Over-clocking *101*

[Toaster]

Hello folks!

The following applies to ALL CPU manufacturers, Intel CPUs are used for clarity but the basics apply to virtually ALL CPUs.

Time and time again, I get via E-mail a question: "What is over clocking and how is it done?"
Another question very common is: "Is over clocking safe?"
First, we'll explore over clocking from a laymen point of view.
(I consider a devices nominal/design parameters to be simply a starting point)
Over clocking quite simply is modifying a device or setting of a device to perform over and above its intended design. This can be accomplished by either or a combination of physical modifications of the device or its settings.
When dealing with CPUs (Processors or "Central Processing Unit"), this is often achieved by setting the system board jumpers or BIOS information to values in excess of the CPU makers stipulations. An example would be thus:
An Intel PentiumII CPU designed for 266mhz is fixed at 66mhz FSB and a clock multiplier or "ratio" of x4.0. That is, 66mhz *times* 4.0. The total CPU frequency is then 266mhz. Setting the "ratio" to 4.5 results in a CPU running at 300mhz. This particular CPU (PII-266) tolerates this mild increase in "ratio" usually quite well. Another way is increasing the "FSB" or "Front Side Bus".
This "FSB" setting varies among CPUs. "FSB" is the "system board" mean frequency of which is often "Multiplied" by the CPU to arrive at a given CPU operating frequency. All CPU/Memeory transfer rates are tied to the FSB setting.
This FSB setting can be anywhere from 33mhz all the way through 266mhz.
A very common CPU to overclock was the Celeron 300a.
This CPU had a "nominal" setting of 66mhz (FSB) and a multiplier or "ratio" of 4.5 times the FSB setting. Simple math reveals 66x4.5=300 (there abouts)
Overclockers nabbed this little gem and set the "FSB" to 100 and left the "ratio" unchanged. The net result was 450mhz! Some pushed a bit farther by upping the FSB to 112mhz to get an end result of 500+mhz.
That is a very noticable speed improvement. Not only did the CPU speed increase by almost 50%, the rate at which the CPU could access main memory increased as well almost 50%. The increased FSB beyond 66mhz done many things to improve the performance markedly. The CPU speed was increased, the rate at which the CPU could access main memory increased, the speed of AGP video increased and the OVERALL speed of system I/O increased. So much so that the Celeron 300a could now "outperform" a PentiumII-450 CPU with ease in many apps.
Reason being is the design of the L2 cache on the Celeron which is accessed directly as opposed to the PentiumII which ran at 1/2 that of the FSB setting.
While the cache was 1/4th the size, it was more efficient then that of the PII.
Most Celerons and PIIs tolerate an FSB setting of 75mhz when the design is 66mhz with ease and offers a nice little boost.
However, with speed comes additional heat. This heat needs to be delt with in the most efficient manor practical.
Often, heat increases are usually minor and the need for other then "stock" cooling methods is often not needed.
Things change in the CPU makers facilities which result in the CPU being "resistant" to over clocking. To overcome this "resistance", users often increase the "core voltage" by minimal amounts. This "core voltage" is a setting specified by the manufacturer of the CPU in which the CPU was intended to operate. The above example using the Celeron 300a, the "stock" "core voltage" was 2.0 volts. Some found that increasing this voltage slightly increased the stability of the CPU and made it "less resistive" to over clocking.
This setting is quite sensitive and utmost caution should be used when adjusting/modifying this voltage. Most found that a setting of 2.10 volts was all that was needed. This was one tenth 1/10th of a volt ABOVE the specified and recommended setting. Increasing the core voltage should not be considered lightly and should only be used as a last resort. The increase of core voltage very often results in the heat created by the CPU to increase by a fair amount.
This mandates on many occaision, the use of aftermarket "coolers" or "heatsink/fan" combinations to deal with the increased heat output of the CPU.
Overclockers the world over soon found a number of CPUs of Intel Manufacture that responded well to over clocking. These were the Celeron 300a, the PII-300 (SL2W8) and the PIII-450. The PII-300 went right for 450mhz+ without a backward glance and took on the PIII-450 and gave it a run for its money.
The PIII-450 responded well to FSB settings of 133mhz for a result of 600mhz.
Some users have these CPUs running at 650mhz+.
Then came along the CeleronII (FC-BGA) based CPU. This CPU was initially released at 533mhz and more common was the 566mhz variant. This CPU had nominal settings of 8.5 *times* 66mhz. Overclockers soon discovered that the CPU often run at 100mhz FSB without modification resulting in a speed of 850mhz!
This CPU running at 850mhz ate the PII at any speed for lunch and asked for seconds. At this speed, the PIII-600 also fell to the throughput of the CeleronII. Then the release of the 2nd generation of the PIII.
This chip, also in FC-BGA (Flip-chip, Ball Grid array) was soon found to also respond well to over clocking. I personally have a dualie PIII-700 running at 1.3ghz each resulting in a quite fast system at neary DOUBLE the intended spec.
The bottom line when over clocking is thus:

