pain, there are half a dozen Paths of Pain here, and a couple are on the critical path.

The third act

Without spoiling it: the thing that happens after what appears to be the ending recontextualises the entire map, and the game you play afterwards is meaningfully different from the one you'd been playing. It is the strongest late-game in the genre, and it's where the difficulty debate resolves itself. Everything the game asked of you in the first twenty hours was practice.

Performance

Locked at 60 on every platform we tried, including Switch 2 in handheld mode and Steam Deck OLED at its 90 Hz setting with a 60 fps cap. Load times are a second or two on Switch 2 and effectively nothing on PC. The one place we saw the frame-time graph twitch was a late boss with a lot of particle effects, on Steam Deck, and only for a moment. For a game this fast, that reliability is part of the design: you can trust every input.

Verdict

Silksong is not Hollow Knight 2. It is a faster, angrier, more demanding game built on the same bones, and a year of patches has sanded off the parts that were merely irritating without touching the parts that were hard on purpose. There is more here than in any other game of its kind: more map, more bosses, more systems, more secrets, and for A$29.95 that is an absurd amount of game. The lack of any difficulty options will keep some people out, and that's a real cost. For everyone else, this is the standard the genre will be measured against for a long time.

Disclosure: all three copies were bought at retail. No code was requested or offered. Tested on PC (Ryzen 7 7800X3D, RTX 4070 Super, 1440p 144 Hz with VRR), Switch 2 (handheld and docked) and Steam Deck OLED.

Every year, the Paintress paints a number on a monolith, and every person of that age turns to petals. The number goes down by one each time. When the game opens it's 33, and the expedition of thirty-three-year-olds leaving the city of Lumière to kill her know perfectly well what happened to Expedition 34 and every one before it. That is the whole premise, and the game trusts it. There's no map full of icons, no faction reputation, no crafting bench. There's a road, a party, and a countdown.

Turn-based, with your hands

Combat is where Sandfall's design bet pays off. You pick actions from a menu, as you would in any turn-based RPG. Then, on the enemy's turn, you dodge, parry or jump every incoming attack in real time. A dodge avoids damage. A parry, timed tighter, avoids it and builds toward a counter. A perfect chain of parries against a multi-hit attack ends with your whole party punishing the enemy at once.

The effect is that no fight is passive. You're reading animations the way you would in an action game, but with the breathing room to plan between rounds. Enemy movesets are long and deliberately deceptive: feints, delayed strikes, attacks that speed up on the third hit. Learning a boss means learning its timing, and the game is happy to let you die a few times while you do. Difficulty options exist and are honest: the lower setting widens parry windows rather than nerfing enemies, so the game stays itself.

Underneath sits a build system of Pictos (equipment that grants a passive) and Luminas (those passives, learned permanently after a few fights). By the middle of the game you're assembling parties around specific synergies, and by the end the numbers get very silly, in the way the best RPGs allow. Maelle's stance system and Monoco's shape-shifting wheel are the standouts; each character plays like a different sub-genre.

A world painted, then cut

The art direction is Belle Époque France run through a dream: a drowned city with a hovering Eiffel Tower, meadows of enormous flowers, stone giants frozen mid-fall. Regions are linear paths with side branches rather than open zones, and the game is better for it. Nothing is placed to fill space. The overworld map, a small-scale walkable world between areas, is the closest thing to a nostalgic flourish, and it's handled with real charm.

Lorien Testard's score is the best of its year and probably of several years. It moves between orchestral, French chanson, and something close to electronic rock without ever feeling like a playlist, and the boss themes with vocals are the kind of thing you'll find yourself humming a month later.

The story earns it

The cast is small and the writing is confident enough to sit in silence. Gustave, Maelle, Lune, Sciel and the two who join later are grieving people who make jokes, which is rarer in games than it should be. The performances, in particular from Jennifer English and Ben Starr, carry scenes that lesser games would have cut to a codex entry. The ending is a genuine choice with a genuine cost, and the game does not tell you which one it wants.

What holds it back

• Platforming. There are jumping sections in the later areas, and the character controller isn't built for them. They're short but they're clumsy.
• Camera and lock-on. In fights against large enemies the camera occasionally hides the wind-up you need to see. It cost us a few unfair deaths.
• Performance on PC. Unreal Engine 5 traversal stutter is present on our main bench, less so after the post-launch patches, and more so on the budget bench. On PS5 Pro, performance mode is soft but consistent. It's playable everywhere; it's not clean everywhere.

