stevenmac666 wrote:Thanks alanore.Much appreciated. especialy as i know nothing about oc'ing and im new to all this computing malarky. Keep up the most excellent work.
Mon the biffy!!!
amd athlon 64 3000 winchester
asus a8n deluxe
geil 2x 512mb dual channel pc3200
xfx 6600gt pci-e
sony dvd rom drive
tagan 480w psu
silverstone temjin tjo5 case
- Posts: 2
- Joined: Wed Mar 07, 2007 10:15 pm
finlay666uk wrote:Well a Raid 1+0 is actually better as it is read of two disks at the same time, and density of the disks matters a lot in terms of performance as does defragging them well and often
Although I believe it is slower when writing as it needs to write the data twice?
Just to add:
For anyone overclocking within Windows, don't!
You'll get a less stable overclock, as you will not know if the overclock would be capable of booting into Windows, meaning you'll be running an unstable overclock for longer - potentially damaging your components.
Also, if you get a registery error about a corrupt file, put back to stock and start again - for me, this always removes the error, although does scare me a bit each time
- Posts: 1286
- Joined: Fri Mar 18, 2005 5:56 pm
- Location: South Wales
your guide to help with Overclocking
Way back on Wed Jan 19 , 2005 11:34, attenboroughp wrote:i remember chris and pepper posted a very long post that they both contributed to that was similar to this. it was very popular but never stickied
hope they get their finger out this time round.
I got bored, did some searching and dug up this from the depths of the forum:
EDIT - I've just found the post which I was referring to above. <<Here it is>>
On Mon Jul 07, 2003 8:29 pm, chris_ah1 wrote:It has often been asked on this forum:
How do I overclock?
Well, I decided to write a mini howto for it.
Before you start overclocking, make sure that your CPU is not getting close to its maximum tolerance temperatures at its normal speed. Most modern BIOSes will have a temperature monitoring device and I advise that you set reasonable limits to let yourself know if your CPU is overheating. Although Intel CPUs have a cut-off mechanism when it gets too hot, AFAIK AMDs are lacking this safety feature and so caution must be used to avoid a burnout.
Next, make sure that there is good thermal contact between your heatsink and CPU.
Next, make sure that you have good quality memory if you intend to overclock that too -any good brand such as OCZ, corsair or Crucial performance Dimms should do the trick. Remember, that only PC3200 is validated and other more exotic numbers are not -such as PC3500/3700. These more expensive Dimms might not be official, but they will allow further overclocking.
Next step is to make sure that your PSU is not about to give out and can supply the CPU with enough power and keep it stable. For this, common sense is necessary and you will have to work it out yourself. But it is safe to assume that a generically loaded PC with a 330W PC with moderate components should be enough. It is also adviseable that the PSU can provide stable power.
OK, now we are ready to start. Check out your BIOS. overclocking is not necessarily possible on some boards. If you have a Dell or some similar computer, there is no point in continuing.
If not, then we can begin.
Make sure that you BIOS has an advanced chipset options which will allow you to change the multiplier or FSB speed as well as AGP/PCI frequencies.
If you cannot see any FSB frequency selections, then you will have to stop.
Now, BIOSes are often different. Some provide a simple setting of 100 or 133 or 166Mhz. For this guide however, I will assume that we can adjust the FSB in 1Mhz increments.
It is also adviseable at this time to lock the AGP and PCI frequencies so that they continue to work within specifications.
Now, some people recommend increasing the voltages for the processer to get better overclocks. However, I have rarely found this to be useful and so we will continue.
The way the CPU speed works is by multiplying the FSB speed by the CPU multiplier.
Therefore, for most intels since you cannot adjust the multiplier you will be forced to increase the FSB. Therefore, an increase of 1Mhz in FSB will result in an overall clock increase of multiplier x 1.
Increase the FSB setting until the system becomes unstable. One thing to ensure is that your memory remains within its clock specifications as well. To do this set the board to automatically set the memory ratio to select 400Mhz if we are using PC3200.
This is done by using divider ratios -but for now it isn't important.
Now, hopefully you have increased your FSB to a point where you start to see noticeable performance gains. To test stability, run a few hours of torture tests with Prime95 and then check for academic performance gains with pi-fast benchmark or even the Prime95 one. It might not be a perfect indicator of increased performance, but it will give you a feel for your increased performance.
