K10 dies 45 nm
| K10 dies 45 nm | |
|---|---|
 The annotated X6 45 nm CPU die used in Phenom II X6 and 8400-series Opteron models. | |
| Overview | |
| Introduced | 2009/January/08 |
| Process Node | 45 nm SOI |
| Microarchitecture | |
| Chronology | |
| Predecessor | K10 dies 65 nm |
| Successor | K10 dies 32 nm, Bulldozer 15h |
The K10 45 nm dies are the most common models of the K10 microarchitecture. These 45 nm dies include the popular Phenom II and Athlon II processors.
Phenom and Athlon models without "II" in their names are usually 65 nm dies, and should therefore be avoided.
As of 2026, approximate prices for pre-owned models are:
- USD 13 for Phenom II X4 940T (3.4 GHz turbo @ 95 W)
- USD 13 for Athlon II X4 650 (3.2 GHz @ 95 W)
- USD 8 for Athlon II X3 460 (3.4 GHz @ 95 W)
Opteron dies are generally the same as Phenom II and Athlon II dies, but socket and motherboard support can differ.
Supported Sockets
- AM3+ (for desktop motherboards, newer)
- AM3 (for desktop motherboards)
- AM2+ (only for Athlon X2 5000+/5200+)
- C32 (for server motherboards)
- G34 (for server motherboards)
- Socket F (for server motherboards, older)
L3 and L2 cache
Almost all 45 nm Phenom II models have 6 MiB of L3 cache, while 45 nm Athlon II models have no L3 cache.
Some Athlon II models (those using the X2 die) have 1024 KiB L2 per core, while others have 512 KiB L2 per core. The larger L2 cache typically gives a small performance gain, around 2%.
At the same thermal envelope, 6 MiB of L3 cache typically improves performance by about 7–17%, depending on the workload. The lack of L3 cache is therefore not as important as many people think.
C2 vs. C3 vs. D0 vs. D1 vs. E0 steppings
More recent steppings are generally better.
For example, many top-line C3 dies can reach 4 GHz easily, while C2 dies struggle at high frequencies.
E0 steppings usually have better thermals than C3.
Thermal Limits
The adhesive securing the dies softens/melts at about 95°C, which is likely to cause irreversible damage.
The built-in thermal protection should trigger an emergency shutdown when a core reaches 90°C, but this is unreliable because K10’s on-die thermal sensors are essentially broken (see below).
The generally accepted safe limit for long-term operation is a max. 70°C for the dies. However, because the on-die sensors are malfunctioning, the reported temperature usually does not correspond to the actual die temperature with that 70°C limit — see below for detailed instructions.
The 70°C thermal limit does not mean the CPU will immediately fail at 71°C. Short periods at about 80°C are generally safe. The 70°C limit means the CPU will likely run without major damage for at least five years of continuous operation at that steady temperature.
Thermal Sensors
Thermal sensors aren’t quite working properly — and yes, there’s more than one, so it depends on which sensor you’re looking at.
The K10 on-die sensor appears as "Core" in HWMonitor on Windows; some sources call it "Tdie"—we will use "Tdie" in this article.
On Linux it’s usually exposed as "k10temp". To read its value (in millidegrees):
cat /sys/class/hwmon/hwmon0/temp1_input
That's the broken sensor — but not completely useless: it reports relative temperature correctly and reacts faster than the other sensor (Tmb). AMD gave this quirk the polite name "temperature offset" — a fancy label for the sensor being a bit off.
Since the K10 on-die sensor has the "temperature offset," motherboard makers added another sensor under the CPU socket. We’ll call that sensor "Tmb" — it measures the actual temperature that corresponds to the 70°C safe limit.
