NPU marketing loves big numbers, and the jump from 40 to 45 TOPS gets presented as a meaningful upgrade. But TOPS — trillion operations per second — is a theoretical peak that rarely describes your actual experience with Recall, Live Captions, or Cocreator. Understanding what the number means (and does not mean) saves you from overpaying for headroom you will never use.

The practical truth: Microsoft defined Copilot+ at 40 TOPS because that level runs every current AI feature with room to spare. The 5 TOPS gap to 45 is a ~12% theoretical difference in a component that is not the bottleneck for anything you do today. This guide explains when NPU power matters, when it does not, and how to make sure your NPU — 40 or 45 TOPS — is actually working.

If you are choosing between specific chips, see our Snapdragon X Plus vs X Elite comparison, and for broader buying advice, the refurbished AI laptop buying tips guide. Zoom users should also check Studio Effects Zoom meeting issues .

Why this happens

  • Marketing needs differentiation: Small spec gaps get amplified to justify price tiers.
  • TOPS is theoretical peak: Real throughput depends on model optimization, precision formats, and memory bandwidth.
  • Software targets the baseline: Developers optimize for the 40 TOPS Copilot+ floor, not the ceiling.
  • NPU rarely the bottleneck: For current features, the NPU waits on software and data, not the reverse.
  • Confusion with GPU/CPU: Buyers apply CPU/GPU “bigger is better” logic to a component with different dynamics.
  • Future-proofing anxiety: Fear that 40 TOPS will soon be obsolete drives unnecessary upselling.
  • Spec-sheet shopping: TOPS is easy to compare; real differentiators (drivers, thermals) are harder to quantify.

How to fix it

Work through these in order — diagnose first, then act.

Fix 1: Understand what TOPS actually measures

TOPS counts raw multiply-accumulate operations per second at a given precision — usually INT8. It does not measure how fast Recall searches, how quickly Live Captions transcribe, or how good Cocreator’s images look. Those depend on model quality and software integration far more than raw throughput. A well-optimized 40 TOPS NPU beats a poorly-utilized 50 TOPS one in real tasks.

  1. Check your NPU’s rated TOPS in Task Manager’s Performance tab under the NPU tile.
  2. Understand that INT8 TOPS (the advertised number) differs from FP16 performance, which some models need.
  3. Recognize that memory bandwidth often limits NPU performance before compute does.
  4. Do not compare TOPS across vendors directly — measurement methodologies differ.
  5. Focus on whether your actual AI features work well, not the number on the box.

Fix 2: Verify your NPU is actually being used

A surprising number of “is 40 TOPS enough?” worries come from NPUs sitting at 0% while features run on the CPU instead. Before judging NPU power, confirm it is actually engaged — an idle NPU makes any TOPS debate moot.

  1. Open Task Manager’s Performance tab and watch the NPU tile while using Live Captions or Studio Effects.
  2. If the NPU stays at 0%, the feature may be running on CPU — check the app’s AI settings.
  3. Update NPU drivers from your laptop maker; outdated drivers fail to route work to the NPU.
  4. In Windows AI settings, confirm on-device processing is enabled where offered.
  5. Test with a known NPU workload (Live Captions during video playback) and watch for NPU activity.
  6. If no NPU appears at all, see neural processors missing from Device Manager .

Fix 3: Decide whether 45 TOPS is worth paying for

For almost every buyer, the answer is no — not as a deciding factor. The 12% theoretical gap does not translate into 12% better Recall or captions. Spend the price difference on RAM, storage, display quality, or battery capacity, which you will feel every day.

  1. Compare the actual price gap between 40 and 45 TOPS configurations.
  2. Ask what else that money buys: 16GB more RAM beats 5 more TOPS for every real workload.
  3. Consider that software targets the 40 TOPS baseline for years to come.
  4. Only value 45+ TOPS if you run specialized local AI models (LLMs, Stable Diffusion) that can use the headroom.
  5. For those specialized workloads, also weigh GPU options — GPUs often outperform NPUs for large models.

Fix 4: Optimize the NPU you have

Whether 40 or 45 TOPS, an optimized setup extracts the real-world performance the number promises. Driver health, power settings, and thermal headroom determine actual NPU throughput far more than the rated peak.

  1. Keep NPU/chipset drivers current through your laptop maker’s validated packages.
  2. Use Balanced or Best performance power mode when running sustained AI workloads.
  3. Ensure good ventilation — thermal throttling hits the NPU along with the CPU.
  4. Close competing heavy apps during intensive AI tasks to free memory bandwidth.
  5. Keep Windows updated; Microsoft continuously improves NPU scheduling.

Fix 5: Understand when NPU power genuinely matters

There are real scenarios where more TOPS helps: running local large language models, batch-processing thousands of images, or future heavier on-device models. But these are enthusiast and professional niches, not Copilot+ features. Know which camp you are in.

  1. Local LLM inference (7B+ parameter models): more TOPS and more RAM both help significantly.
  2. Batch AI photo/video processing: throughput scales with NPU power.
  3. Real-time heavy models (advanced background effects at 4K): headroom prevents dropped frames.
  4. Standard Copilot+ features: 40 TOPS is already more than enough.
  5. If you are in the enthusiast camp, also consider laptops with stronger GPUs for these tasks.

Fix 6: Stop spec-sheet anxiety and evaluate the whole laptop

TOPS is one number among dozens that determine your experience. A 40 TOPS laptop with 32GB RAM, a great display, and excellent drivers will make you happier than a 45 TOPS laptop with 16GB RAM and a dim screen. Evaluate holistically.

  1. List your top five daily tasks and rank which specs affect each.
  2. Notice how rarely the NPU appears in that ranking for typical use.
  3. Prioritize RAM, SSD, display, keyboard, battery, and thermals over TOPS.
  4. Read long-term owner reviews, not just launch benchmarks.
  5. Check the maker’s driver support reputation — it matters more than 5 TOPS.

Closing

45 TOPS vs 40 TOPS is a 12% theoretical gap in a component that is not the bottleneck for any current Copilot+ feature. Both run everything Microsoft offers, identically. Buy the laptop with the better RAM, display, battery, and drivers — not the bigger TOPS number — and make sure whatever NPU you have is actually healthy and engaged. That is the entire story.

Frequently asked questions

What does TOPS mean for an NPU?
TOPS stands for trillion operations per second — a measure of the NPU’s theoretical AI math throughput. Higher TOPS means the NPU can process larger AI models faster, but real-world speed also depends on software optimization and memory bandwidth.
Is 40 TOPS enough for Copilot+ PCs?
Yes. Microsoft set 40 TOPS as the Copilot+ requirement precisely because it comfortably runs Recall, Live Captions, Cocreator, and Studio Effects. Laptops with 40 TOPS run every Copilot+ feature fully.
Will 45 TOPS make AI features faster than 40 TOPS?
Rarely in noticeable ways. Current Copilot+ features are not bottlenecked by NPU throughput at 40 TOPS, so the extra 5 TOPS mostly sits idle. Future, heavier on-device models might use it, but that is speculative.
How can I check my NPU's TOPS rating?
Open Task Manager, go to the Performance tab, and click the NPU tile — recent Windows builds show the TOPS rating there. Your laptop maker’s spec sheet also lists it.