What is a safe temperature for a CPU or GPU?

There is no single safe temperature that fits every processor and graphics card, because each chip has its own limits. Pulse by Zaweba deals with this by reading the hardware's own limits wherever it can, such as the CPU's maximum junction temperature (TjMax), and colouring each sensor normal, warm or hot against them. Where it can't read a limit it uses sensible defaults, for example 75 and 85 degrees C for a GPU core, so you can see at a glance where yours stands.

So the honest answer to the question is "it depends on the chip", and the useful answer is to use a tool that knows the chip's own figures.

Why one number doesn't fit all

Chip makers design their parts to run up to a certain temperature and to protect themselves above it, usually by slowing down. That ceiling differs between models and generations. A figure that is perfectly comfortable for one laptop processor may be near the top for another chip. Graphics cards differ again, and modern cards report more than one temperature: the core, a hot spot (the warmest point measured) and the memory.

That is why a reading on its own can mislead. Seventy-something degrees sounds alarming if you don't know the limit, and sounds fine if you think every chip can take a hundred.

How Pulse sets its bands

The Temperatures page in Pulse (Ctrl+2) has a tile for each sensor: CPU package and cores, GPU core, hot spot and memory, the motherboard or ACPI thermal zone, and each drive. Every tile shows the reading, where it came from, and its band with the warm and hot thresholds printed on it.

For the CPU, Pulse uses TjMax when it can read it, so the bands sit relative to what your particular processor is built to take. For NVMe drives it uses the drive's own warning temperature. Where no limit is readable, the defaults are: GPU core 75 and 85 degrees C, GPU hot spot 85 and 95, motherboard 70 and 85, and drives 55 and 65.

CPU temperatures need the optional sensor helper, turned on in Settings > Sensor readings, which asks Windows for permission and also needs the free PawnIO driver. GPU and NVMe readings work without it.

What normal, warm and hot mean in practice

Normal is where the chip should sit most of the time. Warm is not a problem in itself; a CPU or GPU under a heavy game or a long render is expected to climb into it. What deserves attention is a sensor that sits in warm while the PC is doing very little, or one that reaches hot and stays there. Those are signs of dust, a blocked vent, a fan that isn't keeping up or a program working the hardware when it shouldn't.

Time matters as much as the figure. A brief peak while a game loads is different from an hour in the hot band. That's where Pulse's history helps: the History page lists temperature peaks for today, yesterday or this week alongside the three busiest apps in that minute.

Letting the PC tell you

You don't have to keep checking. The temperature alert on the Alerts page is on by default and fires when a sensor stays past its threshold for five seconds. Leave the value at 0 to use the sensor's own hot band, or set a fixed figure in degrees C if you want an earlier warning. The alert arrives once as a Windows notification and goes into the alert log.

If you see temperatures you don't like, the usual fixes are physical: clean out dust, give the case or laptop room to breathe, and check that the fans are turning. Pulse by Zaweba can show fan speeds in RPM too, once the sensor helper is on and if your hardware reports them.