This blog post is part of a series on my experiencebuilding the perfect computer.
Before you start to decide what type of performance yourequire, you should put yourself into one of two categories:
Gamers are people who use their computer for playing video games. For these, the number of threads and quantityof memory is less important than their speed.
Creators are people who will be using their computer mostlyfor work, whether it is for building code, rending video, etc. This category of user cares wants morethreads on their CPU (at least 20) and a higher quality of memory (at least32gb). They are also willing tosacrifice some cooling performance of exotic, more complicated cooling systemsfor the sake of reliability. GPUs areless important for most, and will often be chosen based on the number ofmonitor outputs.
VCORE is the voltage the CPU will receive, set in bios.
VID should drop under load.
VDROOP: The vcore difference between idle and load, whichcan be eliminated by Load Line Calibration.
Many people say that it is not worth overclocking many ofthe top-end CPUs, since they come with features that automatically increase itsspeed based on the efficiency of your cooling.
AMD has a system called Precision Boost Overdrive(PBO). However, I found that PBO endedup
Part of this is due to what is called the SiliconLottery. You see, each individualCPU can have small variations in its manufacturing that can have very largeimplications for performance.
A CPU’s speed, also known as frequency, is measured in gigahertz GHz). The higher the speed, the more performanceyou get.
The power put through a CPU is measured
Originally, each CPU had a single core. As technology improved, we were able to fitmultiple cores onto a single CPU. Current CPUs can have several cores to several dozen – each core canbasically complete one unit of work at a time, known as a thread. Newer CPUs have the ability to create twovirtual cores for each physical one (know as Hyperthreading in IntelCPUs). This allows processors toeffectively have 2 threads per core (for instance my AMD 3950x has 16 cores and32 threads). If you are a creatorbuilding a workstation, the more cores the better. If you are a gamer, you should opt for ahigher speed over more cores – since many games do most of their processing ona single core.
I found that if I statically set the frequency and voltageon my CPU, it runs cooler and faster than if I leave it on automatic. In some circumstances, this can be
Overclocking is based on twoparameters: voltage and speed.
The first step is to determinethe generally accepted voltage range for your CPU. This is sometimes provided by themanufacturer but in my case, I had to look on forums to find that theprevailing wisdom is that it should be between 1.3v and 1.45v.
As you increase the voltage, themaximum possible speed of the processor also increases but so does yourtemperature and the possibility of damaging your CPU. If you go above the maximum voltageprescribed for your model, you will almost certainly damage your CPU, but therehave been some instances where voltages towards the upper limit can causedegradation over time, although these can be hard to verify.
In order to determine which specsyou should use for your CPU, choose a voltage within your range and a CPUspeed. Boot up your computer and run aStress Test to determine whether the combination that you chose is stable. If your processor is not stable at the levelsthat you choose, your computer will either crash or the stress test tool willinform you that it has detected some instability.
If any instability is found, youcan either decrease the speed or increase the voltage (while staying within arange that you are comfortable with). Iwould recommend changing in very small increments (e.g. 0.1) to find acombination that works for your specific CPU. Once you find a combination that seems to work, I would recommendrunning an extended stress test (e.g. 8 hours) to make sure that it is good forthe long-haul.
For each combination of voltage and speed, you shouldmeasure the following metrics:
- Idle Temperature
- Load Temperature
This is some quantification of overall performance. Many people use Cinebench,which gives you a single score to compare.
The most important part of memory is that quantity that youhave, which is measured in gigabytes. The more memory you have, the more programs can be actively running onyour computer.
The second thing to consider is speed, which is measured inMegahertz (Mhz). 3,600 Mhz memory isconsidered relatively good, but you can get memory up to about 5,000 Mhz orfaster if are willing to water cool it (but that is very rare).
Finally, there is memory timing, which isthe most obscure aspect to performance. This includes four different components, which is normally representedin a hyphenated form (e.g. 18-19-19-39). This includes the following measurements involving Column AddressLatency (CAS) and Row Address Latency (RAS):
- CAS Latency
- RAS to CAS Delay
- RAS Precharge
- RAS Active Time
I don’t pretend to understand the subtleties of thesemetrics, but I can say that lower timings are better.
By default memory will actually run at a slower rate thanadvertised, known as the Serial PresenceDetect (SPD) Rate. In order to getthe promised performance out of your memory, you need to manually ensure thatit is running at the advertised or tested rate. Some memory has what functionality called Extreme Memory Profile (XMP)which automatically loads the advertised rate, but make sure to verify that itis working correctly – many people have reported that after activating XMPtheir memory performance still is not as advertised (you can verify the timingsset on your memory using a tool like CPU-Z).
To be sure, you just have to set the advertised values foryour memory into your bios. As long asyou have effective air cooling in your computer, this will not cause anydamage. If you want to take things astep further, you can even overclock your memory, but I did not go down thatpath.
The most common overclocking is done on CPUs. However, you can also overclock a lot ofother parts wof your computer like your memory, GPU, and even motherboardchipset.
Weirdly enough, there isn’t really a single ultimate toolfor measuring performance while overclocking your computer. Instead, each tool has its strengths andweaknesses and functionality usually overlaps between them. Even amongst the most popular tools, you willoften get vastly different metrics. Therefore, I would suggest downloading a bunch of different tools andfiguring out what works best for you.
A paid tool that does pretty much everything, all in one place.
A widely-used tool to compare performance.
An excellent simple tool for measuring CPU Load,
| || |
| || |