What is the disadvantage of overclocking?

[abdullahseba]

What is the disadvantage of overclocking?
Is it all to do cooling or is there more t it?

[Simon]

the processor or other device becomes unstable, the clock speed is limited by how quickly it can transition a signal from 0-1 and from 1-0, it also needs to syc with other signals so the faster you go the more chance of things not happening when they should like gear cogs slipping.

Increased speed makes more heat, requires more power, may require more voltage to make the transitions happen faster, more more heat and power per clock and lower efficiency.

[Muxr]

There is speculation that it can reduce the life of the component as well, since often times to achieve higher overclocks you also need to up the voltage.

Now days most CPUs have some sort of a temperature driven clock control (what they call turbo), on CPU intensive tasks a CPU will jump to a higher clock while monitoring the temperature, as the temperature increases the CPU underclocks itself. So cooling is paramount if you intend to get the max out of your CPU. Otherwise if you overclock your CPU might get stuck on lower p-states due to high temps.

[deephaven]

Intel officially support overclocking http://tinyurl.com/qxmrq9j

[Simon]

Intel officially support overclocking http://tinyurl.com/qxmrq9j

Of course, they have realized that a fair chunk of us want to get morte for our money and feel we have made a win if we get it to go faster and think we got one over on the manufacturer. I have my AMD overlocked by 300MHz (in 4.3GHz), they have never said don't, they have only said at your own risk. Most users want a stable PC so they edge on the side of caution as they don't know what your specific working environment will be like and if you want to push for more that is up to you.

Overclocking RAM is also nice, no point in that fast CPU when it's ram is slow, I have my timings slightly reduced and it's noticeable.

[wraper]

Overclocking RAM is also nice, no point in that fast CPU when it's ram is slow, I have my timings slightly reduced and it's noticeable.

What noticeable? RAM overclock barely brings any performance increase in the most real tasks (not synthetic RAM benchmarks), like 50% RAM clock increase = <5% performance increase. Yet tremendously reduces stability of the computer. RAM overclock may become PITA because it can cause something like one BSOD in a two weeks, but if you try to run RAM tests, nothing wrong is revealed even if running them for days.

[wraper]

Core i7 2600k:










http://techreport.com/review/20377/exploring-the-impact-of-memory-speed-on-sandy-bridge-performance/3

[Rick Law]

Not to forget are the expansion cards and peripherals.  Some cards may draw clock signal directly from the PC's clock instead of an on-board clock.

I use an old ASUS P5 motherboard.  I have three graphic cards two of which will work at clock+10%.  The other one cannot function even at clock+5%.  And then there are disk drives.  I have an extra IDE port that I use to plug in my different drives with different backup files.  Some drives will occasionally failed.

So, while you tested your motherboard to be "capable" of running at clock+x%, there may be subsystems on or outside your motherboard that would fail under certain circumstances.

[edavid]

Undetected computation errors could be considered a disadvantage...

[Simon]

Not to forget are the expansion cards and peripherals.  Some cards may draw clock signal directly from the PC's clock instead of an on-board clock.

I use an old ASUS P5 motherboard.  I have three graphic cards two of which will work at clock+10%.  The other one cannot function even at clock+5%.  And then there are disk drives.  I have an extra IDE port that I use to plug in my different drives with different backup files.  Some drives will occasionally failed.

So, while you tested your motherboard to be "capable" of running at clock+x%, there may be subsystems on or outside your motherboard that would fail under certain circumstances.

that is why you use an unlocked processor and up the multiplier.

[Simon]

Core i7 2600k:










http://techreport.com/review/20377/exploring-the-impact-of-memory-speed-on-sandy-bridge-performance/3

try 3d modelling and voice recognition, if you need to work on large amounts of data randomly the faster you get the data from the ram the better.

remember dual channel ram speeds are 16 paralely transfered bytes and not as fast as they can look

[Rick Law]

Not to forget are the expansion cards and peripherals.  Some cards may draw clock signal directly from the PC's clock instead of an on-board clock.

I use an old ASUS P5 motherboard.  I have three graphic cards two of which will work at clock+10%.  The other one cannot function even at clock+5%.  And then there are disk drives.  I have an extra IDE port that I use to plug in my different drives with different backup files.  Some drives will occasionally failed.

