Graphics Card Overclocking

Overclocking a graphics card is a great way to unlock higher framerates. In many cases, you are able to increase the performance of a GPU to that of the next better model. For example, AMD 6950 uses, basically, the same hardware as the 6970, just with a couple parts on the GPU core disabled and its at lower clocks. Well, if the hardware is the same then it should be able to hit the same clocks as the 6970, right? In many cases, this is true. For example, the AMD 5870 has a stock clock of 850 mhz, while Im able to get my 5850s to 870 mhz on stock voltage. Looking at the hard data, a 5870 has a 68 Gigatexels/second fill rate and a 153.6 GB/sec memory bandwidth. My 5850s are at 62.6 Gigatexels/sec and 153.6 GB/s (from GPUZ). At the same time, brand new, my cards were $150 cheaper. Pretty good argument for overclocking, no?

A word of warning: As much as overclocking is supported, it is not officially supported. This means that should you damage your card and send it in for a replacement, and the support tech staff is able to determine that you overclocked your card, then your warranty is void. OCing through software makes it very difficult for them to determine this, whereas a flashed BIOS makes it extremely easy.

Ok, so, regardless of why you want to overclock, lets get into the basics of it.

Key Terms

Core means the actual GPU core, where most of the magic happens. The speed, in mhz, that your GPU core is running at is going to greatly affect the texture and pixel fillrate performance, which is a huge bottleneck in gaming.

Im not paid for complete testing ;)

Memory (VRAM) is where textures are stored as well as rendered frames. There are plenty of performance benefits to having a higher memory speed, considering that pretty much everything you see on screen is at one time in the memory.

I ran some tests on FurMark, at 1920x1080, showing how both core speed and memory speed can affect performance. As you can see in the graph, the framerate went up as the memory speed increased, with barely any change from core speed. It is important to note that this particular test was run using 4xMSAA. I also did some tests at 0xMSAA in FurMark, and it showed that only core speed had an improvement. Basically I just want to point out that for optimal overclocking, you need to focus on both memory and core speeds.

Voltage is what gives you stability. As the core speed increases, it may need an increase in voltage to remain stable, much like in CPU overclocking. Most higher end GPUs allow voltage modification, however, not all GPUs allow this. In some cases you might be able to also modify the VRAM voltage, but this is fairly uncommon. If you can, then in much the same way as core voltage, it can help you achieve higher clocks.

VRM is the Voltage Regulator Module. This is what adjusts your input voltage to that going into your GPU. These are very prone to over heating when overclocking, especially if you increase your voltage. 

How To Overclock

MSI Afterburner

The first things youll need is a program to adjust GPU clocks. AMD offers Overdrive in the Catalyst Control Center, and many manufacturers offer their own tweaking programs, like Sapphire Trixx, MSI Afterburner, and EVGA Precision. Personally, I use Afterburner - with my Sapphire cards. It is probably the best all around program, and the only reason not to use it is if you have non-reference, non-MSI cards and want to adjust the voltage.

GPU-Z Sensors

You will also want to download something to monitor your cards. Afterburner takes care of this, but you should also consider getting GPU-Z because it monitors more than just the core temperature. In most cases, you will also be able to see your VRAM and VRM temperatures as well. Ideally your temps wont surpass 80-90C at the absolute most. For my 5850s, I found I would get a hard system crash or lockup when VRM or VRAM hit around 120C, although my core temperature was still in an acceptable area - so keep in mind that you should check on those temps too.

Before you start, make sure that you do NOT have "apply overclock at startup" selected!

The basics of GPU overclocking are very similar to CPU overclocking. Personally, I recommend starting with Memory clock because it gives nice gains and often isnt affected by voltage - since usually you can only change core voltage. To overclock the memory, I like to start with 50mhz jumps, with 5 minutes of FurMark testing at 4xMSAA at full screen - also, let the cards stay heated up a fair bit rather than giving them a long cool down time between tests. You may want to set up your tests using the benchmark feature. Continue this until you get either a lock up, crash, artifacts, or the screen goes blank. Doing a hard reset on your PC isnt fun, but it shouldnt wreck anything. I usually find that ctrl+alt+del still works and you can reset from there.

FurMark stress test

When running your 5 minute benchmark tests, you will get a score at the end. As you increase the memory speed, you should see an improvement to the score. However, at some point even if you dont get a crash or lockup, you might notice your score either didnt increase, or worse yet, decreased. This is a vital thing to notice. The reason for this, as I understand it, is that GDDR5 memory - that is, the most common memory used for GPUs - is meant to be crash resistant. It does this by being able to resend data that failed the first time due to being unstable. In this way, if you are nearing the point of crashing, leading up to it the VRAM will need to resend data repeatedly, and therefore it will reduce performance compared to every piece of data being sent only once. Therefore, if you notice this occuring, slowly reduce your clocks by 5mhz until you realize the highest score.