For AMD processors it becomes far mor difficult -especially for the unlocked ones. However, the general theme remains the same. Increase FSB as much as you possibly can and still keep the system stable and within heat limits. If you can adjust the multiplier, consider making a processor that normall has FSB of 133 run synchronously with DDR333 memory by reducing the multiplier and increasing FSB to 166.
e.g. for a 2400+ thoroughbred T (pre Barton) reduce the multiplier to 12 and up FSB to 166. This should be stable in most cases -but of course it depends on the chip you have. you might even be able to get a 12.5/13 multiplier to work well.
However, be warned that although this will increase performance and has the potential of doing so whilst remaining stable, you will shorten the life of the CPU as you run it beyond its specificiation. Remember, that you also void any warranty on the product, and I cannot be held responsible for any good or bad outcomes of your overclocking.
Therefore, make sure that your system can be overclocked and that it is a decent enough platform to overclock stably. Next, just remember that you cannot have wild expectations of massive overclocks with stock equipment.
That is really all there is to overclocking -it is mostly a question of preparation.
Its all about the FSB (and the multiplier).
The next one in this series will be about cooling and the different technologies.
If you have any precise questions, just post them.
O, the dalliance and the wit, The flattery and the strife! Was that too long?
On Mon Jul 07, 2003 8:59 pm, chris_ah1 wrote:Cooling.
Well, you can get decent overclocks with stock AMD and Intel coolers. They certainly represent advanced design, and in a recent tomshardwareguide AMD cooler review the stock cooler performed quite well.
However, if you want that little bit (or quite a lot) extra, you will have to look for other cooling solutions.
AIR: Although it isn't a perfect way of cooling, it is the easiest and safest. Any air cooler with a large copper and well lapped heatsink and good fan should provide high levels of performance. Currently, new designs as shown in the AeroCool7 are showing that AIR cooling is far from dead.
Unfortunately, air and heatsink and fan is sometimes not enough.
AIR+Peltier Hybrid: To up the power of the AIR+heatsink cooling combination the AC4G was launched. This has a peltier unit with a heatsink and fan that is regulated by a controller board. The peliter units work by having a large potential difference between the two sides of the plate. This then speeds up heat transfer between the CPU to heatsink. This has its merits, but the product is slightly immature and noisy at full CPU load. Therefore, something else is needed.
Water: Yes, not the safest options or the cheapest, but certainly highly performing and now relatively easy to set-up. high performance kits such as the Asetek have been released as well as kits from Danger Den and Innovatek and wet and chilly chips. All of these kits take the pain out of setting up a water cooled system.
The amazing Crucial external water cooling kit that beats EXOS is about to hit the market as well -although it is plagued by noise problems. Water cooling is also getting cheaper, but still remains expensive.
So, what does your cash get you? It gets you a system that provides higher cooling potential becuase of water's higher heat transfer potential. It is far more efficient at absorbing heat that air -often considered an insulator.
This means that overall a waterblock system will have a lower C/W rating, or for every watt of CPU heat disspated to the heatsink/CPU block fewer degress centigrade are created. Therefore, the lower the figure the better. If new CPUs or overclocked ones can create 100W or more, the lower the number the better. For example, a C/W ratio of .23 means that at 100W of heat dissipation from the CPU the cooling medium will only increase by 23centigrade from ambient.
This is important in giving the CPU more thermal headroom to be overclocked and overheated. Or, it alows you to run a stock system much more quietly than an air+heatsink system. Now, water cooling kits are coming without fans on the radiators.
Unfortunately, although performance is improved, it can be difficult to isntall in some cases -especially with the radiator in internal kits. That is why Crucial and other vendors, like Koolance, stepped in to provide external solutions.
Yes, water cooling can be noisy -but if the radiator fan is stepped down to 7V from 12V, and the pump is quiet, the whole system will be quiet and yet still give good results. The only problem that you have to be aware about is having a good airflow within the case for other components.
Peltier+water: Again, this is similar to the AC4G solution, but uses a waterblock instead of the fan+heatsink. Danger den and silver Prop and to an extent Swiftech are the best peltier capable waterblocks. This not only provides maximum efficiency to transfer heat from the CPU to waterblock, but has the advantage of having water.
