How to choose a desktop CPU

A CPU purchase turns on five questions: what socket it needs, how many cores it has and how they're split, how much cache backs them, what it actually draws under load against what the label promises, and whether it needs a graphics card to show a picture at all. Answer those in order and a spec sheet stops being a wall of numbers.

Socket and generation: what a chip fits into

A socket is the physical and electrical interface between a chip and a motherboard. It is a compatibility question, not a performance one — a chip either fits the board in front of you or it does not, and no amount of budget changes that. A socket also typically outlives a single generation of chips: a new architecture ships on the same pins for a few years before the platform moves on, which is why "which socket" and "which generation" are two different questions that get asked together.

AM5 alone already spans two Ryzen generations — Zen 4 and Zen 5 chips both use it — so a socket outliving one generation of chips is normal, not a warning sign. The last row is different in kind from every row above it: BGA means the chip is soldered directly to the board rather than seated in a socket at all, which is how every laptop and other mobile part in the catalog is mounted. Every laptop chip in the catalog carries the BGA value; every desktop chip carries a socketed one. A soldered chip is never a separate purchase — it comes with the machine it's built into.

Cores, threads, and the Performance/Efficiency split

Every chip's page leads with two numbers: total cores and total threads. Cores are physical execution units; threads are how many instruction streams the chip can juggle across them, and on most chips that's simply twice the core count through simultaneous multithreading. It isn't automatic everywhere — on Intel's newest Arrow Lake chips, Hyper-Threading is gone entirely, so thread count matches core count instead of doubling it.

Total cores is only half the picture on Intel's hybrid desktop chips. They split into large Performance cores, built for single-thread speed, and smaller Efficiency cores, built to add throughput without adding much power draw or die area. A chip described as having 8 Performance and 12 Efficiency cores is not the same shape as one with 20 cores of a single type, even though both add up to 20. AMD's current desktop lineup uses one core type throughout, so the split stays empty on every AMD desktop chip in the catalog; it only appears on Intel's hybrid desktop parts and on AMD's own hybrid mobile chips, which mix full-size Zen 5 cores with smaller, efficiency-focused Zen 5c cores the same way.

Cache, and the X3D story

L3 cache is the last, largest pool of on-chip memory before a request has to go out to system RAM, and it's one of the figures recorded for every model. What makes it worth a section of its own is AMD's 3D V-Cache line, sold under the X3D name: an extra slab of cache die stacked onto or beneath the compute die, which enlarges the pool the cores can draw on without touching the core count, the clock speed, or (in most cases) the power envelope. A 16-core, 170-watt chip and its X3D sibling can share the same core count, the same boost clock and the same power envelope, and differ only in how much L3 cache sits on the die.

That extra cache pays off unevenly. Workloads that fit inside a bigger cache pool run faster because they stop reaching out to slower memory as often, and a lot of games are exactly that kind of workload. Workloads that are already memory-bandwidth-bound rather than cache-bound — heavy rendering and compilation among them — see far less from the same extra megabytes. Which of those describes what you actually run is the question worth answering before the cache figure decides anything.

TDP versus what a chip actually pulls

TDP, as the catalog records it, is a base-power rating: AMD's own "TDP" term, Intel's "processor base power," or a laptop chip's nominal figure. It was never meant to describe worst-case draw, and treating it as one is the single most common cooling mistake in a first build. Where a maker also publishes a separate peak figure — Intel's maximum turbo power, AMD's package power tracking limit — the site records that too, as its own "Max power" figure alongside TDP. Not every chip states one; it only appears on a chip's page when a source actually published it. Where a maker does publish it, the maximum figure the catalog records is never at or below the base rating — it sits above it, sometimes by a wide margin.

Cooling and power delivery should be sized against whichever number is higher for a given chip, not the one printed larger on the retail box. A chip with no published maximum is not a chip that never exceeds its TDP; it's a chip whose maker didn't publish the other number.

Integrated graphics: a fallback display, not a discrete GPU

Whether a chip has integrated graphics at all is recorded as a plain yes or no, and where the answer is yes, the specific graphics silicon is named alongside it. Every laptop chip in the catalog reports one — a soldered part always ships without any assumption that a discrete GPU sits nearby. Desktop is where the split shows up: some current chips from both AMD and Intel disable the graphics silicon outright and record integratedGraphics as false, so building around one of those means a discrete card isn't optional, it's part of the parts list from day one.

Where a chip does carry one, treat it as a fallback rather than a second GPU. It exists to put a picture on a monitor before a graphics card is installed, to keep a slot free in a compact build, or to cover an office machine that will never run anything demanding. It is not sized, cooled, or powered like hardware meant to carry a modern game.

What the catalog holds, and what a listing still won't tell you

Every desktop chip's page leads with cores and threads, then lists whatever else a source actually stated for that model: socket, boost clock, cache, TDP, and where published, a maximum power figure, base clock, an integrated GPU model, and the rest of the fields a spec sheet can carry. A figure that isn't in the source appears nowhere on the page — the row is left off rather than filled in with a guess or borrowed from a sibling chip. Every number behind it traces back to a manufacturer datasheet or the maker's own press material, never a benchmark run here.

That also marks the edge of what a spec sheet settles. None of the figures above are a frame-rate number, a render time, or a score from running anything — they describe what a chip is built to do, not how fast it did it in someone's test rig. Matching a build to a workload still takes reading a review built around that workload, not just the numbers on this page.

Desktop and laptop chips are tracked as separate surfaces rather than one mixed list, for a reason beyond the socket difference above: a laptop chip is soldered into a specific machine and was never sold on its own, so it carries no price of its own to track over time the way a boxed desktop chip does. Every desktop CPU in our catalog is on its own page, with specs read from the datasheet and, where the chip is sold on its own, a price tracked over time rather than a one-time list price. Laptop chips get their own page on the same terms, minus the price history a soldered part never has.