So, while you tested your motherboard to be "capable" of running at clock+x%, there may be subsystems on or outside your motherboard that would fail under certain circumstances.

that is why you use an unlocked processor and up the multiplier.

Well that wouldn't be overclocking the system in my view since everything except the CPU is running at normal speed.

Come to think of it, even when the entire system is running at a different clock speed, there are many other unchangeable - my disk is still spinning only 7.2K RPM.

[Simon]

The CPU is the "prime mover" in data crunching, you start there. My pc can just keep up with my SSD anyway so ?

[wraper]

try 3d modelling and voice recognition, if you need to work on large amounts of data randomly the faster you get the data from the ram the better.

remember dual channel ram speeds are 16 paralely transfered bytes and not as fast as they can look

Even if you take the most RAM speed dependent application, you won't get very significant performance increase. You'll need extreme overclock to make a difference. 10% clock increase won't give anything worthy. Only idiot will do 30-50% RAM overclock (to make it worthy) for any real work. BTW I have four channel RAM in my PC.
EDIT: worth to mention. RAM likes to degrade very much.

[Simon]

According to sinthetic tests my ram runs at 40% of theoretical bandwidth and is started going down since the introduction of DDR technologies so yes the ram is already at a disadvantage. If you have 4 channel maybe you have already maxed out the required performance, but again simultaneously transfering 32 bytes of data may only get you 1 in 32 of the bytes you really need. Dual channel and DDr rely on the system predicting which bytes are wanted together, the memory still works at 200-400 MHz it's just that you have more ram chips than access channels so the channels go faster and switch from chip to chip, much like the rigol scope using 10x100MHz ADC's to get 1000MHz effecive speed.

[Rick Law]

The CPU is the "prime mover" in data crunching, you start there. My pc can just keep up with my SSD anyway so ?

The "so" is: in the past, I overlooked those other pieces and view overall clock increase in isolation - I didn't think of them (I considered the IO speed but the disk RPM is overlooked) until I was posting just now.

CPU's importance really is application dependent anyhow.  Most of my stuff these days are broad-band speed limited or IO limited, not the CPU.

[Bassman59]

What is the disadvantage of overclocking?

From a design verification standpoint, consider:

When you do a synchronous logic design, you do static timing analysis on all of the paths. The prop delay through each logic path has to be shorter than the clock period over the whole range of variations in process, supply voltage and temperature, and the STA will tell you that. When you overclock, you run the risk of clocking too fast for some of the logic paths, which leads to errors. Some of those errors might not be very obvious, which is why they're dangerous. I think it's actually better that the machine just comes to a screeching halt when overclocked than having some insidious errors occur.

Also consider that the clock tree has its own limits on clock frequency.

Finally consider that the PLLs and other clock generation blocks have their own limits.

[Simon]

Try running solid edge up to version 7, it uses 1 core only and we have x2 4 core processors so 4 cours that look like 8 and the main program only uses 1, so the best thing for that program is dual core as fast as possible. The processor has to do all the calculation for models that can be 100's of MB to move them on the screen, the ram at that point is as important as the CPU speed, you can't cache 100MB......

[Simon]

increasing the memory controller speed in my CPU from 2400 MHz to 2600MHz was a good move, and it has it's own supply voltage.

[wraper]

According to sinthetic tests my ram runs at 40% of theoretical bandwidth and is started going down since the introduction of DDR technologies so yes the ram is already at a disadvantage.

Better check something from the real world.

If you have 4 channel maybe you have already maxed out the required performance.

4 ch RAM in LGA2011 actually don't have very significant effect on real performance compared to cheaper 2ch Intel sockets. Yes it rocks in synthetic benchmarks but it the real word not worth too much.

[wraper]

increasing the memory controller speed in my CPU from 2400 MHz to 2600MHz was a good move, and it has it's own supply voltage.

And increasing this voltage is one of the fastest ways how to kill CPU.

[dr.diesel]

My 2500K (stock frequency of 3.3Ghz) has been running at 4.7Ghz since 2011, 24-7, so far so good.

