Overclocking Sadness (FX-8350/GA-990FXA-UD3 Rev. 3.0)

If Gigabyte were a person I could punch them right now.

So I've been trying to OC my FX-8350 and the method I wanted to use was via FSB but for some reason anything past 210mhz causes BSOD a minute or so after loading to desktop and raising voltages did nothing at all it always BSOD. I've given up FSB for now as I have no clue what to try next, however at 210mhz it's stable so I put it there and used the multiplier, right now I'm at 4.62 GHz @ 1.42V.

If it wasn't bad enough that I was having massive stability issues via FSB I also have to deal with voltage drop MASSIVE voltage drop at load it will go from 1.42 to .9 and then back and forth. Yesterday I spent a good chunk of time going through an owners thread on OCnet and found out my particular board doesn't have the option to disable APM which is what is causing the massive throttling. It seems to be just for us FX users though as apparently enabling HPC fixes the throttling for other CPUs. Another user also posted an email from Gigabyte saying a BIOS update with the feature should be released in a week or so and I really hope it is. Overall the BIOS is really weak it's missing a lot and there are a few important settings that can't be changed statically which really blows I dislike offsets it doesn't allow us to be precise which makes fine tuning sorta impossible. 

If you have any suggestions that could help with my FSB issue feel free to post and I'll give it a shot.

Dude, even the wildest overclocking phantasy can't override Ohm's Law. 6.2 GHz at 1.42 V means your FX-8350 would be drawing about 200 W of power lmfao

Ha, shit I didn't even notice that lol I meant 4.62GHz of course lol.

don't move the fsb up, use the multiplier. You might get better results that way. Trust me on that

No worries, I just thought it was funny.

The FX-8350 is a bitch to overclock in the same way as the Phenom II X4 or FX-6300, it draws a lot of power and runs very hot very quickly. To be honest I wouldn't overclock an FX-8350 for a customer to more than 4.6 GHz on a Noctua NH-D14 unless I knew for sure that machine was going to be in an air conditioned cool room all of the time.

It has 33 % more power consumption and heat emission than an FX-6300, and 40 % more than a Phenom II X4 965, and that puts a lot of stress on the chip itself, on the motherboard (especially the power delivery circuit), on the PSU if it's not really beefy, and on all the other parts if the case doesn't have extreme airflow.

I've also noticed that the FX-8350 has burst heating habits, it will heat up very quickly very fast for a short time, then come back to a more stable level. I don't know what causes this, probably some internal thermal management, but it causes a just-stable overclock to break down. I don't think a BIOS update will solve all the overclocking issues of the FX-8350, and believe me, that Gigabyte board is a really good quality board.

Like Skull says, the safest way to overclock an FX processor (any FX processor for that matter), is augmenting the multiplier. You can use AMD OverDrive to let the system augment the multiplier automatically to test when it crashes, but on an FX-8350, after rebooting after the creshtest, don't come down 0.1 GHz from the crash fequency like on other AMD CPU's, but come down 0.2 GHz, because an FX-8350 will not be stable for prolonged periods on maximum CPU load at 0.1 GHz under the crash frequency. With AMD OverDrive, if you have really good cooling, it will probably crash at about 5.1-5.2 GHz, and a safe overclock at that point would be 4.8 GHz, but I would stick with 4.6 GHz for a bit of thermal margin and a bit of electrical energy relief on the VRM, because in my experience 4.6 GHz is often reached without voltage boost and 4.8 GHz isn't, so the amount of electrical energy required by the CPU goes up exponentially once you have it set to a clock speed higher than 4.6 GHz.

The FX-8350 is a vanguard product SKU, it has a much smaller silicon than the Opteron CPU's, but still has 8 fully functional instruction cores at a base clock frequency of 4 GHz, and at 32 nm that's stretching a few practical principles. That's what I like about AMD, they keep pushing the envelope at world wide market level, even though their products need much more silicon and circuit complexity than Intel or nVidia's, and as a consequence they lose money with every product they sell, but they don't give up.

different results for me

any multiplier changes resulted in odd performance in games and some games not even running at all (borderlands 2).

altering the cpu bus however allowed me to reach 4.6 (stable) easy

All I'm hoping for from the BIOS update is for an option to disable APM so overclocking is even worth it.

Oh and my CPU is water cooled, I'll probably add more to the loop later as well.

Buy cheap overclocking mobos you get a cheap mobo

What? What are talking about the board is good the BIOS just needs updated.

Don't worry about the FSB, just up the multiplier out of the ass. I have my mult set to x26 and the FSB to 200. Complely stable on a 6100, can't imagine it's that much different on an 8350

I've had my FX 8350 as high as 4.75 GHz on a GA-990FXA-UD3 Rev. 1.0

FSB was 225-230

FSB voltage was @ 1.150v (+.050)

CPU multi to set the CPU to 4.7-4.75ish @ 1.5vcore

CPU NB voltage was @ 1.2875-1.3125

FSB and HTT multi set to 11 or in the 2475-2510 range

HTT Voltage was @ 1.25v (+.050) 

RAM multi was set to be around 1800 or alightly less @ 1.55v


I also went with slightly looser RAM timings which seemed to help stability

Started at 4.4GHz @ stock voltages and slowly worked my way up, increasing multis then voltages back and forth until a 15 min stability test passed with no errors.  These short tests were strictly to see if the multi and voltages were close to the right ranges as I slowly worked my way up.

Cooling was on an H100i and i stopped once it started to linger over 56-57c under full load

The NB and VRUs didn't seem to get too hot, as in I could touch them immediately after the stress test and they were warm but not too hot to keep my finger on them.  I have really good backside board cooling and a spot fan on the NM and VRMs.

It's tricky, challenging, but do-able if you take your time, take notes, and slowly work your way up while checking all the important temps, including chipset and VRMs.  Keeping the ram below 1800 seemed to help the NB temps a lot.

PS - To back Loagan up, I didn't see a significant performance difference going the FSB/HTT route as opposed to strictly using the multiplier.

I did it that way because I wanted to make smaller jumps up, in the 50MHz range.  Going multi only you have to make 100MHz jumps and that is more likely to cause BSODs if it's too big a jump without a voltage bump.  Since I'm breaking new ground for myself OCing this CPU I opted for the more granular, 'smaller steps' approach.  ;)