Verdict

Expedition 33 is the best RPG of its year and one of the most complete debut games we've played. It knows exactly how long it needs to be, it respects the player's time and intelligence, and its combat is a genuine idea rather than a hybrid. A studio of around thirty core staff made this. That should embarrass some much larger ones.

Disclosure: the PS5 code was supplied by the publisher, Kepler Interactive, with no conditions beyond the embargo date. The PC copy was bought at retail for performance testing after launch. Played on PS5 Pro (performance mode, VRR on) and PC (Ryzen 7 7800X3D, RTX 4070 Super, DLSS Quality at 1440p).

Most reviews you read were written after about eight hours, against an embargo, on hardware the publisher picked. That isn't a scandal. It's a schedule. Review code arrives a week or two before launch, the embargo lifts a few hours before the shops open, and the outlets that publish first get the traffic that pays for the next review. It works well enough for games that show you everything they've got by hour three. It is useless for games that don't, and the games that don't are usually the ones worth writing about.

Where games are decided

The back half is where a design either holds together or falls apart. Three examples from our own recent verdicts:

• Elden Ring Nightreign is thrilling for the first five runs with a squad. Its solo balance problem, the thing that decides whether we can recommend it to someone without two friends on voice chat, doesn't show up until you've reached the later Nightlords alone and realised the encounter scaling wasn't designed with you in mind.
• Monster Hunter Wilds runs acceptably in the early hunts. The performance problems compound in the endgame, where more monsters, more players and more particle effects stack in the same arena at once. An eight-hour review can honestly say it runs fine. A seventy-hour one can't.
• Clair Obscur: Expedition 33 is good for its first act. It is exceptional because of its third, and because of what its ending does with everything before it. The score is a verdict on the whole thing, and you can't have that verdict early.

Silksong's difficulty argument is the same shape. The launch-week reviews weren't wrong that the early hours were brutal. They were incomplete, because the third act reframes what those early hours were for. We got that piece wrong ourselves in our first impressions, which is why our full review is dated a year after launch and not a week.

What it costs us

Speed. Our review of a forty-hour RPG lands one to three weeks after everyone else's. We don't get the launch-day traffic, we don't get quoted on the box, and we don't get invited to things, which is fine, because we wouldn't go. Every review on this site was played on hardware we bought, in our own homes, at our own pace. Publisher codes are accepted, disclosed on the page, and never accompanied by travel, accommodation or a "preview event". The only condition we accept is an embargo date.

The rules, in full

1. Finish the game before scoring it. For games without an ending, play at least twenty hours across at least two weeks, so that the second week's opinion gets to argue with the first.
2. Name the hardware, the settings and the display mode. Say whether VRR was on. Say which servers the online parts were tested on and when.
3. Say where the copy came from. If a publisher supplied it, name the publisher.
4. Re-score only for a substantial patch, and leave the original score and date visible beside the new one.
5. Never show a review to a publisher before publication. Never accept a score range, a "suggested angle", or a request to hold a criticism.

None of this makes us right. It makes us slow, and it makes us accountable for the whole game rather than the part of it that fits in a press cycle. If you need a day-one verdict, there are good outlets for that. Read them. Then, before you spend A$110, read us.

The one idea

Frames per second is an average over a second. Frame time is how long each individual frame took to produce, in milliseconds. The two describe the same thing from opposite ends, but only one of them tells you how a game feels.

Target	Frame time	What it looks like
30	33.3 ms	Console "quality" modes. Fine if every frame is exactly 33.3 ms. Miserable if they alternate between 16 and 50.
40	25.0 ms	The 120 Hz middle ground. Evenly paced 40 feels far closer to 60 than to 30.
60	16.7 ms	The standard. A single 50 ms frame in a run of 16.7 ms frames is what your eye calls a stutter.
120	8.3 ms	Diminishing returns for most people, but the input latency drop is real.

Your eye is far more sensitive to variance than to level. Sixty evenly spaced frames feel smoother than an average of 75 that spikes to 40 ms once a second. That's why a game can report a perfectly good average in a benchmark and still feel awful: the average is hiding the spikes.