Wow, this is the most expensive and the most destructive. Once you use it, you will never be able to use the CPU in another installation (unless you can get all the grease off).
This simply works by providing a very cold plate above the CPU. Rate of heat transfer depends on temperature difference, and so the heat will transfer quickly from the hot CPU to the cold compressed CPU plate. This can have temperatures of -54centigrade, and can run with massive overclocks mildly at 14centigrade. Check out tomshardware for an AMD and intel massive overclock, but be warned that the systems were expensive, and noisy. However, such a compressor system from vapotech or asetek has given the maximum overclocks for P4s to beyond 4.4 with a 3Ghz chip.
Dry ice: I have heard of a cooling system using this, but cannot corroborate it or do I believe it is practically possible. The point with the above cooling methods is that they are all stable, can provide overclocks -moderate to extreme and can be installed.
I also ordered them in ascending difficulty.
Important things to note: Arctic Silver and other special materials might theortetically provide greate improvements from a statistical analysis, but in reality they rarely give more than a 1centigrade improvement -well within error bounds.
Also, depending on the mobo it is easy to compress and destroy your CPU with a heavy heatsink fan combo -500-600g is the limit for intel and AMD AFAIK.
To get extra cooling capacity with a simple air setup, try adding an air duct over the CPU cooler and improve case air flow to get increased performance.
With water cooling, remember to use water wetter to minimise component damage -I have seen components survive a moderate leak.
Peltier systems -be careful becuase condensation can occur and fritz the system. This problem has been mostly dealt with, but can always occur -as with any hyper-cooling device such as compressor cooling.
Remember, the CPU cooler is the thing that keeps your overclock from fudging up. Don't understimate it, and don't overestimate its abilities.
Stop overclocking as soon as you feel it can't take much more.
Next in the series is memory and GPU overclocking.
If you have any questions, don't hesitate to PM me, or post for the combined efforts of the forum to solve it.
O, the dalliance and the wit, The flattery and the strife! Was that too long?
On Wed Sep 03 , 2003 1:18, o_O_JPepper wrote: wrote this for another guy at a different place in the same situation as you... so i'll save myself the hand ache by cutting and pasting instead of rewriting what I have written before!!
Right, here goes... short and sweet...!
FSB = front side bus, how your CPU talks to the rest of the motherboard.
Multiplier = multiplys the value of the front side bus. Was locked by AMD originally to try and stop people from buying a XP 1700+ and running it at XP 3200+ speeds. Didn't work. Now all new AMD CPUs come unlocked as standard... thus so should yours be.
CPU Speed = FSB Value x multiplier
therefore your barton 2500+ actual speed is 1.826MHz
1826 = 166 x 11 (default)
A cpu will have a limit to it's overclockability, not all CPU's are the same. If you have a friend who has a 2500+ and he got 2.3GHz from it, you may not.
More settings in the bios you may need are Vcore (the voltage going into the 'core' of the CPU, usually around 1.6V)
Simply put, the more power you pump into the CPU, the more stable it is... but then it generates more heat and you have to rid the heat via bigger and better cooling solutions. Try not to let the CPU get anywhere near 50* under stress at all IMO. And don't try to max it... a CPU can only take so much before it goes *poof!* 1.8/9 is maximum IMO.
Step 1: find maximum available speed.
First thing to do is to increase the multiplier one setting at a time (usually by 0.5 increments) to find the maximum speed your CPU will go to.
After each increment, test (stress) the cpu to see if it is stable... quick and easy are 3DMark loops, prime95, folding, SETI... in the beginning it may not be totally necessary, but later on it becomes crucial.
A CPU may boot into windows and *seem* stable, but it may not be at higher overclocks.
If it doesn't boot or isn't stable, increase the Vcore one notch... carry on until cpu will no longer increase in speed; your cooling is not adequate enough; or you can't increase the Vcore anymore.
That's you Maximum speed... remember that.
Step 2: Find maximum FSB.
A system running at 200x11=2200 *WILL* out perform a system running at 100x22=2200. Even though the final speed is the same, the latter has a bottleneck where the CPU is not getting the information from the rest of the motherboard fast enough.
Therefore it be better to increase your FSB and lower your multiplier to achieve the final speed.
Right, lower the multiplier to a nice low figure (say 8 or 9 in your case) and gradually increase your FSB speed first in 5MHz increments up until 200MHz and then in 1 MHz increments.
