Every PC has a weakest link. That's not a design flaw, it's just how components age at different rates. The question that actually matters isn't whether a bottleneck exists. It's how big it is, and whether it's big enough to do anything about.
We get asked about this a lot, usually by someone who has read a forum thread, seen a frame rate graph that looks worrying, and now isn't sure if they need a new graphics card, a new processor, or an entirely new machine. So here's the framework we actually use when we're looking at a customer's system, without pretending it's more scientific than it is.
What a bottleneck actually is
A bottleneck happens when one component, usually the CPU or the GPU, is meaningfully slower than the rest of the system. The faster parts end up waiting around for the slower one to catch up, rather than working flat out. You paid for that performance. If it's sitting idle, you're not getting it.
The tricky part is that some amount of this is completely normal. No system is perfectly balanced all the time, because workloads change from one game or application to the next. A small gap isn't a problem. A large one is. The rest of this piece is about telling the two apart.
A rough framework for judging severity
This isn't a precise formula, and anyone who tells you it is has not actually built PCs for a living. But as a general guide, it holds up well:
- Under 10 to 15 per cent: leave it alone. If the weaker component is only costing you that much performance versus a perfectly balanced system, you won't feel it in real use. This is background noise, not a problem.
- 20 to 40 per cent: a targeted upgrade starts to make sense. This is the zone where swapping just the one weak component, usually the GPU, brings a noticeable improvement without needing to touch anything else.
- Beyond 40 per cent, especially with an older CPU platform: a full rebuild is usually the more sensible move. At this point you're not fixing a bottleneck, you're propping up a platform that's fundamentally behind, and that tends to be a losing game.
Treat those bands as general guidance rather than a scientific formula. Real workloads are messy, and the exact percentage will shift depending on the game, the resolution, and what you're doing with the machine. But the shape of the framework, small gap ignore it, medium gap upgrade it, large gap rebuild around it, holds true almost everywhere we've seen it applied.
How to tell if your CPU is the bottleneck
You don't need expensive software to get a reasonable read on this. Most modern monitoring overlays (MSI Afterburner, the Xbox Game Bar overlay, or similar) will show you GPU usage in real time while you play. Here's what to look for:
- GPU usage sits noticeably below 95 to 100 per cent at your target resolution.
- Frame rates are lower than you'd expect for that graphics card, given what you've seen it do elsewhere.
- The gap is worse in CPU-heavy titles: simulation games, strategy games, and multiplayer shooters with a lot of players on screen at once tend to expose this first.
If that pattern holds across more than one game, especially the CPU-heavy ones, your processor is the one holding things back, not your graphics card.
How to tell if your GPU is the bottleneck
The signature here is close to the opposite. Your GPU usage will sit at or near 100 per cent almost constantly, while the CPU clearly has headroom to spare. Another useful test is dropping settings that mostly affect CPU load, things like draw distance or simulation detail, and seeing whether frame rates move. If they barely budge, the CPU was never the limiting factor. The GPU is working as hard as it can and simply doesn't have any more to give.
This is the more common bottleneck for people gaming at higher resolutions, since GPU demand scales heavily with resolution while CPU demand doesn't move nearly as much.
Why platform age matters more than the CPU alone
Raw benchmark numbers only tell you part of the story. An ageing motherboard chipset can box you in even if the CPU itself isn't catastrophically behind. Fewer PCIe lanes limit what storage and graphics cards you can actually fit. Older memory standards cap how much bandwidth you can feed the system. And if there's no realistic CPU upgrade path on that socket, you're stuck with what you've got the moment the CPU does become the limit.
This is often the detail that tips a borderline case towards a rebuild. The CPU on its own might sit in that 20 to 40 per cent gap, upgradeable territory, but if the platform around it can't take a better chip and can't take faster memory either, you're not upgrading a component, you're extending the life of a dead end.
Why a rebuild isn't automatically the safe choice
It's tempting to treat a full rebuild as the responsible, thorough option and a single-part upgrade as the cheap-and-cheerful one. That's not always true. A rebuild often means re-buying a case, power supply, and cooling as well, not just the CPU and motherboard, especially if those parts are old enough to be near the end of their own useful life or simply don't fit the new components.
A fair comparison has to weigh what you can genuinely reuse against what you'd actually be buying new. Sometimes that tips things back towards a targeted upgrade even when the CPU gap looks large on paper. This is exactly the kind of arithmetic we covered properly in our piece on the cost side of this decision, so we won't repeat it here. What matters for this framework is just that platform age and reusability both belong in the decision, not just the raw performance gap.
Our approach when someone asks us
We would rather tell a customer that their existing system just needs one part changed than sell them a full rebuild they don't need. That costs us a sale in the short term and we're fine with that, because it's the honest answer and it's the reason people come back to us later for the bigger job when it's actually warranted.
Equally, when a system genuinely is past the point where bolting on one new part makes sense, we'll say that plainly too. Nobody benefits from us tiptoeing around a rebuild that's clearly the better call just to make the conversation easier.
Build it with us
If you've worked through this and you're still not sure which band your system falls into, that's a normal place to land, the diagnosis isn't always clean cut. Try our PC configurator to see what a rebuild would actually look like spec for spec, or get in touch and talk it through with Matt or Michael. Every system we build is hand-built to order in our Stevenage workshop, stress-tested before it ever leaves us, and backed by our 5-year warranty.
