This is a guide I did as a writeup of all the features of my Pro4-M FM2 board I was rigorously tweaking for a build for a friend. I did plenty of research on what each feature does, and it should apply to anyone using the Asrock UEFI firmware on an AMD platform, and they might even translate to intel ones. This focuses on manual overclocking and gives insight to all the important settings in the tweaker / advanced panels.
- EZ OC Mode: either manually configure overclock options or chose preset overclocks. The presets from my experience are overzealous with voltage and reset RAM preferences. They are designed to be "safe" but be careful using the highest presets because they put the voltage at 1.4 + offset of .2 which can put you over safe operating frequencies under load, especially if you use load line calibration which can skew voltage even higher. That can get really dangerous in terms of processor burnout. So I recommend going manual and tweaking individual settings.
- Overclock Mode: manually tweak the fsb frequency or leave it default and allow spread spectrum. Spread spectrum prevents voltage spikes on the various system buses off the southbridge, and overclocking the pci bus is often dangerous because it can fry add-in cards or cause hardware transfer rate errors since it also affects the sata and usb interfaces. You never need to overclock the PCI bus and spread spectrum is good at rated clock rates, so auto overclock mode and auto spread spectrum.
- AMD Turbo Core Technology: enable the cpu boost feature when only one core is maxed out. Even if you do a hard overclock, you might want to try this, because it only becomes active when the other cores are underutilized and the heat generation off one core going about 400mhz faster usually won't kill a system. I enable it, and use the aformentioned 400 mhz above full utilization clock.
- APM Management: an enhanced version of cool'n'quiet that supports per-core downclocking and undervolting in a more granular and sophisticated way. If you use cool'n'quiet, you want this on. The tradeoff here is that by powering down underused cores revving them back up isn't instantaneous and a lot of people have misplaced fears of the voltage shifts wearing out the chip faster. Modern chips are much better designed for undervolting to save power and the only time to heed caution here is if you have an absurd Vcore. Leave it on to save juice.
- Multiplers: You want this manual, else you are not really overclocking. The goal is to get the base clock multiplier as high as possible while staying stable and with manageable heat and voltage rates. One unmentioned aspect in most cpu stories is that if you downclock a processor enough you can drastically cut the wattage usage by dropping the voltage to match - if you took an A10 that usually rates at 1.325 to 1.375 volts down to 1.25 or 1.2 volts and find a stable clock rate (probably around 3ghz) you could easily see wattage and heat output cut to around 60 - 65 watts, maybe 90 watts under load, compared to the stock idle 100 watts and 125 under load.
- Vcore and offset: Boosting voltage boosts overclock headroom for stable clocks but also boosts heat and power consumption. You want these as low as possible for a given clock rate while keeping the system stable at its maximum frequency. Generally, the given configuration is overgenerous and excessive and it should be tweaked through torture testing to find a stable vcore at your desired clock rate as low as possible. You should actively try to adjust the offset rather than the vcore, because most power management functions operate as a fraction of vcore, so overclocking it reduces power efficiency.
- CPU NB Frequency: this is the northbridge data rate. You want this as high as possible while remaining stable for maximum bandwidth between the memory and cpu, and the best way to know if it is excessive is if you get hardware errrors on memory transactions that don't show up in memtest. In my case, I was able to get total system failure at 2600mhz when running SIMD instructions off the gpu, because those are usually more bridge intensive than CPU based memory operations and want a lot more bandwidth. Unlike some platforms, you can't explicitly set the northbridge or cpu base frequency, which is really a good thing because both are integrally tied to a lot of underlying parts that can easily break if tweaked.
- CPU NB/GFX Voltage: The gpu cores and northbridge have their own voltage. As per usual, you want this as low as possible while maintaining a stable system with a given nb overclock and gfx overclock.
- APU Load-line calibration: This is used in modern cpus to offset vdroop. Only use load line calibration if your voltage isn't dangerously high - traditionally, your given voltage is a fixed limit that the processor won't pass, but may dip below (droop) especially under constant heavy load. LLC will up voltage to offset droop, which means your rated voltage isn't your limit anymore, but you have less chance of voltage dropoff under load. So use LLC at low overclocks but avoid it with higher ones where the voltage spike it causes puts the cpu over safe limits.
- Memory Profile: JEDEC profiles are mandated for DDR and are base profiles meant to work on anything of a certain DRAM spec. You want to use the XMP profile as the template the manufacturer meant the RAM to run at. In my case, my Corsair Vengeance sticks were giving the wrong XMP RA tRAS value and I needed to tweak it from 30 to 27. Make sure to try at your memories given ratings and overcloclking the CAS and such latencies usually isn't worth trying.
- DRAM Frequency: has preset frequencies that go with the standard. The rated frequency all Piledriver based chipsets are rated at is 1866, but they all support overclocks up to at least 2400. For example, I had 2133 Corsair RAM, so set this to the rated speed of your ram. It is always a tiered thing, and the only way to really fine tune adjust dram frequencies is to overclock the northbridge base frequency, and you really take a lot of risks if you do this, even if you can do this. My P6T can, but my Pro4-M can't. You might for laughs try putting RAM modules rated for something like 1866 at 2133, but most likely that will error out in memory tests, especially if you don't try adjusting the latencies and voltages upwards to compensate.