[GoneTomorrow]

Got a sweet overclock on my Phenom II x6 1090T. Increased CPU bus speed from 200MHz to 240MHz (no peripherals are based on this clock). Lowered multiplier from 16 to 15. This resulted in a CPU clock speed increase from 3200MHz to 3600MHz, and also a memory controller increase from 2000MHz to 2400MHz with the stock 10x multiplier. Lowered memory multiplier to from 4 to 3.33 to keep the 800MHz (1600MHz DDR) clock speed.

At the end of it I was also able to lower the CPU voltage by 75mV, which resulted in about a 3 degree Centigrade lower load temp. Ended up about 15% faster. This CPU can go over 4GHz with a voltage increase, but I prefer a silent computer.

Overclocking can get addictive, I guess that could be a disadvantage 

[free_electron]

what's the point ? 99.99% of the time the machine is waiting for input from the meatbag typing with two fat fingers on a small key board (or, in this case :one fat finger as the other hand is holding an icecream cone...)

[GoneTomorrow]

what's the point ? 99.99% of the time the machine is waiting for input from the meatbag typing with two fat fingers on a small key board (or, in this case :one fat finger as the other hand is holding an icecream cone...)

That's a very incorrect statement. Many computers spend most of their time at 100% CPU usage; compiling code, rendering 3D images, encoding video.

[Simon]

what's the point ? 99.99% of the time the machine is waiting for input from the meatbag typing with two fat fingers on a small key board (or, in this case :one fat finger as the other hand is holding an icecream cone...)

My PC's spend  100% of their time running boinc which aids various medical research and scientific research. At work if I had better ram than the measly 1333MHz I had I could do 3D CAD modelling AND run boinc, sadly due to the ram not being bast enough I have to reduce the amount of cores usable by boic just to leave some RAM bandwidth to solidedge. My home PC with 2133MHz ram can model 3D CAD and use all 4 cores to run boinc,

[Psi]

Unless your having to water cool it to get rid of the massive heat from upping the voltage to something insane I wouldn't worry about any sort of damage.


You can safely up the voltage a little, stick a better heatsinkfan on it and push the clock until just before it becomes unstable. Your typical serious overclocker does this for the life of the CPU with no problems at all.

[Simon]

I'll up voltages as much as 10% and that is all. at the end of the day if i want serious performance increase I'd buy the faster CPU.

Overclocking is a bit of an art and hobby, or for extreme cases of obtaining a performance that is not commercially available. At the end of the day the money spent on extra cooling and the extra power used by the processor over it's lifetime will cost much more than the equivalent processor that was designed for that performance in the first place.

I buy "better" heatsinks as a matter of course but that is more to reduce noise, I can run 100% load in thw inter without it screaming, in the summer i tend to drop the amount of processor time boinc can use to reduce noise and heat in the room, and yes I notice the difference.

[BradC]

what's the point ? 99.99% of the time the machine is waiting for input from the meatbag typing with two fat fingers on a small key board (or, in this case :one fat finger as the other hand is holding an icecream cone...)

For general use I kinda agree, but I have a machine here (i7 3770k) which is water cooled and running at 4.6Ghz. That brought an autodesk automation related process I have to do from 60 minutes to a little under 50 minutes. AutoCAD is single threaded, and every MHz matters. If I have to run that process 6 times a day it saves me an hour, that allows me to get another hours actual work done that I can charge for. So for specific tasks it can actually be quite valuable.

Unfortunately because AutoCAD is single threaded, moving to a Xeon or other more expensive enterprise processor buys me nothing except being unable to overclock, but overclocking a cheap consumer processor makes me money. I know my case is very atypical, but it's one more data point.

[Simon]

what's the point ? 99.99% of the time the machine is waiting for input from the meatbag typing with two fat fingers on a small key board (or, in this case :one fat finger as the other hand is holding an icecream cone...)

For general use I kinda agree, but I have a machine here (i7 3770k) which is water cooled and running at 4.6Ghz. That brought an autodesk automation related process I have to do from 60 minutes to a little under 50 minutes. AutoCAD is single threaded, and every MHz matters. If I have to run that process 6 times a day it saves me an hour, that allows me to get another hours actual work done that I can charge for. So for specific tasks it can actually be quite valuable.