Do the usual stress tests to check for system stability.
Here is where it gets tricky, you need good RAM that will overclock and you need to know how to change Vmem (voltage going to the RAM, usually around 2.6/7V) and how to manipulate the latency timings of the RAM (lower latency = increase bandwidth = better) But these I will ignore for now as they are more advanced and you can get by without them. (might to a follow up if you're interested)
But more importantly you need to know if you're PCI/AGP bus speeds can be 'locked' down at their selected speeds (33/66MHz respectively) or if there is a 'divider ratio' where it is a divisible value for the FSB (Jeez! MY head hurts now! ) usually looks something like 6/3/2 in the bios, where it equals FSB speed/AGP speed/PCI speed. Got that??
I don't know the motherboard you got so I cannot tell, but you will need to make sure you can keep the PCI/AGP bus speeds close to their original values as they *do not* like to be clocked much more than they were originally intended to run.
Well if you understood (or ignored that!) you should find a situation where you cannot raise the FSB anymore and maintain a stable system... This is you maximum FSB speed, remembver this.
You can try raising the voltage to your northbridge to try and get more MHz, usually chipset voltage in the bios, but make sure you have adequate cooling on the northbridge too.
Now try combining the two together with the aim of getting a higher FSB and a lower multiplier.
CPUs, RAM and motherboards are funny things... some settings work others don't... but at each stage *test* the system for stablity and *watch* the temps on your CPU!!!
Enable the CPU thermal protection feature if you have one in the bios and get a decent monitor such as MDM (mother board monitor 5) installed.
Phew!!! got it done... but this is just the beginning... there are lots more to do to get a perfectly overclocked system, and once you've been bitten, it's hard to just not tinker!!
Well, happy overclocking!
Hope this helps!
EDIT: in increasing your FSB, you may find you do not get to the speeds as seen by many other people on the forums. The usually limiting factor is the motherboard (as stated above) and also the rated speed and make of your RAM modules. I'm running at 211MHz (and still going) at the mo' with Corsair XMS 3200, but when I was running generic PC3200 ram I could only get up to 190MHz FSB at a push!
Ahhh!! no disrespect to chris, but I have just read fully the post and realised you have written one too... but feel free to add/amend what I have written here...
Jeez, that takes me back.
EDIT - Here's some extra stuff from the link I provided in my edit note (near the top of this page)
OK. before I continue on about memory and GPUs, let us revisit the AMD multiplier.
Rather than going into detail, I shall use a readily available resource and direct you all to it. Remember, that I take no responsibility for what you feck up while unlocking the CPU:
Also, the XP-TMC Adapter Socket from Upgradeware unlocks the Multiplier as well.
Yes, memory timings can be important. However, a recent performance review that I cannot verify came to the conclusion that performance benefits from overly aggressive memory timings were academic, and detrimtental to stability especially with early Canterwood and Springdale problems with aggressive timings -although it is claimed to have been fixed.
Yes, timings do make a difference, and good quality RAM rated to work at CAS2 will be able to be overclocked and still work at that latency. Lower latency means less delay between data being sent to the memory and back to the CPU. Memory overclocks increase the bandwidth available for the data to travel over. Latencies just reduce delays. That is why synchronous memory operation is so desireable becuase the delay will be further minimised as the memory will not have to wait for the CPU to catch up, or vica versa.
Yes, crucial high-performance Dimms can be pushed that little bit extra. However, their value-Rimm line rated at CAS3 will not give that much extra oomph (unless it is a golden sample memory stick).
Now, as I already hinted at, the bandwidth that memory has available to transmit information to the CPU is calculated by its clock speed. PC3200 RAM will have a maximum single bandwidth with the CPU of 3.2GB/s theoretical. Dual RIMMS will have 6.4GB. Therefore, if the CPU is an intel with its quad pumped FSB, it needs RAM that will saturate the bandwidth so that the CPU will again become the limiting factor. This is provided by faster memory Mhz speeds, and then to increase the bandwidth that the CPU can use before it is saturated, the FSB must be boosted. and so the cycle continues.
more memory bandwidth means that data can be moved more quickly from CPU to memory, and so faster system performance. This is one reason why HyperOS feels it can claim such amazing performance improvements.