Moving on to GPU core overclocking, the process is much the same. You can still start with 50 mhz jumps, but with the testing you should run at 0xMSAA in FurMark. Also try to keep temperatures high throughout the testing phase.

The most common issue of an unstable core is artifacting. This is when pixels essentially get stuck, and youll notice multiple squares on screen that dont sync up with the image. At this point you will want to stop the test and will also have to reboot the PC. You have a choice here, either to reduce the clock or else increase the voltage. For 24/7 overclocks, I highly recommend reducing the clock and finding the highest stable clock on stock voltage. However, if you want to increase the voltage, I suggest doing it in 0.05V increments. Basically just increase the voltage and retest at the same clocks, and repeat until stable. It is also vitally important to watch all your temperatures, because as I mentioned before, its possible that you crash because of heat and not instability.

Extra Info

There are, as always, a few other things to consider. FurMark isnt perfect in determining if your GPU is stable. Ive passed fairly long sessions in FurMark, as well as a few games, before finally having an issue in one specific game. It didnt take long to realize that my GPU OC was to blame, and dropping my memory clock by only 5mhz made it nice and stable from there on out. So, actually playing games will be your final stress test.

If you run crossfire or SLI, there might be issues with instability. Generally it should be ok, but Ive read about issues with crossfire that has to do with the switch between low power and high power, and only occurs when a custom voltage is set. It seems the way around this is to disable ULPS, which you can google for yourself.

Another area of problem is that sometimes after an overclock is applied, the graphics cards wont automatically change to low power state at idle, or else the idle state clocks are higher than they used to be. This is something that is at the mercy of people making the overclock programs, and the only way around this (and other) issues is to manually reflash your GPU BIOS. This is not a recommended practice, but I will just put here that if you have a very stable overclock, it is a handy thing to do because your GPU will essentially think that the overclock settings are its default settings, so it behaves like a totally stock card. There are tools for changing and flashing the BIOS, and for more info check out techpowerup.com.

If you have concerns about your particular brand or model of graphics card, just google your card and key terms like "overclocking" "highest clocks" "stability" to find a variety of sources. I always advise learning as much as you can before doing something you dont quite understand, and causing damage.

Read More..

Beginners Overclocking Guide Part 3

Testing

The first step to overclocking is to download appropriate apps to test your settings. Theres a large variety of programs available, so Im just going to list a few here.

First of all youre going to need system health monitoring programs. You can try:
HWMonitor
CPU-Z
Real Temp
Speed Fan

GPUZ, HWMonitor, and Real Temp

Secondly, youre going to need some stress tests. My favorite is IBT, but heres a few options:
Intel Burn Test
Prime95 32bit / Prime95 64bit
OCCT
Memtest86+ Bootable CD

Testing is actually fairly simple. Id suggest running HWMonitor and Real Temp at the same time, to make sure you have accurate sensor readings. Leaving these programs open to monitor temperatures, run a stress test. Generally speaking, the more RAM you use, the more stability will depend on RAM. Specifically with IBT, you can set a low RAM usage (standard) with a very high amount of passes (20-30) to check almost exclusively CPU stability. If CPU passes, then it is recommended to do at least 10 passes on Maximum stress level to make sure everything is good.

Prime95 is much less sensitive to errors than IBT, but it is still a good test to run. However, it takes quite a bit longer. It would be a good idea to run it overnight, as it is typically recommended to get an 8 hour test without errors to be almost guaranteed stable. This, obviously, should be the last test you run as it is the most time consuming.

With any CPU, there is a thermal limit you shouldnt go too far over. Typical for Intel CPUs is 72C. AMD CPUs are a little different, but overall if you keep a CPU under 70C you should be ok. If stress tests are below 80C, it might be a little bit of a risk but you can still call that acceptable because under normal loads like in gaming and encoding videos, the temperature should be lower - just make sure you verify that.

Overclocking

Ok, so you should know understand most of the important options in your BIOS, how they are related, and how to test an overclock once you make changes. But what changes should you make? This will depend highly on your system, so I can only give some more general advice. Remember that OCing is more of an experience than anything. Take your time with it, dont jump too high very quickly, and always, always test.