Unfortunately because AutoCAD is single threaded, moving to a Xeon or other more expensive enterprise processor buys me nothing except being unable to overclock, but overclocking a cheap consumer processor makes me money. I know my case is very atypical, but it's one more data point.

We have ishit at work (sorrt I just can't say i7  ) so I supposedly have 8 cores, solid edge up to version 7 only uses 1 core so 12.5% of PC resource. I have now disabled that hyperthreading crap that stops my 4 cores looking like 8 cores and have received a performance boost! because now i am using 25% of my total processor for modelling. Even boinc is working better and I'm getting more work done overall. Hyperthreading crap may have worked wen you ram ran 200MHz but these days it's just marketing bollocks unless the software can specifically use it. If i remember Dave had to disable it to improve video rendering.

[Mr.B]

... unless the software can specifically use it...

And that is the whole point...
Multi CPU,
Multi core,
Hyperthreading,
Overclocking...

It is all about whether your applications will benefit from the configuration.

There are valid arguments for all configurations, it is all about what you need to achieve with the application you choose to use.

[Simon]

Hyperthreading means having one core swaping between 2 tasks and showing up in the OS as 2 cores, it does NOT mean you have 2 cores. the original idea was that because the RAM could not get data on the specific task fast enough the processor could switch to another task that was qued, so one core two instruction pipelines. but this does nothing for single threaded programs which the vast majority are and the whole purpose of the design is surmounted by sufficient RAM bandwidth.

for those arguing that RAM does not need to be that fast remember that the first computers ran an FSB and RAM at the same speed as the processor, it was only when processor speeds started exceeding ram speeds that the multiplier concept was introduced and everyone stopping thinking about the ram.

If i understand correctly when your CPU graph shows 100% this is more about 100% FSB usage, not neccessarily the actual core, this may have changed with later processors but on your old P3's and athlons was the case.

My first PC had a duron with a 200MHz DDR FSB, the ram was SDRAM 133 MHz, because they did not set up the PC properly the RAM ended up at 100MHz like the base bus frequency, but ones I upped my ram to it's correct speed of 133MHz (while the FSB was 200MHz) I got an immediate and significant performance boost, 33% infact.

You all seem to forget that RAM and CPU are the core subsystem of a computer, you need both to have a system that can even run a program even if it does not interface with the outside world......

[BradC]

We have ishit at work (sorrt I just can't say i7  ) so I supposedly have 8 cores, solid edge up to version 7 only uses 1 core so 12.5% of PC resource. I have now disabled that hyperthreading crap that stops my 4 cores looking like 8 cores and have received a performance boost! because now i am using 25% of my total processor for modelling. Even boinc is working better and I'm getting more work done overall. Hyperthreading crap may have worked wen you ram ran 200MHz but these days it's just marketing bollocks unless the software can specifically use it. If i remember Dave had to disable it to improve video rendering.

Yeah, I have hyperthreading turned off and Windows Server 2003 running in a VM assigned only 2 cores on a Linux host. I benchmarked several windows OS version on the bare metal, and several Windows OS version in the VM and this is the fastest configuration for my specific workload.

[Simon]

Of course cpu speeds and the amount of work they do per clock well overtook ram speed years ago that could only boast transfers of 8 byte in parallel, now we have 8 and 16 core machines on an increased ram bandwidth so we are still back to sqaure one if all of your cores are fully loaded. The truth is the ideal is ram speed close to that of the CPU and that is prohibitively expensive if not impossible as of yet.

[BradC]

Of course cpu speeds and the amount of work they do per clock well overtook ram speed years ago that could only boast transfers of 8 byte in parallel, now we have 8 and 16 core machines on an increased ram bandwidth so we are still back to sqaure one if all of your cores are fully loaded. The truth is the ideal is ram speed close to that of the CPU and that is prohibitively expensive if not impossible as of yet.

I saw a significant performance boost moving from a similarly clocked i5 to the i7. All down to the cache size. More ram closer to the processor.