Memory overclocks and to improve memory overclocks:
Of course agressive timings, and especially RAS and precharge timings are important in increasing efficiencies towards theoretical maxes. To get more performance however, you have to overclock. This can be done by one method -using memory divider ratios. Modern BIOSes have a soft option for this, so I shall concentrate on it. A 1:1 ratio in the BIOS means that a 200FSB speed will equate to 400MHZ memory speed after dual pumping for DDR RAM. Rambus uses quad pumping -that is why with a 133Mhz FSB you can get 1066Mhz Rambus -certainly the performance choice for quite a while.
Therefore, if you increase FSB on a new C stepped P4 to 250Mhz with 1:1 ratio, memory will work at 500Mhz. However, until DDR-II or QDDR this is not possible quite yet.
Therefore, you change the ratio so that a 250FSB will become something that the memory can handle. For example, let the memory be unratified DDR433, then make the divider give a memory clock speed close to 433 e.g 7:6. This means that for every 7Mhz of FSB, you get 6MHz of memory clock -in this case you get memory running at 2*(6*(250/7)=429. It might not perform as well as synchronous memory, or to its maximum ability, but it is faster.
to improve memory overclocks, as with CPu overclocking, you can increase the voltage going through it. Some people swear by the improvements it makes, but I have never found it to give a significant edge to overclocking. What it does do however is increase heat production. high speeds+high p.d. = hot
Therefore, boutique products such as active ram coolers, and large heatsinks have been developed. However, for the gains they produce IMNSHO they are not worth it. Some special memory modules dubbed 'performance' or 'Hyper' will have heat-spreader built in. Their performance gain is questionable, but copper does look nice if you have a side with a window. Also, copper has a high heat conductivity constant so it is good at transferring heat, although there is no point dissipating heat to heatsinks if there is not adequate air-flow within a case
Note -this is only possible on GPUs that haven't been locked. I will also not go into the risky procedure of converting a Radeon 9500 to 9700 becuase the cards that could do this are in very short supply -if available at all with the advent of the 9600.
With current advances in chip manufacturing, the fab. process has produced smaller and thinner chips -with most GPUs being created with a .13micron process. this means that higher clock speeds are possible. As with DDR-II that is appearing on high-end cards high clock speeds are also possible.
However, be warned that many high-spec. video cards based on thse advanced technologies will already be running hot and you will have to dissipate extra heat through extra cooling to get the maximum out of the card.
As with CPu overclocking, don't be wild in your dreams for the fastest GPU, but be reasonable in expectations. Also, try to stop once you ge artifacts on the screen.
To overclock, simply slide the button on the clock speed slector as far as you want to go -for memory and GPU clock speeds. this can simply be done in windows with certain tools.
for NVIDIA cards use the overclocking applet in the detonator drivers (search for the registry hack on the internet)
For ATI cards, use the applet that I think Rage3D (RAGE3D TWEAK) has for the cards. However, it is also possible that the card manufacturer has their own tuning utilities and have left the card open -if so, use their tool.
One thing to note is that cooling might now become an issue, and you can get some overclocking simply by sliding the GPU general performance tab to maximum performance -although it varies from card to card (the sure way to overclock remains with the direct tools).
To cool a GPU, you could add nvidia's 'hoover' cooling design. Or, you can use small peltier elements with water cooling, plain water cooling, improve the heatsink/fan combo or use a passive heat-pipe solution such as the Zalman one. Ram heatsinks can be also used and are often included on deluxe models.
The Zalman heatsink is a bugger to install, and so that is why Sapphire offer a Radeon with this solution. It might not give huge overclocks, but it is as good as stock cooling (in good air-flow cases) and is quiet. Water cooling is easy if you already have it for the CPU and offers good performance. However, as with the CPU cooling improvement, air cooling improvements for the GPu are the easiest and cheapest -just be careful with the chip.
Next, we shall have a general guide for maximising air-flow within the case -I know it is common sense, but it often helps to have it written down and it is such an important issue for overclocking stably. Shame we can't pin these.
