Overclocking The AMD AM1 Athlon & Sempron APUs

[phoronix]

Phoronix: Overclocking The AMD AM1 Athlon & Sempron APUs

With the initial Linux tests of the AMD Athlon 5150 / 5350 & Sempron 2650 / 3850 out of the way, I ran some basic overclocking tests on all four of these week-old AM1 APUs.

http://www.phoronix.com/vr.php?view=20171

[schmidtbag]

Considering the minimal speed bump and the short pipelines of the architecture, I'd say the performance improvement was pretty good. But, you should've been able to get higher than that. You'd get a lot higher if you lowered RAM frequency and increased voltage.

[Michael]

Considering the minimal speed bump and the short pipelines of the architecture, I'd say the performance improvement was pretty good. But, you should've been able to get higher than that. You'd get a lot higher if you lowered RAM frequency and increased voltage.

I tried lowering the RAM frequency to DDR3-1333MHz as well as bumping up the voltage (up to 3 steps) to no avail. I did get it to boot a few times at 107~110MHz but would reset itself usually a few seconds after getting back into the UEFI setup.

[dungeon]

The same 7% OC experience i have with AM1M-A board and Athlon 5350 i've running it @2206MHz with no problem. So just to mention, both Asus AM1 boards can be overclocked .

Why oh why wattage sensor doesn't work

[dungeon]

Michael you mentioned fglrx tests coming... so please turn on HyperZ for radeon if you test on 1920x1080 resoultion, it makes a difference in that case .

But if you test on lower like 1024x768 dont worry, there is not as much difference there .

[grndzro]

I cant find how many power phases that motherboard has.
It is kind of silly not to say how many phases but include overclocking.

It might be a factor in overclocking.

[mczak]

I cant find how many power phases that motherboard has.
It is kind of silly not to say how many phases but include overclocking.

It might be a factor in overclocking.

I highly doubt that given the quite low wattage.
Pretty sure it's just the reference clock which limits because everything depends on it. Wondering what the "true" clock limit of the cpu cores actually is - but with (almost no) multiplier adjustments on locked chips, no unlocked ones and just minimal reference clock adjustments there's no way to find out...
I think gpu overclocking would be quite interesting too, after all the clock is only 600Mhz for the fastest part, whereas discrete graphic chips using near 100% identical logic easily reach 1000Mhz (on the same or very similar 28nm process too). But looks like that's not possible neither :-(.

[dungeon]

Michael you mentioned fglrx tests coming... so please turn on HyperZ for radeon if you test on 1920x1080 resoultion, it makes a difference in that case .

But if you test on lower like 1024x768 dont worry, there is not as much difference there .

I installed fglrx in Lubuntu . and compared it with radeon in Debian with hyperz enabled . At 1920x1080 resolution In Openarena 0.8.5 i have 79.4% performance of fglrx and in Openarena 0.8.8 that is 86.3% - pretty good results .

[dungeon]

I installed fglrx in Lubuntu . and compared it with radeon in Debian with hyperz enabled . At 1920x1080 resolution In Openarena 0.8.5 i have 79.4% performance of fglrx and in Openarena 0.8.8 that is 86.3% - pretty good results .

Ha without Hyperz enabled in both cases we are at ~56% level .

[Luke]

DDR 3 is touchy about both voltage and timings

I tried lowering the RAM frequency to DDR3-1333MHz as well as bumping up the voltage (up to 3 steps) to no avail. I did get it to boot a few times at 107~110MHz but would reset itself usually a few seconds after getting back into the UEFI setup.

I've actually played with the same CPU (a Phenom II x4) in both a DDR2 and a DDR3 motherboard. The older 785 chipset would allow CPU give power management and a higher multiplier to be used together, but you could run the base clock up to about 236-238 MHZ to overclock from 3.2GHZ to almost 3.8 GHZ, in the process overclocking DDR2-800 to almost 1000MHZ on the same voltage and timings with no issues.

Put the same chip in the FX 990fx board, and the base clock could not go over about 110 MHZ if even that without instability. On that board the unlocked multiplier could be used w/o losing power management, so this was no big deal. Same RAM with that chip and my FX8120 CPU, it will go as high as DDR3-1833-but NOT without selecting a complete DDR3-profile with not only the corrrect voltage but the correct timings for this! It is DDR3-1600, and will error out even at stock clocks if a DDR3-1333 "profile" of voltage/timings is used at 1600 MHZ.

Therefore, it seems to me that all of these boards would require first selecting a higher DDR "profile" (multiplier/timings/voltage), then manually reducing the RAM multiplier before a base clock overclock could be used effectively. This would bring the RAM back to the high profile's starting point and bring up the CPU clock. Haven't tried this as the unlocked CPU multipliers on all my CPU chips allow setting the CPU and RAM frequencies one at a time, a better approach when available it seems to me.