[Simon]

Of course cpu speeds and the amount of work they do per clock well overtook ram speed years ago that could only boast transfers of 8 byte in parallel, now we have 8 and 16 core machines on an increased ram bandwidth so we are still back to sqaure one if all of your cores are fully loaded. The truth is the ideal is ram speed close to that of the CPU and that is prohibitively expensive if not impossible as of yet.

I saw a significant performance boost moving from a similarly clocked i5 to the i7. All down to the cache size. More ram closer to the processor.

you make my point perfectly, if 16GB of cache could be feasable we would all be happy.

[dr.diesel]

Back to the OPs somewhat vague question. 

I've been OCing since the mid 90s, at times taking core voltages to the extreme, and never once fried a CPU.  From helping others on sites like [H], I'd say frustration with system stability/inexperience, is the main downfall.

And as someone else listed above, quite addictive.   

[eetech00]

What is the disadvantage of overclocking?
Is it all to do cooling or is there more t it?

I've overclocked my I7-3770K to 4.5Ghz.

I think the only disadvantage is the expense of doing so.

In order to support reliable overclocking, you have to spend money on components that are designed for overclocking...MB, CPU, RAM, Graphics Card and you'll probably need to upgrade Disk, Power Supply, cooling, sometimes even a case....they're all more expensive than standard components.

Oh...and..as another poster said....its addictive.. 

[edy]

Can someone please verify this?

I heard years ago chips came out of the production line and they had no definitive "speed". The company would test them at different speeds and review the performance and would then stamp the highest reliable speed they could on the chip and sold them at that.

So what happened is when there was a greater demand for slower cheaper chips, the company would sell faster chips with just a lower number written on them to suit the supply chain demand. These could be more reliably over clocked because in actuality they were higher speed tested chips branded with slower numbers purely for matching market demand.

Any truth to this? Do batches of chips from the same die get tested and sorted and stamped with different speeds, either based on tested reliability or to match inventory with market demand?

[BradC]

Any truth to this? Do batches of chips from the same die get tested and sorted and stamped with different speeds, either based on tested reliability or to match inventory with market demand?

Yep, it's called "binning".

[Psi]

I'll up voltages as much as 10% and that is all

You often don't have to go above the manufacture spec at all.
CPUs get bin'ed by voltage too.
Lets say the CPU core is rated for 0.85 to 1.5V (like my core2 Q6600 is)

CPUs that get binned to run at full speed down to 0.85 - 1.0V are supplied to laptop manufactures due to being really efficient.
1V-1.3V might be general desktops and 1.3v-1.5v are either rejects, or get their clock lowered until they will run ok at a lower voltage then sold as that.  etc...

So if you get a CPU that's been "set" to run at 1.1V you can safely up the voltage to 1.4V (27%) and still be inside the spec.
However, what you do have to watch is the idle vs load voltage, it dips a lot at high current depending on how good your motherboard is.

My Q6600 is stable under load at 3.2ghz and 1.3V but i have the bios set to 1.45V to get 1.3V under load.
The CPU could go faster at 1.4V under load but not without exceeding the voltage spec when idling.

The top market overclocking motherboards have better voltage sense lines so this is less of an issue.

[Simon]

What I heard was that the chips that come from the centre of the wafer are the faster ones so you will always get a spread of speeds and specs from one wafer, that is why as the speed goes up a bit the price goes up faster, supply and demand. and of course those chips that really are slower will be set for stable operation so if you design your system appropriately and are willing to take the risk you can get more speed. Of course the manufacturers interest is making sure that chip is stable. Me and my employer would be pretty pissed of if I lost days of work because my processor was unstable and corrupted all the data of what I was working on......

[wraper]

Any truth to this? Do batches of chips from the same die get tested and sorted and stamped with different speeds, either based on tested reliability or to match inventory with market demand?

They are binned not only for speed but also defects. There are models with cores disabled or partially disabled cache because of the defects in them. But this doesn't mean they are binned only by results of the tests. Perfectly functional cores and cache can be disabled too to meet the market demand for lower end chips. This more often happens when the chips are in production for longer time and manufacturing process gets more polished resulting in higher yields of non defective and higher speed chips. But there is still demand for lower end ones.