The numbers to trust

• Average fps. The least useful number on the overlay. Fine for comparing graphics cards, useless for judging how a game feels.
• 1% low. The average frame rate of the slowest 1% of frames. This is the single best number for "how bad are the bad moments". A 1% low within about 70% of your average is healthy. A 1% low half your average means the game hitches regularly.
• 0.1% low. The very worst frames. Shader compilation and asset streaming show up here. If this is terrible but the 1% low is fine, you're looking at occasional hitches rather than constant unevenness.
• The frame-time graph. A line that shows each frame's time. Flat is good, regardless of height. Spiky is bad, regardless of average. Once you've learned to read this line you'll never trust an fps counter on its own again.

What the shape of the graph tells you

• Sharp, isolated spikes the first time you see a new effect or enemy are shader compilation stutter, on PC only. Games that make you sit through a "compiling shaders" step at launch are doing you a favour. It usually settles after the first hour in an area.
• Regular spikes as you move through the world, especially crossing between regions, are traversal stutter: the engine loading the next chunk. Common in Unreal Engine titles. A faster SSD helps less than you'd expect, because the bottleneck is usually the CPU preparing the data, not reading it.
• Frame times climbing in towns, crowds or big battles while GPU usage drops below about 90% is a CPU bottleneck. Lowering resolution won't help. Lowering crowd density, draw distance, shadow-casting lights or ray tracing will.
• A sudden collapse in 1% lows after raising texture quality is video memory overflowing. The card is swapping textures over the bus. Drop textures one notch and the collapse disappears.
• A flat line at a level you didn't ask for, say a solid 60 when you wanted 90, is a cap: V-sync, a frame limiter, or the game's own ceiling.

Tools

• Steam's built-in overlay. Since 2025 the Steam performance monitor shows frame times, 1% lows and CPU/GPU load with no extra software. Settings → In Game → Performance Monitor. This is enough for most people.
• Intel PresentMon. Free, works with any GPU, and the reference tool for measuring frame pacing. Its "frame time" graph is the one to learn on.
• CapFrameX. Free, records sessions to a file so you can compare settings changes properly instead of by feel.
• MSI Afterburner + RivaTuner. The old standard. Still fine; also the easiest way to set a global frame cap.
• On console there's nothing built in. Enable VRR, pick the performance mode, and rely on reviews that report frame times rather than frame rates. Ours do.

Fixing it, in order

1. Turn on VRR if your screen supports it. Variable refresh rate lets the display show frames as they arrive instead of forcing them into a fixed grid, which is what turns unevenness into judder. Most TVs support VRR between 48 and 120 Hz. Below the bottom of that window, VRR stops helping, which is why a game dropping into the 40s on a TV still feels rough.
2. Cap the frame rate a few frames below your refresh rate, inside the VRR window: 117 on a 120 Hz screen, 141 on 144. This keeps you off V-sync's latency penalty. On PC, use Nvidia Reflex or AMD Anti-Lag where the game offers it, or a driver-level cap otherwise.
3. Pick a target you can hold, not the highest you can hit. A locked 40 on a 120 Hz screen feels better than an unlocked 55 that swings between 40 and 70. Consoles have started offering 40 fps modes for exactly this reason. Use them.
4. Lower the right setting. If GPU usage is pinned at 99%, lower resolution or use an upscaler. If GPU usage is below 90% while frame times climb, the CPU is the limit: lower crowds, draw distance and ray tracing. Our settings guide ranks these by cost.

Preset names lie. "Ultra" is a marketing tier, not a visual one, and on most games the gap between High and Ultra is invisible in motion while costing a quarter of your frame rate. The order below is the order we work through on every new game, on our main bench and on the 8 GB budget card, and it's rarely wrong.

1. The free ones: post-processing

Motion blur, chromatic aberration, film grain, lens flare, vignette: off. They cost almost nothing, so this isn't about performance. It's about clarity, and about upscalers: grain and blur are noise, and an upscaler has to reconstruct through noise. Turning them off makes DLSS and FSR look measurably cleaner. Per-object motion blur is a matter of taste; camera motion blur is not, and it's the one that makes people motion sick.

