Tweet
Turns out this was somewhat old news, but it wasn't very well documented.
In the next patch release of PerformanceTest we are adding improvements to the detection of overclocking on some of the newer AMD's CPUs. (Or 'APU's as AMD now like to call them).
While testing on a AMD A4 3400 CPU we found we could overclock the multiplier to x47 in BIOS. Raising the CPU's speed from 2700Mhz to 4700Mhz.
(base clock of 100Mhz with a x47 multiplier)
What we found strange however is that benchmark scores didn't increase when the CPU was overclocked. CPU didn't get hot either, despite running at nearly 5Ghz on air cooling, the CPU remained at around 35C.
CPUZ, Windows and our own PerformanceTest code reported the new speed as 4700Mhz however.
After some investigation it turns out this is a bug, or maybe even 2 bugs. The first bug is in the BIOS. AMD A4 3400 has a locked multiplier, so the BIOS should not allow the user to adjust the multiplier, but it does anyway.
Second bug is in the CPU itself. CPUz and PerformanceTest use what's known as MSRs, "Model specific registers" in the CPU to report on the CPUs current speed. The wrong information is being reported in these MSRs. So all the monitoring tools also report the wrong information.
After some searching on the internet it seems that this problem has been known about for months, but no one could be bothered fixing it.
This might effect all the Socket FM1 Llano chips. These include
A8-3850, A8-3800, A6 3600, A6-3650, A4-3400.
Over at Techpowerup they called this "Empty Overclocking".
Other related links
http://www.techpowerup.com/148636/A8...ultiplier.html
http://www.xbitlabs.com/articles/cpu...8-3800_22.html
Note 1: that you can still overclock these chips and really have an increase in performance, but you need to overclock the base clock and not the multiplier. You can also under clock the multiplier.
Note 2: We aren't sure at the moment what effect all this has on the new unlocked 'K' A-series APUs like the A8-3870K & A6-3670K.
Update: We have included a possible workaround for this in PerformanceTest. See this post for more details.
http://www.passmark.com/forum/showthread.php?t=3635
Update 2: There is also another performance related issue with the Llano CPUs. A patch for a CPU bug can impact the performance. See this post for details.
http://www.passmark.com/forum/showthread.php?t=3656
In the next patch release of PerformanceTest we are adding improvements to the detection of overclocking on some of the newer AMD's CPUs. (Or 'APU's as AMD now like to call them).
While testing on a AMD A4 3400 CPU we found we could overclock the multiplier to x47 in BIOS. Raising the CPU's speed from 2700Mhz to 4700Mhz.
(base clock of 100Mhz with a x47 multiplier)
What we found strange however is that benchmark scores didn't increase when the CPU was overclocked. CPU didn't get hot either, despite running at nearly 5Ghz on air cooling, the CPU remained at around 35C.
CPUZ, Windows and our own PerformanceTest code reported the new speed as 4700Mhz however.
After some investigation it turns out this is a bug, or maybe even 2 bugs. The first bug is in the BIOS. AMD A4 3400 has a locked multiplier, so the BIOS should not allow the user to adjust the multiplier, but it does anyway.
Second bug is in the CPU itself. CPUz and PerformanceTest use what's known as MSRs, "Model specific registers" in the CPU to report on the CPUs current speed. The wrong information is being reported in these MSRs. So all the monitoring tools also report the wrong information.
After some searching on the internet it seems that this problem has been known about for months, but no one could be bothered fixing it.
This might effect all the Socket FM1 Llano chips. These include
A8-3850, A8-3800, A6 3600, A6-3650, A4-3400.
Over at Techpowerup they called this "Empty Overclocking".
Other related links
http://www.techpowerup.com/148636/A8...ultiplier.html
http://www.xbitlabs.com/articles/cpu...8-3800_22.html
Note 1: that you can still overclock these chips and really have an increase in performance, but you need to overclock the base clock and not the multiplier. You can also under clock the multiplier.
Note 2: We aren't sure at the moment what effect all this has on the new unlocked 'K' A-series APUs like the A8-3870K & A6-3670K.
Update: We have included a possible workaround for this in PerformanceTest. See this post for more details.
http://www.passmark.com/forum/showthread.php?t=3635
Update 2: There is also another performance related issue with the Llano CPUs. A patch for a CPU bug can impact the performance. See this post for details.
http://www.passmark.com/forum/showthread.php?t=3656