Over the last few parts of these HOWTOs the basic ways and considerations for overclocking have been dealt with. This encompasses not only how to overclock, but also how to do it properly and safely. Adequate cooling is the centre of an overclocking platform. Already RAM cooling, GPU and CPU cooling have been debated, but hardly the cooling of all the other parts. This is why we are going to talk about good ventilation in the case which often people neglect. Often the worst examples of case cooling can be found with smaller integrated systems (shuttle), because it is almost impossible to get extra air-flow and in OEM brands where corners are cut because the system was not designed to be stressed. However, good case cooling is critical to good overclocking because not only will it keep cooler air moving over the heatsink in an air-cooled system, but it will keep the rest of the components cooler if they are being stressed.
Today's processors usually require massive heatsinks with high-flow fans, especially for those interested in maximum overclocking potential, or more exotic cooling systems. These can dissipate large quantities of heat, but they still require a consistent supply of cool air to get rid of some of their heat to remain efficient. The easiest and cheapest way to circulate cool air around the system is through the installation of case fans to either intake or exhaust air from the room. Of course you can fill the computer with cold air from an air-conditioner, but that is a bit extreme and unnecessary.
Most low-cost OEM designs rely on a single fan to cool both PSU and move air inside the case. This is bad for overclocking because one or two fans just isn’t enough to create adequate air-flow for heatsink+fan or anything else -especially since the PSU creates so much heat anyway.
So, we know that we need a few more fans, but what do we have to know before we start?
The larger the fan, often the better performance is an adequate starting point. A fan with 120mm diameter will have far better cooling ability than a 60mm fan because the amount of air it can push/suck increases greatly -assuming ceteris paribus, e.g. RPM (revolutions per minute), blade design etc. However, size is not the most important issue, noise is. Most people prefer large fans with low rotation speeds, as these tend to move the same amount of air as compared to smaller units, but with a lower noise level. However, it is important to note that noise is also affected by fan design and efficiencies. Some dual bearing advanced fans with super-efficient blades can spin at double the rate of a similar diameter fan and yet make half the noise. Noise is created not so much by the turning of the motor, but more by turbulence and vibrations.
Voltage is also important. A fan working at 12V will spin much more loudly and faster than if working at 7V or 5V. The only compromise with lower voltages is that cooling abilities are reduced. However, for a case fan 7V is more than adequate and is far quieter since there is less stress in the fan and much less turbulence. N.B. if a switch of voltage causes resonance to occur by creating air turbulence at the natural frequency of the PC case stop the fan immediately and change voltage.
Therefore, chose between cooling ability and noise levels -although fan design can make a quieter fan more efficient.
My personal recommendations are for controllable fans, or 'smart' fans such as the coolermaster and thermaltake 'smart case II fan'. They are identical from both companies; although the coolermaster product is more practical because the thermostat has a longer sensor. These are nice because not only can you control them via a rheostat, but you can let them automatically vary their speed based on temperature -the hotter, the faster. Nice!
However, for most people just a fan with some capability of control is necessary. For this, I recommend any 80mm case fan from panaflow, sunon, or Papst. These are the best quality fans in my opinion for a compromise between great performance and quietness. One thing to note is that you will be limited to what your case can handle in fan selection -although adapters are available -or you could always create your own blowholes. For maximum performance, or if you want to learn the joys of flight, get a Delta screamer. As the nickname suggests, they are loud at 55dB+ but they do shift over 80CFM of air per minute -certainly the best raw performer.
Some OEM cases will only have one installation point for a fan, whilst larger higher quality cases will have several more. Hopefully, your case will have at least one free front fan fixture, one free back fixture, and a side one. For my purposes, we will assume that we are using a 3fan setup for a moderate system.
I recommend putting an intake fan at the bottom front to suck cool air in, and another similar fan at the back and close to the top to exhaust air. It would be ideal if the case has a side panel fan fixture so that you can have cool air coming directly onto the heatsink+fan combination of the CPU. This has the added benefit of utilising hot air's natural tendency to rise up and be extracted directly by the top rear fan.
Be careful that the fans are balanced though. You don't want a delta screamer as your exhaust and a weak 1000RPM fan as your intake. This could create a lower pressure environment and seriously hamper cooling performance of all components as less air enters the system. Likewise, if the situation was reversed you could suffer from pressure increase and get extra dust on the components -even worse.
To prevent a dust problem, or slow it down, some type of filtering is necessary -foam or otherwise. However, this should not replace regular maintenance to clean up the PC. Don't forget as well to put a protective shield on the fan to prevent fingers from being lost.
