FLOPS measures floating-point operations per second and TOPS measures tera-operations per second, two ways of stating how many math operations a chip can run each second. The converter below reads a value in any FLOPS or TOPS unit, reduces it to operations per second using the metric prefixes, and shows the equivalent across the gigascale, terascale, and petascale. Everything runs in your browser.
What FLOPS and TOPS Mean
FLOPS stands for floating-point operations per second. A floating-point operation is a single arithmetic step, such as a multiply or an add, performed on a number that has a decimal point. TOPS stands for tera-operations per second, and the operations it counts are usually integer operations rather than floating-point ones. Both terms describe throughput: how many basic math steps a processor completes in one second. The number in front, such as the 100 in 100 TFLOPS, is scaled by a metric prefix, so the same chip can be quoted in FLOPS, GFLOPS, or TFLOPS depending on which prefix keeps the number readable.
How to Use It
- Type the value you want to convert into the value box.
- Pick the unit it is given in, such as TFLOPS for a GPU spec or TOPS for an NPU spec.
- Read the result. The big number shows the most natural scale, and the list shows the same rate in gigascale, terascale, and petascale.
- Use the base rate line, in operations per second, when you need the raw figure for your own math.
The Metric Prefixes
| Prefix | Symbol | Multiplier | Operations per second |
|---|---|---|---|
| Kilo | K | 1e3 | 1,000 |
| Mega | M | 1e6 | 1,000,000 |
| Giga | G | 1e9 | 1,000,000,000 |
| Tera | T | 1e12 | 1,000,000,000,000 |
| Peta | P | 1e15 | 1,000,000,000,000,000 |
Each step up the table multiplies the rate by one thousand. So one TFLOPS equals one thousand GFLOPS, and one PFLOPS equals one thousand TFLOPS. The converter applies these same multipliers whether the unit is a FLOPS unit or a TOPS unit, because the prefixes describe the count, not the kind of operation.
FLOPS vs TOPS
Where These Numbers Appear
FLOPS shows up on graphics card spec sheets, where a GPU is often quoted in TFLOPS for FP32 or FP16 work, and on supercomputer rankings measured in PFLOPS. TOPS shows up on NPUs, the neural processing units inside phones and laptops, and on standalone AI accelerators built for inference. A laptop marketed as having a 40 TOPS NPU is stating integer inference throughput, while a GPU listed at 80 TFLOPS is stating floating-point throughput. Knowing which unit a spec uses tells you which kind of workload the number describes.
Why They Are Not Directly Comparable
Because TOPS and FLOPS count different operations at different precision, you cannot treat one TOPS as one FLOPS even though both sit on the same tera scale. An INT8 operation moves fewer bits than an FP32 operation, so a chip can claim a higher TOPS number than its TFLOPS number while doing less precise math. This converter places both on the same numeric scale so you can read the magnitude, but it does not pretend the operations are equivalent. Treat the result as a scale conversion, not a performance equivalence between floating-point and integer work.
When to Use It
Reach for this converter when a spec sheet quotes a unit you do not normally read, such as GFLOPS on an older card or POPS on a data-center accelerator, and you want it in a scale you recognize. It is also useful when you are comparing two products that report in different prefixes, since putting both at the terascale makes the gap obvious. Keep the precision in mind when the units differ in kind, and use the base rate line whenever you need the plain operations-per-second figure.
Last Thoughts on FLOPS and TOPS
FLOPS and TOPS look interchangeable because they share the metric prefixes, but the unit you read tells you whether a chip is being measured on floating-point or integer work. Convert the scale freely, and hold the precision fixed when you compare. A 40 TOPS NPU and an 80 TFLOPS GPU are answering different questions, and the right comparison keeps the operation type the same.
Convert your next spec figure above, then pick the scale that reads cleanest for your notes. For related hardware math, see our VRAM for local LLM tool and the LLM cost calculator, or browse the rest of our free online tools.
Key Takeaways:
- FLOPS counts floating-point operations per second; TOPS counts tera-operations per second, usually integer.
- The metric prefixes K, M, G, T, and P each multiply the rate by one thousand from the step below.
- One TFLOPS equals one thousand GFLOPS, and one PFLOPS equals one thousand TFLOPS.
- FLOPS appears on GPUs and supercomputers; TOPS appears on NPUs and AI accelerators.
- TOPS and FLOPS share a number scale but measure different operations at different precision, so they are not directly comparable.
- Compare TOPS to TOPS and FLOPS to FLOPS at the same precision, such as INT8 or FP16.
Frequently Asked Questions (FAQs)
Is TOPS the same as FLOPS?
No. TOPS and FLOPS both count operations per second and share the metric prefixes, but FLOPS counts floating-point math while TOPS usually counts integer math at lower precision. They sit on the same number scale, so one TOPS and one TFLOPS are the same magnitude, but they do not describe the same work.
How many GFLOPS are in one TFLOPS?
One TFLOPS equals one thousand GFLOPS. The tera prefix is one thousand times the giga prefix, so multiplying a TFLOPS figure by 1,000 gives the GFLOPS figure, and dividing GFLOPS by 1,000 gives TFLOPS.
Why do NPUs use TOPS instead of FLOPS?
NPUs are built mainly for inference, which often runs in low-precision integer formats such as INT8. Because the operations are integer rather than floating-point, vendors report them in TOPS. GPUs aimed at training report in FLOPS because that work uses floating-point formats like FP16 and FP32.
Can I convert TOPS to FLOPS?
You can place both on the same scale, since 1 TOPS and 1 TFLOPS are the same count of one trillion operations per second. What you cannot do is treat them as equal performance, because TOPS usually counts integer operations and FLOPS counts floating-point operations, which differ in precision and in the work each step does.
What does PFLOPS mean?
PFLOPS means petaflops, or one quadrillion floating-point operations per second. The peta prefix is one thousand times the tera prefix, so one PFLOPS equals one thousand TFLOPS. It is the scale used to rank supercomputers and large data-center clusters.
Does a higher TOPS number mean a faster chip?
Not on its own. A higher TOPS number means more operations per second at the precision it was measured, usually INT8. A chip can show a high TOPS figure and a lower TFLOPS figure because the integer operations move fewer bits. Compare figures at the same precision and the same operation type to judge speed fairly.
