Choosing motherboard for FreeBSD (AM3+)

If you like to buy a ASUS mainboard always check the QVL list for the model.
Asus Boards are known RAM divas.
For all other vendors it is also a good idea.
 
Issue is not as much in Asus boards, but in AMD memory controller. Somewhere in AMD support side is interesting table I could no longer find. It listed AM3(+) CPU's and described which is the maximum supported speed when you are using 1,2,3,4 sticks, one-sided or dual. It also gave default supported speed of RAM for different CPU's. All I can remember of it is that FX8350 memory controller could work with 1866Mhz RAM stick until you used 1 stick. When you had 2 sticks then you had to use 1600MHz sticks. With 4 sticks, RAM speed reverted to 1333Mhz. That's what AMD guaranteed would work although on different boards you could get different results. Worse or better. With Athlons and such, those speeds were much much slower. Even 1066Mhz rams.

Also, Asus AM3 boards tend to have decent VRMs, so you could install FX8350 without issues. Gigabyte I would not trust in that regard. Some of their AM3 boards took bunch of revisions before they became more-or-less-stable with a powerhungry FX riding on top of it.
 
aht0 said:
It also gave default supported speed of RAM for different CPU's. All I can remember of it is that FX8350 memory controller could work with 1866Mhz RAM stick until you used 1 stick. When you had 2 sticks then you had to use 1600MHz sticks. With 4 sticks, RAM speed reverted to 1333Mhz.
Click to expand...
Memory clock is regularly overestimated. It's more of a marketing trick that attracts overclockers and creates the illusion that the actual system speed would increase.
The decrease of memory clock is physically caused. Each additional module introduces additional capacitances and inductances reducing the maximal achievable bus speed. This is why normally the "marketing speed" of 1866 MHz can be reached only if one single module is installed.

Actually such "fast" systems with a single module have substantially slower throughput compared to slower-clocked systems with multiple modules, which can use interleaving, which boosts memory bandwidth much more.

Maybe the simplest non-technical, extremely simplified way to describe this would be: 4 modules with 1066 MHz working concurrently in parallel have about double the throughput of 1 module at 1866 MHz.
 
You must log in or register to reply here.
Back
Top