Ok, so it is theoretically set up to be perfect, we must make it practically work. This can be done by making the cables in the PC orderly and bundled up -preferably with rounded cables. This causes less air-flow disruption and so increases efficiency as turbulence is decreased. Also, make sure that HDs and optical drives are not crowded but space them out within the limtis of your case. The same could also be said for spreading out PCI cards, especially away from hot GPUs with large cooling systems.
OK, the system should be fine now. But if you want that little bit of extra peace of find, there are always more exotic systems available -this might especially be necessary if you have a very hot HD running, or have a hot GPU that you want to cool a bit more. For this, you can get a slot cooler that installs within a PCI slot below warm devices. It then blows cool air into the system -they move moderate amounts of air quite quietly.
Baycoolers that slot into free 5.25" slots are for cooling HDs, or getting extra air into the system. If you are running a RAID5, or dual RAID1/0 arrays this could be quite important if you have many HDs below it, and optical drives above.
Whatever you decide to do, poor case cooling can limit the efficiency of even the best processor cooling system installed -whether it is simple air cooling or something more exotic, well, especially important for exotic cooling methods. Having a continuous supply of cool air will reduce overall system heat stress under load, improving stability at any speed, and helping higher overclocks can be reached. Whether you are a DIYer, or bought your PC from an OEM, don't underestimate case cooling -it might be cheap, but it can be critical.
It seems that this board has gone Crucial crazy -and yet I have to question why go crazy over a great brand and yet all the choices, numbers can often seem impenetrable and too complicated. What is the solution to this problem? Well, its going to be me explaining generally what all the numbers mean very simply and all the less prominent numbers. Hence this 'Need to Know' as THG would call it.
Memory comes in different types, but the most mainstream modules are DDR ones. These operate on the principle of doubling the FSB speed to double theoretical bandwidth maxes over normal memory modules.
DDR266=2x133 and is also called PC2100
Therefore, DDRxxx refers to FSBx2 and PCyyyy refers to the maximum theoretical bandwidth that the memory can use to communicate with the processor. In this case it is 2.1GB/s. Efficiencies for this memory speed is very high with many machines (especially laptops) reaching speeds of 2.06GB/s.
After that, new memory standards were created and ratified:
DDR333=2x166=PC2700 with 2.7GB/s
DDR400=2x200=PC3200 with 3.2GB/s -latest ratified standard.
After this, you see exotic DDR433,466, and now even a DDR500 model. However, the voltages required to keep the last module stable without interference is getting too high, and DDR-I is reaching its maximum performance levels. This is why DDR400 running in dual configuration on canterwood and springdale at very aggressive memory timings has had so many problems -the nforce2 ultra and its stability only shows nvidia's prowess at board manufacturing, although they don't have to contend with a quad pumped CPU FSB.
Ready to take over are DDR-II and QDDR (the topic of a later discussion).
Performance benefits are seen when the FSB of the processor is run synchronously with the memory.
Therefore, a 166FSB Athlon Barton would run synchronously with DDR2*166 and give better performance than running with DDR2*200, despite its numbers being bigger. This is the key to looking at DDR modules, it is always 2*mem. FSB and if it is the same as CPU you are happier than if 2*FSB is greater or smaller than CPU external clock speed or FSB.
If you want to run memory at non 1:1 ratio speeds, motherboards use dividers that create a ratio of CPU FSB: memory FSB.
However, inherently it is possible to see the problem with this and why synchronous operation is preferable.
If there is divider, then there is going to be a gap between the time that data is available for the memory, and when the memory is available to accept the data (or vica versa).
Another trick to improve performance is to utilise dual DDR technology -have two modules act as one in effect, thus doubling previous bandwidths, e.g. 3.2GB becomes 6.4GB/s. This greatly improves performance, and can now theoretically saturate the quad-pumped FSB of the Intel P4 at (4*200)Mhz.
These delays or latencies are a major part of performance loss, and it is probably true to say that computing performance is more limited by the speeds of memory rather than CPU. Even if you made the CPU infinitely efficient and performing, you would still get similar performance to what you do today.
So, what other important settings do we have to play with?
CAS latency (CL) Number of clock cycles that pass from when the memory column is addressed to it being sent out through the output register. This is the most marketable and obvious rating memory manufacturers use. Ratings of 1, 1.5 are very aggressive and beyond the capabilities of all RAM officially at the moment. Performance dimms will deliver 2 -the rest will be at 2.5 and 3.