The benifits:
1. Increased speed of the system CPU.
2. Increased memory speed performance.
3. Increased video performance.
4. Increased general I/O (input/output) performance.
5. Savings and great fun!
5. A sence of accomplishment!
And if approached from a "sane" point, no decrease in stability.

The pitfalls:
1. Increased heat production of all FSB controlled devices/sub-systems.
2. Reduced stability if overdone.
3. Slower devices are now "obviously" slower and may mandate replacement.
4. Increased FSB requires faster memory in many instances.
5. Instability with devices that can NOT tolerate high FSB settings or "bus rates".


In many instances, increases can be almost 2 fold for the cost of a less capable CPU. (Why buy a 850mhz CPU when a 566 will run at 850+?)
However, one needs to use thier head in this. It is unwise to "overclock" a mission critical system as ANY reduction in stability is often "unacceptable".
The benifits are many and the risks few to those that do thier homework which use caution and common sence.

Some terms:
"FSB" - "Front side bus" or the main system "clock" of which is "multiplied" to arrive at the CPU/system operating frequency.

"Ratio" *or* "Multiplier"- This is what it sounds like. The number of *times* the FSB setting is "multiplied" to arrive at the CPU operating frequency.

Example for FSB *and* Multiplier/Ratio:
FSB of 66mhz *times* the "Ratio/Multiplier" of 4.5 (4 point 5) equals 300.

More terms:
VCORE- The voltage at which the CPU is rated for by the manufacturer.
This voltage varies by maker and can range from below 1.5 volts to well over 3,3 volts. (5 volt CPUs are first gen Pentiums and 486 and under CPUs.)

L1 Cache- This is often referred to as "data" or "instruction" cache or a combination of the 2. This cache is most often accessed at the speed of the CPU.
Often, 2 or more such "caches" are used, one for CPU "instructions" and one for "data". This cache GREATLY increases the performance of the CPU and if disabled will have a MARKED reduction in performance, often a factor of 3.

L2 Cache- This is often a "memory to CPU" "scratch pad" in which data is held for the CPU to process without going directly to main memory.
The design of this cache is PARAMOUNT to performance. The "bigger is better" anology does NOT always apply and can actually degrade preformance.
A smaller, more efficient design is better then a large inefficient design.
Properly designed, an efficient L2 cache can have a VERY profound performance increase. (an anology: A needle in a haystack or a haystack in a ball park)


Main memory- Pretty much what it sounds like. The system memory installed in your system.
A little note here: Many recent systems use whats called "Interleaving".
This is where when pairs of memory modules are installed, the memory performance increases by a fair amount because the memory is accessed "accross" pairs of modules instead of "through" odd numbers of modules.

Look at memory in this fashion to better understand "how" your system uses memory:

Go look out a window with a "bug screen". There are a multitude of little squares in the mesh. Now image several layers of this screen, say...4 layers.
Now you have the basics of a memory "array". Each "square" is an address.
There are "rows" and "columns". A "row" is "across" and a "column" is "through" the mesh. Your system locates "data" stored at a particular "row and column".

"Interleaving" is the ability for the system to "page" multiple "columns and rows" in and out of memory reducing the amount of "time" needed to access this data.
So....bank1 is "page one" and bank2 is "page two", swap these in and out quickly or "page" them in and out of the memory "array".

