AMD Athlon 5350 / 5150 & Sempron 3850 / 2650

[AJSB]

What i see on that picture also wrongly setuped is also that he plugged one DIMM in DIMM_A1 , but manual explictly says: Ensure to insert the DIMM into DIMM_A2 socket if you install only one memory module .

Unless he took picture and then realized that was in wrong way and changed to correct one...if not, ROFLMAO !

I don't have the MoBo and i read the manual

However, i didn't noticed the picture of Michael's setup (good catch from you)...if he indeed made the 1st run of tests with DIMM in that position (and 2nd run in correct position), that could explain the problems in 1st run and that improvements in second run was so high and that were not related to Kernel version...

[dungeon]

Unless he took picture and then realized that was in wrong way and changed to correct one...if not, ROFLMAO !

I don't have the MoBo and i read the manual

However, i didn't noticed the picture of Michael's setup (good catch from you)...if he indeed made the 1st run of tests with DIMM in that position (and 2nd run in correct position), that could explain the problems in 1st run and that improvements in second run was so high and that were not related to Kernel version...

Seems like that is it, he putted memory in not recommended bank if one memory used .

I compiled my localmod 3.15rc1 kernel (with HDD, Michael even use SSD) under all three those kernels and it is the same ~11-12 minutes . So there is not any major difference between 3.13, 3.14 and 3.15rc1 CPU wise. All are the same/very similar in my benchmarks for the GPU and for the CPU .

Of course 3.13 kernel in Ubuntu 14.04 and Debian Sid only needs one tweak - dpm enable. And both colortilings in xorg.conf .

I also tested with kernel compile time how it goes with default (@2050MHz) and overclocked (@2206MHz) so it saves ~50 seconds in compile time. Which is ~7% stable overclock for me for the same ~7% gain .

[Michael]

From my testing with AM1M-A and Athlon 5350 with Lubuntu 14.04 and Debian Sid kernel 3.14 is not explicitly needed , so xUbuntu users can stay with default kernel and default mesa, etc... but they have to tweak few settings: radeon.dpm=1 as kernel parameter and ColorTiling together with ColorTiling2D in xorg.conf - that way performance is roughly the same like you see in this article .

I am using it right now with 3.13 kernel no problem, and performance is like you see in this article i even overclocked it to 2206 MHz and it works fine .

With the Oibaf packages on the upgraded stack both ColorTiling/ColorTiling2D are enabled. With Linux 3.14 the DPM is also enabled. Only when I was explicitly testing the Ubuntu 14.04 "out of the box" experience I didn't make any changes.

[Michael]

What i see on that picture also wrongly setuped is also that he plugged one DIMM in DIMM_A1 , but manual explictly says: Ensure to insert the DIMM into DIMM_A2 socket if you install only one memory module .

The DIMM is how it arrived as installed by AMD's PR team and is still in that slot.

[dungeon]

The DIMM is how it arrived as installed by AMD's PR team and is still in that slot.

Hm, OK then... but you better put it in A2 which is what manual says http://dlcdnet.asus.com/pub/ASUS/mb/..._AM1I-A.pdf#15

The AM1 APU temperatures are exposed under Linux using the k10temp driver. The fam15h_power driver that's supposed to report the approximate power draw of the APU individually was also automatically loaded for this hardware, but it didn't actually work. The AMD "Fam15h" Linux power driver on the 3.14 kernel was just always reporting 0.00 Watts while recognizing the 25 Watt limit of all four APUs. The Radeon PCI thermal driver was also reporting the reported GPU temperature on the Linux 3.14 kernel that I loaded past the Ubuntu 14.04 install.

Wattage reading does not work even in 3.15-rc1 kernel as i see , but they added support for Mullins - "Kabini support added in kernel 3.12. Mullins support added in kernel 3.15. "

[haplo602]

Hm, OK then... but you better put it in A2 which is what manual says http://dlcdnet.asus.com/pub/ASUS/mb/..._AM1I-A.pdf#15



Wattage reading does not work even in 3.15-rc1 kernel as i see , but they added support for Mullins - "Kabini support added in kernel 3.12. Mullins support added in kernel 3.15. "

kabini is single channel, so memory placement in this case should not affect anything.

[dungeon]

kabini is single channel, so memory placement in this case should not affect anything.

Yes i know that . It also doesn't matter even if it is dual channel, because only one module is used and caution is about that in manual . So your statement that kabini is single or dual channel in this case doesn't matter at all .

Maybe to translate what manual says: if you use only one module, put it in A2 and not A1 .

[AJSB]

kabini is single channel, so memory placement in this case should not affect anything.

So, why the manual says otherwise ?

[chithanh]

My uneducated guess would be that signal quality could degrade due to reflections at the end of the memory bus.

http://en.wikipedia.org/wiki/Signal_reflection

[dungeon]

So, why the manual says otherwise ?

They always says for one module to goes to A2 . If you have dual channel, first goes to A2 second in B2, they never recommends A1 only or A1 and B1 only for dualchannel, for the example:

http://dlcdnet.asus.com/pub/ASUS/mb/LGA1150/H87M-PLUS/E8528_H87M-Plus.pdf#23