If you have an unlocked CPU, this becomes rather simple. First and foremost, set your RAM to manufacturer specifications and test it - without overclocking the CPU. This is important to make sure your RAM is ok. Once youve done that (lets say 10 IBT passes at Max usage - or at least 2 passes of Memtest86+) you can start increasing your CPU multiplier. I would suggest not making more than 400mhz jumps above stock up to 4ghz, and not more than 100mhz after that. Test after each change.

If you have a locked CPU and need to start increasing the FSB/BCLK, overclocking gets much more complicated. The first thing to consider, which I mentioned previously, is that the speed of your RAM is affected by the BCLK. This of course poses a problem, and is one reason why overclockers like to buy fast RAM (1600+ MHz). For now, the RAM will have to be set to a lower than rated speed to ensure stability, while you overclock the CPU. Often, 200 BCLK or less is easily acheivable, so consider what values you can use to eventually get your RAM to 1200, 1400, 1600, or higher speed (if rated for that). Well worry about actually hitting those values later. Much like for the unlocked CPUs, you dont want to take huge steps in speed increase. Guide Part 4

Read More..

Beginners Overclocking Guide Part 2

Some Terminology

There are a few things you need to know first. This can be rather tedious and, for the beginner, daunting. I assure that no matter how complicated it looks now, once you enter the BIOS and have a look for yourself, and start to change some values, it should all start making a lot more sense.

Example of BIOS voltages

Voltages

The first thing Ill be addressing is Vcore. This is the CPU core voltage, and is one of the biggest influences on CPU stability. All CPUs have manufacturer specifications on what the highest Vcore the chip is rated to handle, so make sure to check the documentation online before you start. As a general rule for Intel CPUs, 1.4V is the max you want to go, but some high end chips are ok up to 1.5V.

Secondly, Intel has what is sometimes called IMC voltage (Integrated Memory Controller), other times called QPI (QuickPath Interconnect), and yet other times VTT. On AMD rigs, you have a Northbridge Voltage - the NB is pretty similar to the IMC (and in a way the QPI) in that they both handle a lot of information flowing into and out of the CPU and back and forth to other components on the motherboard. Increasing this voltage along with Vcore will help stability. On Intel rigs, I generally keep this around 0.1 to 0.2V below the Vcore value, but that is not a "hard and fast" rule, just a preference. There is one hard rule for IMC/VTT/QPI voltage, though: it must be within 0.5V of your DRAM voltage.

DRAM Voltage is directly related to your RAM. RAM should always be set to the recommended manufacturer speeds. For example, you might have bought a 2x2gb 1333mhz CL8 1.5V RAM set. The 1.5V designates the DRAM voltage needed for that speed (1333mhz) and timings (CL8). Ill explain more on this later in the nitty gritty of OCing.

There are a few other voltages you can probably tweak. In the example picture, that is my BIOS. PLL means "phase-locked loop", and PCH "platform controller hub". The PCH refers to the "south bridge" which is where many information sources have to go through - for example most PCIe 1x slots will go through there as well as USB and SATA hubs. The PLL voltage rarely has to be changed, maybe 0.1V increase at high base clocks. Same goes for the PCH - although in my case I have a PCIe 4x slot that runs through it, with a GT 240 installed to run PhysX. I found that by increasing my PCH voltage to 1.15 helpled stabilize the GPU.

Clocks

Base Clock (Intel) and Front Side Bus (AMD) are quite similar. The BCLK is what all other speeds are multiplied from to get their final speed. So your CPU might run at 22x multiplier - this means 22xBclk. RAM, too, is multiplied by the BCLK. Many Intel CPUs have a 133mhz base clock - so 22x133=2.9ghz. 10x133=1333mhz RAM. The QPI is also a multiple of the base clock. Front Side Bus, or FSB, works in much the same way although usually they use ratios instead of multipliers. This can be a little confusing, but when I say "change the multiplier" consider this to also mean the ratio if you have an AMD rig.

The PCIe clock is a bit special. I have a first gen "i" core - an i5 750 - in which the PCIe bus is unrelated to the base clock. However, as I understand it, this is not the case for the current Sandy Bridge intel cores like the i5 2500k. I have mentioned PCIe clock before in a previous blog post, but suffice it to say you shouldnt touch this right now. I will delve into this deeper at a later time (links will be provided at the end).

CPU Frequency is, of course, the speed your CPU runs and this is the value you are trying to increase. Only a few years ago, 4ghz was considered "the top" and going beyond that was "insane". On any current generation CPU, 4ghz is actually fairly easy to achieve. That isnt to say all CPUs will make that number, though - bear in mind that all hardware is slightly different and has different potential for overclocking. You need to be slow and cautious to get the most out of your hardware.