2. Anti-aliasing: use the upscaler, not MSAA

On a modern deferred renderer, MSAA and SSAA are brutally expensive and don't touch the shimmer on foliage and specular highlights that bothers people most. Use the upscaler instead, either at its Quality setting or in native-resolution mode (DLAA for Nvidia, FSR Native AA for AMD). DLSS 4's transformer model, released in early 2025, was a real step up in stability and works on every RTX card back to the 20-series. FSR 4 closed most of the remaining gap, but it needs a Radeon RX 9000-series card; older Radeons get FSR 3.1, which is fine at 4K and soft at 1080p. Intel's XeSS runs on anything and is a good fallback.

3. Shadows: Ultra to High

Shadow quality beyond High mostly raises the resolution of shadow maps you'll only notice within a few metres of the camera, while costing a chunk of both GPU and CPU. This is usually the single largest saving available for zero visible loss. Ray-traced shadows are a separate toggle and a separate discussion (below).

4. Volumetric fog and clouds: Ultra to Medium

Volumetrics are among the most expensive effects per pixel of visible difference. Medium looks identical in motion and different only in a side-by-side screenshot, which is not how you play games. If a game separates "cloud quality" from "fog quality", cloud quality is the one to lower first.

5. Ambient occlusion: keep it on, keep it cheap

Ambient occlusion matters visually. Without it, objects look like they're floating a centimetre above the floor. But the difference between a good screen-space method (SSAO, HBAO+, GTAO) and ray-traced AO is small, and the cost isn't. Keep AO on, pick the screen-space option.

6. Ray tracing, one feature at a time

Ray tracing is not one setting. Ranked by how much they change the image:

1. Global illumination (Lumen hardware mode, RTXGI, path tracing). The biggest visual change, especially indoors and at dusk, and the biggest cost. If you can afford one RT feature, this is it.
2. Reflections. Visible on wet streets, glass and water; invisible in a forest. Turn on for city games, off for everything else.
3. Shadows and ambient occlusion. Marginal gains. Last to enable, first to disable.

Path tracing, where a game offers it, is the whole pipeline replaced. On a 4070-class card it's playable at 1440p with DLSS Performance and frame generation. It's spectacular and it's optional; don't feel obliged.

7. The CPU-side settings

Draw distance, crowd density, foliage density and the number of shadow-casting lights are largely CPU costs. Lower them if your GPU is sitting below about 90% usage while frame times climb in cities. Lowering resolution in that situation does nothing; you're not GPU-bound. Our frame-time guide covers how to tell.

8. Textures: as high as your memory allows, no higher

Texture quality costs almost nothing in GPU time and everything if it overflows video memory. The symptom is a collapse in 1% lows rather than a lower average, so it's easy to misdiagnose. On an 8 GB card in 2025–26 releases, High is the ceiling; Ultra is for 12 GB and up. On 16 GB, max it and forget it.

9. Frame generation: only from a good base

DLSS and FSR frame generation insert an interpolated frame between two rendered ones. The catch is that the interpolated frame doesn't respond to your input, so latency stays where the base frame rate put it. Enable it to go from 60 to a smoother 100-plus on a high-refresh screen. Do not enable it to go from 30 to 60; you'll get 60 fps with 30 fps input feel and artefacts around the HUD. Multi-frame generation on RTX 50-series cards changes the numbers, not the rule.

10. Resolution scale: the last resort

At 4K, an upscaler working from 1440p (Quality mode) or even 1080p (Performance mode) looks excellent. At 1080p output, the upscaler is reconstructing from 720p or less, and it shows. If you're on a 1080p screen, do everything above first, and use the upscaler's Quality mode only when you must.

If HDR on your console looks grey and washed out, or blown-out with no detail in bright skies, it's almost never the TV's fault and almost never the game's. It's that two devices are each adjusting the picture for a display they can't see. The console guesses how bright your TV can go. The TV then re-compresses the picture the console already compressed. The fix is to give the console the right numbers and tell the TV to stop guessing.

1. Peak brightness, on the console

The console needs to know your TV's real peak brightness so it can map the game's highlights to it. Every platform has a calibration screen:

• PS5: Settings → Screen and Video → Video Output → Adjust HDR. Three screens. On each, press until the symbol is just barely visible, not until it's comfortable. That's the point where the TV stops being able to show anything brighter.
• Xbox Series: Settings → General → TV & display options → Calibrate HDR for games. Same idea, with a few more screens.
• Windows: the free Windows HDR Calibration app from the Microsoft Store. Games that use the system-level HDR profile will pick it up.