RAS-to-CAS delay (tRCD) Number of clock cycles that pass between the row address being determined and the column address value being sent out. 2 is the most aggressive setting, but can yield speed improvements of 2 to 5%. 5 is the maximum value.
RAS precharge time (tRP) (2/ 3) Number of clock cycles used to precharge the memory circuits so that the row address can be identified. A setting of 2 is very aggressive with 3 being the default on most mobos without performance tuning.
As with all of these, the lower the number, the faster the performance because less clock cycles are being used so less inefficiency. However, this comes at a cost, and that is stability. It is similar to wireless networking with short and long preambles. A long pre-amble might be slower, but in a heavy network environment it is much more reliable than short preamble because there is more certainty a packet is for your NIC.
The same is for memory -the more cycles used, in general, the more stable the performance. This is inherently true for all of them because to access precisely the right part of the memory, you have to be accurate, and more time to do a calculation will make it more accurate in this instance.
So, for your machine and performance boosting you have to find a good balance. If stability is key, go for conservative settings that are inline with your official memory ratings. If you want that little bit extra, change settings to different combinations until the machine becomes as fast as it can before being stable (defined by me as being able to run PRIME95 torture test for at least 12hrs.)
Now, I’m sure some people are thinking: You said 2 to 5% performance increase, is that noticeable? Is it worth me sacrificing stability? Is it worth the hassle if it will only be an academic performance difference? Well, I personally think that no noticeable gains will be yielded if only one setting is made more aggressive. You have to have good quality memory (expensive –not Crucial they all shout), and make all settings as aggressive as possible. Even then, the improvement is debateable. There was a very detailed review of the effects of these settings and their permutations, and I shall try and post a link to it.
Choosing memory can be daunting since the choices are so numerous, and so I hope that this has helped explain, condense the major differences, and the advanced BIOS options for maximum memory performance.
Side point about Rambus: I haven't discussed Rambus here because despite continued development of the architecture, and DRDRAM1200 production, I feel it will never be mainstream again now that Intel has dropped its development. Sis is the last firm developing a platform for it, and it could not compete with Dual DDR platforms. For detailed discussions and information of DRDRAM search Google.
Revived after 4 years!
- Posts: 6966
- Joined: Sun Dec 07, 2003 8:37 pm
- Location: Manchester
Up to date Guide for modding (Core Duo/1333 FSB/DDR3 etc)?
Is it possible someone can post an updated guide that allows for the latest mobo's, memory (ie latencies seem pretty irrelevant for example - if they aren't however it just proves that it needs explaining in the context of the Core Duo's/Quads, DDR3 etc), graphics cards. etc.
Thanks a bunch.
- Posts: 24
- Joined: Thu May 17, 2007 3:29 pm
Haha, attenboroughp thats a blast from the past!
With regards to neil_board, yes the guide is hopelessly out of date, and in need of some revisions. Although the main fundementals are the same, DDR3 isn't thats diffrent to DDR. I would consider the graphics overclocking guide more out of date, with the advent of the stream processors.
I have no plans to rewrite the guides, as I have no intrest in computers. Although I am considering the idea of making the guides copyleft, which would allow anyone to re-write this guide without permission and use it as they see fit, it would also be nice if the current version of "turbo nutters" posted up on the forum too.
- Posts: 5233
- Joined: Sun Apr 11, 2004 11:55 am
- Location: MON THE BIFFY
- Posts: 5233
- Joined: Sun Apr 11, 2004 11:55 am
- Location: MON THE BIFFY
Re: How-To Overclock
Be more afraid that it may never begin...
Ever get the feeling you might be missing something?
- Posts: 4092
- Joined: Sun Mar 12, 2006 3:34 pm
- Location: Close to the Edge, down by a river
Re: How-To Overclock
Basically, no. Your laptop will have only enough cooling to cope with it's designed specification.
By the way, Links on signatures are against the forum rules, until you have 'earned' them by posting lots of useful and fairly regular contributions.
Be more afraid that it may never begin...
Ever get the feeling you might be missing something?
- Posts: 4092
- Joined: Sun Mar 12, 2006 3:34 pm
- Location: Close to the Edge, down by a river
Who is online
Users browsing this forum: No registered users and 2 guests