Feature Terms

Theres a lot of features present in most BIOSes. Many of them you dont need to worry about, some you do. Some are completely optional or unrelated to overclocking. For example, the ASUS RoG series of motherboards (extremely top end) have all sorts of features you can access - like being able to overclock from a smart phone. Im not going to touch on these. Instead Ill focus on what is primarily related to CPU stability when overclocking.

What not to do with your PCIe frequency

On the previous picture, youll see one very important term - Load Line Calibration (or LLC). What this means, is that the motherboard will try to counter act the natural drop in Vcore once a load is applied to the CPU. This drop actually has a term - called VDroop - and it is in fact a specification made by Intel. Im not sure if AMD has this or not. One important thing about VDroop is that when an Intel CPU is running with Turbo Boost, it can modify its multiplier on the fly based on how many cores are needed. If only 2 cores out of 4 are required, the CPU will run at a higher frequency. This can also require a higher voltage, so with LLC off (VDroop on) the voltage will naturally be a bit higher because there is a smaller load (2 cores arent as power demanding as 4). However, when overclocking, it is common to disable Turbo Boost. In this case, enabling LLC can be beneficial to getting a stable overclock. Its not required unless going for close to or above 4ghz. There is a note of caution here, technically speaking when the Vcore fluctuates and the motherboard tries to compensate, it can for just a fraction of a second apply an overvoltage which may damage your CPU. This is extremely unlikely, but remember that the quality of your motherboard plays a huge factor in this.

CPU Ratio I have already hit on, but I want to mention a couple things. Most CPUs can go as low as 9x multiplier. The highest they can go will depend entirely on the CPU. My i5 750 tops out at 21x, the i5 760 at 22x. Both AMD and Intel have a series of CPUs that are "unlocked". From AMD, they are the Black Edition (BE) CPUs and from Intel they are the "K" series (2500k). What unlocked means, is that the multiplier is unlocked and can be raised to a very high number - like 60+. This makes overclocking significantly easier as you dont need to worry much about changing the base clock, which also means no worries about it interfering with your RAM and QPI.

DRAM Frequency is set as a multiplier of the BCLK. Generally on Intel, you only have 3 options: 10x, 8x, 6x. This means that to hit your RAMs rated speeds can require a specific BCLK. If you need to change the BCLK, keep this in mind. As you see in the image, my BCLK is at 200 which makes it really nice to hit my 1600mhz RAM speed with the 8x multiplier. This does limit my CPU OC options, though, to 200mhz increments. I have it at 20x (4.0ghz) and other options are 4.2ghz, 3.8ghz, 3.6ghz, etc. Ill delve into RAM OCing at another time, but for now its just important to know how BCLK will affect RAM speed. With AMD CPUs, you have an FSB:DRAM ratio which acts in much the same way.

The QPI frequency isnt something that needs to be worried about unless you hit an extremely high BCLK (over 200). Generally you want this fast because it will help speed up all data access on the motherboard. On AMD boards, I believe this is often the Hyper Transport speed and it also helps to overclock that.

Both AMD and Intel have features that can change the CPU multiplier automatically. The first feature is a turbo feature, which increases the CPU frequency automatically when less cores are being used in order to speed up single threaded apps. At least for Intel CPUs, it is possible to overclock the base clock and gain a higher turbo-enabled overclock - however there is a limit. I personally hit a wall at 177 base clock, which still delivered a very high turbo clock. Of course when all 4 cores are in use, the overall frequency is lower than doing a straight 4ghz for example.

The other multiplier changing feature they have is EIST/Speed Step from Intel and CoolNQuiet from AMD. In both cases, these features are used to reduce CPU frequency and voltage in order to save power at idle. Generally it is good to have these enabled, but when overclocking to a high frequency that starts to push the limits of your CPU, you may have to disable these functions in order to achieve a stable OC.

CPUs also have a feature called C States, which are related to EIST. The C States are what engage to turn off cores at idle. They are necessary for Turbo Boost and EIST to function properly. There are also different levels of C States, with the higher number (C6) being the most advanced stage of idle mode. Usually these can and should remain active, but you may wish to run C3 instead of C6.

There is one feature that is exclusive to AMD, and it is "Core Unlocker". Essentially, many of the dual and tri-core CPUs from AMD are actually just quad cores with some of the cores disabled. Sometimes this is because they are unstable, but often it is merely done so that they actually have some cheaper dual cores to sell. In either case, many motherboards allow you to unlock cores. If you do this, be sure to test the unlocked cores before getting on with any overclocking. Guide Part 3

Read More..