In the very early years of computers, the "screen anology" was EXACTLY what was done to make a memory "array". (basic design in actuality but not EXACTLY as I described)

If one increases the speed of the CPU one does NOT automatically increase the speed memory is accessed. However, when one increases the FSB *to* increase the speed of the CPU, memory speed increases as well. That is, a higher multiplier setting results in a faster CPU but NOT main memory access rates.

One further note:
CPU manufacturers very often "lock" the "ratio" or "multiplier" to a given value. This is commonplace with virtually all modern CPUs in the "X86" world.

More terms:
X86- This is a term attached to "Intel" compatible CPUs.
If your system uses Win95, 98 it is either an "X86" CPU or it is EMULATING an "X86" CPU. The Pentium, Athlon, Duron, Thunderchicken and all its variants are "X68" "class" CPUs. This class is also known as CISC Processors or "Complex Instruction Set Computer"

RISC- This is a "class" of CPU that resides in the high end "workstation" class of systems. These are "Reduced Instruction Set Computer" systems using CPUs like the Alpha, MIPs and many others.

A "CISC" CPU has many "complex" instructions to complete before a task is complete. A "RISC" CPU does the SAME WORK but requires FEWER instrucions to do the SAME task. The result is a much faster CPU because it is far more efficient doing the same work in FEWER CPU CYCLES. One hz (hertz) of a 500mhz (500,000,000) is a "cycle" or clearer yet 500,000,000 cycles per second is 500mhz or 500 MegaHurtz.
This CPU can do the same work in fewer cycles making it more efficient and thus faster.

Yet another term:
Superscalar- This means how many instructions PER CYCLE a CPU can perform in any one cycle. Thus, a CPU that is superscalar running at 300mhz is faster then a CPU running 500mhz that is NOT superscalar. This is DIRECTLY related to the amount of work that can be done in a CPU "cycle".

AMD, Intel, Cyrix and others are superscalar to some degree. Often, this number is below "2". That is MULTIPLY by a factor of "2" a CPU that is NOT superscalar to get a CPU as fast as a Superscalar CPU. (this was oversimplified, no hate mail please!) The MIPs and Alpha CPUs ARE SUPERSCALAR and some have reached 7 instructions per CPU cycle making a VERY VERY efficient and fast CPU.


(The following is from memory and are approximate, refrain from hate mail please!)

The PentiumII can execute 1.3 Instructions per clock cycle.
The PentiumIII (first gen) can execute 1.4 instrucions per clock cycle.
The Pentium Pro can execute 1.3 Instructions per clock cycle (32 bit) and 1.1 instructions per cycle (16bit).
The PentiumIII (second gen/FC-BGA) can execute 1.5 instructions per cycle using SSE instructions.

Over any manufacturer (AMD, Intel) typical numbers are 1.3-1.6 instructions per cycle. (again these are from memory and are approximate!)

The bottom line is simple, one needs to increase all subsystems to get a usable increase in speed. It is BEST to increase the FSB setting rather then just the multiplier. Todays CPUs are locked at a given multiplier and the FSB route is the only alternative unless you have a means to disable the lock like on the AMD class CPUs.

(I corrected some spelling errors, oops!)


I hope this was helpful.

"I don't know how WWIII will be fought but I belive WWIV will be fought with sticks and stones" -Albert Einstien-

[Hpro]

I might add that the Pentium Pro wasn't optimized for 16 bit and actually was designed to run 32 bit applications only.. on 16 bit sometimes it's slower than a 486.. hehehe at least it looks like this to me - and I'm using them for ages..

BTW - I actually have some message for you on the PC-Harware board - may you are interested - it'a about PSU's and fan speed...

[SARGE]

Now, if I could just find where the multipliers are in BIOS, and voltage settings, maybe I could live dangerously. Can't find 'em, though...

Good piece of work, Toaster

[DrZaius]

Great post Toaster, it should be stuck on top of Tips and Tricks or here so it doesn't drop.