Do this once, in a darkened room, with the TV already in Game mode (see below). If you change TVs or picture modes, do it again. Typical results: 600–800 nits for a mid-range OLED, 1,000–1,500 for a high-end OLED or a good Mini LED, 400–600 for an ordinary LCD.

2. Paper white, in the game

Many games expose a second slider, usually called paper white, UI brightness, or mid-tone brightness. It sets how bright a white interface element or an overcast sky should be, in nits. This is the one people get wrong by dragging it up.

• 200 nits is a sensible starting point for a normally lit lounge room.
• 100 nits is the SDR reference and looks dim next to a lamp.
• Above 300 starts to flatten the picture: if your "white" is already near the TV's peak, there's nowhere left for highlights to go, and the HDR you paid for disappears.

If a game only has one "brightness" slider in HDR, that's usually paper white in disguise. Set it so that a white menu is comfortable rather than dazzling.

3. HGiG, on the TV

This is the setting that fixes the fight. HGiG is a guideline agreed between the console makers and the TV makers: when it's on, the TV performs no additional tone mapping and trusts the numbers from step one exactly. Without it, the TV applies its own curve on top of the console's, and you get the double tone mapping that makes everything grey.

• LG (2019 onward): Picture → Advanced Settings → Brightness → Dynamic Tone Mapping → HGiG.
• Samsung (2022 onward): in Game Mode settings, HDR Tone Mapping → Static. On some models this is labelled "Game HDR" with an HGiG option.
• Sony (2021 onward): Picture → Brightness → Tone mapping → Off, or "Gradation preferred" on models without an Off option. Sony's 2025 sets have an explicit HGiG toggle in the game menu.
• Others: look for "HGiG", "tone mapping off" or "static tone mapping" in the game picture mode. If none exists, pick the option that sounds least clever.

Once HGiG is on, go back and redo the console calibration in step one. The numbers will land in a different place, and that's expected.

The supporting settings

• Game mode and ALLM. Game mode drops input latency by turning off picture processing. ALLM lets the console switch the TV into it automatically. Both on. The picture you tuned above is the Game mode picture; other modes ignore it.
• Enhanced HDMI input. The HDMI port needs to be set to its full-bandwidth mode for 4K HDR at 120 Hz. Samsung calls it Input Signal Plus, LG calls it HDMI Deep Colour, Sony calls it Enhanced format. If your console reports it can't do 4K120 or VRR, this is the reason nine times out of ten.
• RGB range. Leave the console on Automatic. Forcing Full on a TV expecting Limited crushes the shadows.
• Auto HDR (Xbox and Windows) upgrades SDR games to HDR on the fly. It's decent, and it occasionally lifts blacks. Turn it off per-game if a dark game looks milky.

Inner deadzone

The inner deadzone is a circle around the stick's centre position where input is ignored. It exists to hide the fact that a resting stick never sits at exactly zero. Set too large, and the first few millimetres of every push do nothing, so small aim corrections feel sluggish and you overshoot. Set too small, and a slightly worn stick makes your view creep on its own.

Start at 5–8% and only raise it if you see drift. Where a game offers the choice, pick a radial deadzone over an axial (cross-shaped) one; the cross shape makes diagonals feel sticky because each axis snaps to zero independently.

Outer deadzone

The outer deadzone is the threshold at which the stick counts as fully pushed. Physical sticks don't trace a perfect circle at the edge, and a stick that only reads 100% at one exact angle means you rarely hit maximum turn speed. If turning feels slower than the sensitivity number suggests, lower the outer deadzone (some games call it "outer threshold" or "max input threshold") until full deflection feels like full speed in every direction.

Response curve

The curve maps how far you push to how fast you turn. Three common shapes:

• Linear. Half push, half speed. Most predictable. Best if you have a light touch and a small deadzone.
• Exponential (often "precision" or "classic"). Slow near the centre, fast at the edge. Gives you fine control for small adjustments and quick turns at full push. Best for most people in shooters.
• Dynamic, where offered, ramps speed based on how long you've held the stick. Some players swear by it; we find it makes tracking a moving target less predictable.

Try linear with a small deadzone first. If your fine adjustments overshoot the target, switch to exponential and lower your sensitivity a notch.