- DRAM Voltage: Part of the XMP spec is the rated voltage. You want this as low as possible without memory errors. Undervolting memory usually isn't worth it much because memory modules don't generate much heat (anymore, at least, old DDR2 and DDR ram was much hotter due to higher voltages) so stock rates are usually sufficient unless you are a real power nutter. If you get memory errors in memtest or SIMD transactions that aren't related to northbridge frequencies and voltages, it might be an issue with your RAM voltage. For example, those Vengeance sticks I was talking about would error out at 2400mhz northbrige at stock 1.5 volts but a 1.505 voltage was sufficient to avoid those errors.
- APU PCIE Voltage & SB Voltage: Keep these constant at 1.208 volts and 1.1 volts respectively. You don't overclock your PCI bus or SB buses so don't overvolt them unless you are a really crappy power supply that has bad rails. And if that's happening, you have bigger problems anyway. So don't touch these unless you have explicit SATA or PCI device failures, voltage drooping, or performance degradation from too little power on these lines. And like I said, that really shouldn't happen.
Advanced CPU Configuration:
- Core C6 Mod: Enables the lowest power state of the processor (deep sleep mode). This enables S3 suspend to completely shut down cores at 0v rather than leaving the processor in a low power halted state. Leave it on, modern systems support it fine and it is designed not to be unstable. If you need to turn this off your motherboard or cpu are pretty much broken.
- Cool 'n 'Quiet: The precursor to APM, also supports memory downvolting. If you use APM you want this on, if you don't use APM you might still want to use it because it is less aggressive. AMD has supported CnQ for 4 processor generations now and it is a very mature power saving standard that won't burn out your chip by alternating voltages.
- SVM: The AMD hardware virtualization support. Keep it on, it allows you ti run hardware accelerated virtual OSes and doesn't cost anything. This option is only here to prevent virtual guest operating systems.
- CPU Thermal Throttle: This is the fraction of power to take the processor to when in S1 ram suspend mode. You want this as low as possible without causing instability, and lower rates can make resuming slower because it has to restore more voltage difference. Since most systems use S3 suspend as the default mode (power down everything except ram, instead of S1 keep everything on in low power mode and store everything in ram) you don't need to worry about this much, and auto is just fine.
North Bridge Configuration:
- Primary Graphics Adapter: If you have a discrete card and are not using hybrid crossfire, you want this on to use your discrete card instead of the integrated gpu. There is practically no use case for this though, except maybe in laptops where you power off the discrete card unless you have something graphically intensive. But modern GPUs have really good power saving modes anyway. You won't need to change this unless you know you need to change this.
- Share Memory: Amount of system ram to allocate to the integrated GPU. The way integrated GPUs currently work is that they deny the OS access to a portion of installed RAM and use it as a substitute VRAM. Since RAM is cheap, set this to the max with any APU unless you are running a system with little RAM (why?).
- Onboard HDMI HD Audio: Leave this on to support the HDMI port also sending digital audio. Duh. You want this if you plug in a tv or display with speakers via HDMI, but there really is no reason to turn it off. If you don't have HDMI audio capable hardware, it just won't play anything. Turning this off just gets rid of the audio device the OS sees for HDMI audio.
- DVI Function: DVI cables can operate in an emulated HDMI mode. So if you use an HDMI converter on the DVI port, use HDMI mode, and if you use a regular DVI cable, use Dual Link.
South Bridge Configuration: These are case by case and self-explanatory options for the most part. The options I have are for the front panel audio controller, the lan controller, and led controls. Enable or disable features, mostly.
- SATA Controller: You probably have hard disks connected. You want the SATA controller on. You might have an ATA controller here if your mobo supports it, turn it off if you don't have an ATA disk (the old slow crap from the 90s).
- SATA Mode: Traditionally you have 3 choices here. Hardware raid, ahci, and ide. IDE is legacy and slow, ahci is modern for multiple discreet disks and is probably want you want, and if you want to setup a RAID you should know what you are doing in advance. AHCI provides a lot of benefits for monitoring and performance.
- AMD AHCI BIOS ROM: This option loads a rom image into ram before starting anything containing legacy AHCI tables for backwards support. Most modern operating systems don't use or care about these, but if you need them they are there, but you would find out elsewhere if you do. You can try running it to see if you get more sensors out of it, but most OSes can detect AHCI 2 sensor data without the rom load or extra boot time.
- SATA IDE Combined Mode: Allows your "last" sata ports to act like a combined IDE port for PATA devices if you use a translation cable. Almost certainly you don't need this because you don't have PATA devices, and if you did, you would have gotten a mobo with a PATA port.
- Hard Disk SMART: If the AHCI ROM is enabled, you always have smart data available. If you don't use the rom, turn this on always to enable smart monitoring. Smart is for ATA devices to give you important diagnostics and testing for hardware integrity.
Super IO Configuration: usually these are the enables / disables on the backplate. The only thing to worry about are IRQs and addresses, IRQs are interrupt handlers, you want all of them on so if you use devices they provide interrupts as intended in certain circumstances, addresses are hardware locations where these devices are referenced. Make sure they don't overlap (if you uefi even allows it).
Continued in next post!