[Toaster]

Hello there Hpro (and others),
The Pentium Pro was a unique step in processor design.
It was the first of the X86 class CPUs to be a "true" superscalar CPU.
The reason "why" the "Pro" was a bit of a slug in a 16bit environment lies in its cache design and instruction set. When in the 32bit world, the 200mhz P-Pro can give a PII-333 big worries. The L2 cache on the Pro is accessed at full CPU clock but with a somewhat high latency of near 3. Still, the Pro was a radical change in CPU history. Still used this day by many folks in file servers and network management arrays. The P-Pro boasted L2 "on chip" L2 cache sizes of up to 2MB and 4MB cache Pentium Pros were actually made in very limited numbers.
Those of you in the AMD world with your #2 pencils to "connect" or "bridge" the options on the top of your CPUs is a trick used by those using the P-Pro nearly 7 years ago. (your gonna make me dig up my notebook on this...arn't ya?)
The Pentium Pro "design" was used right upto the PIII when Intel finally forsake the P-Pro design in the P4. In its day, the P-Pro had all the cards. The largest "on chip" L1 and L2 caches, optimized for 32 bit code and of superscalar design.
CPUs from all makers use parts of this design today.

[highrisemech]

I agree with the Doctor. This post should be stuck around the top of Tips and Tricks. I want to print it but my printer is presently down and I'm hoping this post does not get lost in the shuffle. Excellent post Toaster.

[Hpro]

just save the full Message Thread to the hard drive and print it when the printer comes back comes back -
FILE > SAVE AS >SELECT FOLDER >GIVE IT A NAME - or copy pst to note pate and save as..

[highrisemech]

Thanks Hpro, got it........

[Roberto]

Great post Toaster. A list of CPU's and their successful overclocked speeds might be a useful complement to this post.

[Toaster]

Hello Roberto,
I listed some CPUs and their "typical" speeds of which they attain but all CPUs are different. Not all will do "xxx-mhz".
Most of the Celeron line overclocks quite well but not all.
The first gen Celerons at near 500mhz are about maxed. Aside from these few exceptions, almost any other Intel CPU flys with effortless ease.
However, with overclocking, there are no guarantees of what speed the CPU will run with reasonable stability.
"My" personal favorites are the CeleronII 566+ parts and the PIII-600+ parts.
These all seem to do 100% above thier nominal speeds for me.
However, I'm a risk taker but to date have only fried one CPU of which I felt I was going to fry but still pushed. At 122%, it went "knokkerz up" and did nothing from that point on. Still, at 2.22 volts (from 1.70 volts) it was a hot running beast right to its colorful end. I as many feel that Intel based CPUs are marked with "starting speeds" and all we can do at this point is go "up".

[Hpro]

Yes may you have to get your notebook out - let me see something I dont' know - actually it hasn't much to do with the O/C by itself - still got the P-PRO 1Mb Cache sitting here and the board refuses to display the cache - OK I can use DMI to get beeing reconized - but this wasn't actually what I was going to do - as Win2k sees the 1Mb Cache and also CCT = CELEM CACHE TEST version 386 _ that's the name of this vey sophisticated test program - seeing the cache .
I know there is actually the need of those expensive MOBOS to buy -which of course I will not do - which can handle the 1 Mb cache - so if you have something up in the sleeve - just bring it on - as for the O/Cing I'm up at the maximum what a P-Pro could do (~400Mhz) without delaying the other components - and I'm very scared seeing Bus speed exceeding 125 MHz and also RAM (-50us EDO) speed when I O/C the (512Cache chip - not the 1Mb)to more thant 440MHZ.!


and Toaster
TIA...

Edit
You can dowload the program -CCT386.exe - at belows link ...
English:
--------

CCT386 is available for free on numerous BBS under the name CCT386.ZIP.

The latest version is always available for free on the author's BBS
+32-4-365.13.95 or Fidonet: 2:293/2202 or by contacting Celem Computers
by fax at +32-4-367.65.15 or on the internet at http://www.rtfm.be/fpiette

Please feel free to contact the author by EMail to report results on
your own computer or to report bugs:
francois.piette@f2202.n293.z2.fidonet.org