Aim acceleration

Separate from the curve, many console shooters ramp turn speed up the longer you hold the stick at the edge. This lets a low sensitivity still turn fast. It also means a full-speed flick lands somewhere different depending on how long you held it. If you flick to targets, turn acceleration off or low and raise sensitivity. If you track targets with small movements, keep it.

Drift and what to buy

Most sticks use potentiometers, which physically wear and eventually report movement at rest. That's drift. The fixes, in order of cost:

1. Raise the inner deadzone a few percent. Free, and buys months.
2. On PC, calibrate the stick in Steam (Settings → Controller → Calibration). Steam Input can apply a deadzone at the driver level to any game, including ones with no stick settings at all.
3. Replace the stick module. The DualSense Edge has user-swappable modules; the standard DualSense, Joy-Con 2 and Xbox controllers don't.
4. Buy a controller with Hall effect or TMR sticks, which sense position magnetically and don't wear. Most third-party pads from 8BitDo, GameSir and Turtle Beach use them now. The first-party controllers from Sony, Microsoft and Nintendo still do not.

Gyro: the setting worth learning

Every DualSense, Joy-Con, Switch 2 Pro Controller and Steam Deck has a gyroscope, and using it for fine aim is the biggest single accuracy improvement available on a controller. The standard setup is stick for coarse turning, gyro for the last few degrees, with gyro active only while aiming down sights. It's built into most Nintendo shooters, most Sony first-party games, and can be added to any PC game through Steam Input. It takes about two evenings to feel natural, and after that stick-only aiming feels like typing with mittens.

Connection

Wired, or the console's own wireless protocol, beats generic Bluetooth. On PC, an Xbox controller through the official wireless adapter or a DualSense over USB will both feel noticeably tighter than either over Bluetooth. If the pad feels vague on PC and sharp on console, the connection is the reason.

Games have got dramatically better at this in the last five years. The problem now is that the good options are buried three menus deep, in a section most people never open because it isn't labelled for them. Open it. Here's what to look for, in the order we check.

1. Subtitle size and background. Turn subtitles on before the first cutscene, then make them large enough to read from the couch. Most published guidelines land around 46 pixels tall on a 1080p image, which is bigger than almost any default. Enable a background box and speaker names if offered. Directional indicators for off-screen speakers are a bonus.
2. Hold-to-toggle. Sprint, aim, crouch, and interact are often "hold" by default. Switching them to toggle means less strain on your hands over a long session and fewer accidental releases in a fight. This is the single setting we change in every game.
3. Camera motion. Screen shake off or low, camera bob off, motion blur off. If a game makes you queasy, raise the field of view (90–100 in first person is a good range) before you give up on it.
4. Colour and contrast. Colourblind modes are standard now; check whether the game also offers a high-contrast mode that outlines enemies, allies and interactables in flat colours. Sony's first-party games and a growing number of others do, and it's useful for anyone in a busy scene.
5. Remapping. Check whether the game allows full remapping, not just presets. If it doesn't, the system can: PS5's Custom Button Assignments, the Xbox Accessories app, Switch's button mapping under Controllers and Sensors, and Steam Input on PC all remap at a level the game can't see.
6. Aim assist and lock-on strength. Usually separate sliders for magnetism (the reticle sticking to a target) and slowdown (sensitivity dropping near a target). Neither is cheating. Set them where hitting things is satisfying rather than automatic.
7. Difficulty and assists, separately. The best games split these: a difficulty preset, and then individual toggles for things like damage taken, timing windows, or invincibility. Celeste's Assist Mode is still the model. Using them is not a lesser way to play; the developers shipped them because they wanted you to finish.
8. Phobia toggles. Arachnophobia modes exist in Grounded, Hogwarts Legacy, Satisfactory and others, replacing spiders with something abstract. If you know a category of thing you can't look at, check the menu before you meet it.
9. Screen reader and menu narration. PS5 and Xbox both have system-level screen readers, and an increasing number of games narrate their own menus. If you or someone you play with relies on this, it's worth knowing which games support it before buying: The Last of Us Part II, Forza Horizon 5, Diablo IV and the recent Call of Duty games are the reference points.

None of this is about playing on easy. It's about removing friction that the developers themselves decided was optional. The menu is there. Use it.

Nintendo Switch 2 Buy for the library

A year after its June 2025 launch, the Switch 2 has become what the original was in its second year: the console with the games. Mario Kart World, Donkey Kong Bananza and the steady conversion of third-party releases that skipped the first Switch have made the machine feel necessary rather than merely new. The hardware itself is competent rather than exciting, and that's a fair trade at this price.