Except not quite. Not all boards include the extra thermal sensor under the socket — notably the ASUS M5A97 EVO R2, which is ironic for a board aimed at overclockers. A glowing TechPowerUp review missed that the board can’t measure temperature correctly, which makes safe overclocking virtually impossible. Another example is the ASUS M5A97 EVO2 M51BC/M52BC — it also lacks the sensor, but at least it doesn’t support overclocking in the vendor's UEFI. The thermal sensor costs about USD 0.30, so omitting it is an obviously excellent design choice.
Second, Tmb isn’t read by the CPU — it’s read by a separate motherboard chip, and different boards use different chips. That means you need some luck that your OS already has a driver for that particular chip.
But wait — Linux won’t load the sensor-chip driver automatically. First detect which sensor chip your board has by running "sensors-detect" from the shell. Follow its prompts, then add the reported driver module(s) to your system’s module autoload list so the sensor is available after reboot.
After that, read temperatures with the "sensors" command. Or install the pSensor GUI app, add the monitored sensors, and view temperature graphs.
Usually a driver exists because there are only a few common sensor chips and modern OSes now support them. However, each sensor chip on Linux is exposed under its own hwmon path, typically like:
/sys/class/hwmon/hwmonNUMBER/SOMETHING
The Temperature Offset
Each CPU has its own offset for the Tdie (k10temp) sensor. On average the offset is about 10°C and can easily reach 20°C or more. Typically Tdie reads up to ~15°C lower than actual or up to ~5°C higher. The good news: the offset is roughly constant, so relative changes are reliable.
Or not—well, kind of. The temperature offset can change if you unlock or disable cores in the BIOS. On E0 steppings, unlocking extra cores makes the Tdie sensor read 0°C, which renders it mostly useless.
Some rare boards (e.g., "ASUS m5A88‑M") start thermal throttling when Tdie reaches 70°C; that can be disabled in the vendor's BIOS. If you are lucky because your CPU’s temperature offset is near 0°C, you can effectively get Turbo Core-like behavior without a real turbo — set the clock as high as you like and the vendor's BIOS on m5A88‑M will keep the CPU at or below the 70°C throttle point. Brilliant.
Thermals
Quick rule of thumb for best performance based on cooling capacity:
- Up to ~ 40 W cooling capacity: choose an X2 die
- ~ 40–60 W cooling capacity: choose an X3 die
- More than ~60 W cooling capacity: choose an X4 or X6 die
Baseline power for X4 dies is about 10 W. That means if the OS uses any core, all cores must be powered, so even at the lowest clock rate (800 MHz) the package still draws roughly 10 W.
Of course, dies draw less than 10 W when lightly loaded because the OS can put cores into sleep states. To reach below 10 W, a modern OS will automatically toggle a die’s power state on and off quickly.
The most efficient clock rate is about 2.2–2.7 GHz.
On top of the baseline power, estimated additional heat from one core under heavy load (Prime95):
- 1 core @ 2700 MHz (Prime95) = baseline power + approx. 10 W
- 1 core @ 3200 MHz (Prime95) = baseline power + approx. 16 W
After about 3200 MHz, most cores start to demand much more extra power for each additional 100 MHz clock rate increase.
The dies are attached to the heat spreader with a thermal solder. The solder’s thermal conductivity varies a lot depending on how it settles on each unit, so dies drawing the same electrical power can run hotter or cooler depending on whether the solder settled well or poorly.
BIOS vs. UEFI
Some vendors used BIOS firmware on older motherboards; more recent motherboards use UEFI vendor firmware.
Generally, UEFI was still fairly new then, so UEFI boards had a plentiful of issues in vendor's firmware — bugs, no automatic ECC support, dumb fan-controller defaults that force poorer operation modes, flaky DDR3 training that often needed extra DRAM voltage, and a bunch of other never-fixed bugs.
BIOS boards tend to have fewer vendor's firmware bugs and more features that actually work. Their downsides: they usually lack IOMMUv1 (only found on some chipsets), and older chipsets (800‑series) have more hardware bugs — though some never-fixed bugs remain even on the 900‑series hardware.
The POP bug
...to do