Things worth knowing that the box doesn't say:

• The 7.9-inch screen is a good LCD with a 120 Hz refresh rate and HDR support, but it isn't an OLED. Blacks are grey in a dark room. Next to a Steam Deck OLED it's obvious.
• VRR works in handheld mode only. Docked, you get fixed 60 or 120 Hz output, so a game that dips will judder on the TV in a way it doesn't on the handheld screen.
• Storage is 256 GB, and only microSD Express cards work for expansion. Your old microSD cards are for screenshots only.
• Game-Key Cards, the cartridges that download the actual game, are common for third-party releases. Check the box if you care about lending or resale.
• Joy-Con 2 attach magnetically, the sticks are larger, and mouse mode works better than it sounds. The sticks are still potentiometers, so drift remains a question of time.

Spec	Detail
Screen	7.9" 1920×1080 LCD, up to 120 Hz, HDR
Docked output	Up to 4K at 60 Hz, or 1440p/1080p at 120 Hz
Storage	256 GB, microSD Express expansion
Battery	2 to 6.5 hours, heavily game-dependent
Launch price (AU)	A$699.95 console; A$769.95 with Mario Kart World

Steam Deck OLED Still the value pick

The OLED revision arrived in late 2023 and is, three years later, still the handheld PC we reach for most. The reasons haven't changed: the screen is the best on any handheld, the battery lasts, the controls are the most complete of anything in the category (trackpads, gyro, four back buttons), and SteamOS gets out of the way. Suspend and resume works like a console. Nothing running Windows does that as well, even now.

The honest caveats: the chip was mid-range in 2022, and 2025–26 releases increasingly need low settings and 30 fps caps to hold steady. Unreal Engine 5 games in particular struggle. Valve has said a successor waits on a real generational leap, and the company's 2025 announcements were a living-room Steam Machine and a VR headset rather than a new Deck. If you mostly play indie games, older AAA, and anything with a "Verified" badge, buy the OLED without hesitation. If you want to run this year's big releases well, look at the Ally X below and accept the trade-offs.

Spec	Detail
Screen	7.4" 1280×800 HDR OLED, 90 Hz
Storage	512 GB or 1 TB NVMe, microSD expansion
Battery	50 Wh; roughly 3 h in heavy games, 8–12 h in light ones
OS	SteamOS (Arch Linux); Windows possible but not recommended

ROG Xbox Ally X Great, if the price works

Asus and Microsoft's collaboration, launched in October 2025, is the first Windows handheld where you don't spend your evening fighting Windows. The full-screen Xbox interface boots straight into a controller-driven launcher that aggregates Steam, Game Pass, Epic and the rest, and it suspends background Windows processes to claw back memory and battery. It's not as seamless as SteamOS, and every so often a Windows dialog pokes through, but it's finally usable from the couch.

The hardware is the strongest in its class: the Z2 Extreme has meaningfully more headroom than the Deck's chip, 24 GB of memory means no VRAM juggling, and the 80 Wh battery is the largest in a handheld. The costs are weight (715 g, noticeably front-heavy), an LCD rather than OLED, and the price. The non-X Ally with the Z2 A chip is much cheaper but lands close to Deck performance with a worse screen and worse software, and we'd skip it.

Spec	Detail
Chip	AMD Ryzen AI Z2 Extreme, RDNA 3.5 graphics
Memory / storage	24 GB LPDDR5X, 1 TB NVMe
Screen	7" 1920×1080 LCD, 120 Hz, VRR
Battery	80 Wh
Weight	715 g

What we test on

For completeness, the rest of the bench: PS5 Pro (2024) on an LG C4 OLED with HGiG and VRR on; Xbox Series X on the same TV; a main PC with a Ryzen 7 7800X3D, RTX 4070 Super and 32 GB of memory on a 1440p 144 Hz monitor; and a budget PC with a Ryzen 5 5600, an 8 GB Radeon RX 7600 and 16 GB of memory on a 1080p 144 Hz monitor. The budget bench is there so the advice in our guides works for the hardware most people actually own.

Disclosure: all hardware on this page was bought at retail. No review units were requested or accepted. Prices are Australian launch prices and may have changed.