[jpb]

Modern computer systems are mainly memory limited. The clock speed of processors has increased much more than that of memory. So, for example, in the past it made sense to have separate maths coprocessors whilst now the bottleneck of getting the data to and from the cpu would negate the increased processing speed. (Though now-a-days GPUs could be considered maths coprocessors but they have their own RAM.)

Personally I prefer to have a stable system that I can leave running at full power for days at a time (I used to do machine learning simulations and despite having a dual processor 8 core Xeon machine with 24GB of RAM it still took days to finish them).

But there must be a lot of headroom for overclocking as chips need to be sold with cheap coolers and to operate in fairly extreme temperatures. If you change to a better (more expensive) cooler and know you'll be running it in a coolish room then overclocking it should be no worse than running it with the stock cooler in a warm room.

I notice that on some of Dell's mobile workstations, which I was looking at on their outlet store, some had overclocked chips. The workstations are designed to be very stable (I use a tower T5500 workstation) so some overclocking is ok.

But generally overclocking seems to be more of a sport or hobby than a means of increasing performance for general use. It is like a runner improving their 100m time below 10 seconds, it is great sport but is probably not used much in everyday life (i.e. even when running away from an escaped tiger being able to do 100m in 9.8 secs rather than 12 secs probably won't make much difference!)

[Simon]

I've run ram for a very long time at a slightly higher speed, I won't do more than a small voltage increase and what I try to do is reduce timings more than anything as this does make a difference and will never damage the ram just make it unstable.

As i say the actual speed you get from ram versus it's theoretical speed has been decreasing with every introduction of new DDR technology. My ram is now under 50% efficiency so I might as well be on DDR2 rather than DDR3, I put this down to the ever increasing complexity in designing controllers that can put data in the right places so that DDR technology can work and every additional DDR level makes this more and more complicated. Also the timings have become longer and longer. i am starting to fail to see the point of DDR.

My SDRAM was 80-90% efficient in speed, DDR1 dropped to 70% DDR2 dropped to 60-70% DDR3 is not at 50% or below and I get this on different systems so it's not a system think it's called marketing bollocks.

[Simon]

Just been poking about in my BIOS, I can disable half of my cores (not hyperthreading actually switch half them off), that actually seems to give a performance boost, I wonder, less cores trying to fight for access to the same RAM ? better organization of data in RAM as only 2 processes at a time are trying to make the most of the multi DDR level complications.

[abdullahseba]

Just been poking about in my BIOS, I can disable half of my cores (not hyperthreading actually switch half them off), that actually seems to give a performance boost, I wonder, less cores trying to fight for access to the same RAM ? better organization of data in RAM as only 2 processes at a time are trying to make the most of the multi DDR level complications.

Multiple SLI graphics cards has worse performance than one graphics card on some games.

[Simon]

Because the bus between them is probably slower than the ram and cpu speed on the cards

[GoneTomorrow]

Because the bus between them is probably slower than the ram and cpu speed on the cards

Cards in SLI/CrossFire don't actually require that much bandwidth between cards at all. So little that in fact that AMD has done away with the CrossFire interconnect and just puts whatever communication over the PCIe bus along with all the other stuff. No performance degradation.

It usually comes down to how games and/or drivers are written and how (un)optimised they are for multiple cards.

[Psi]

What is the disadvantage of overclocking?

The disadvantage is that the poor semiconductor fabs can't sell the fastest cpus and make less money.  Those ivory back scratches don't come cheap

[Jeroen3]

Instead of overclocking, just buy a processor with lots of cache. Cache is the fastest memory available in the system, meaning less time waiting for the ddr memory controller.

You could overlock the cpu core to the family's maximum turbo frequency if you've fitted it with non-stock cooling without much long-term stability issues. Except from huge heat and power if you've bought one out of the cheap bin to start with.
iirc they sort the chips during testing using power consumption and voltage. Where xeon is best, celeron is worst.

[Galenbo]

What is the disadvantage of overclocking?
...

Everybody that passes his time on a serious job instead of playing around with overclocking, has extra multiple 1000 dollars to spend on a new, better or extra system.
How many milliseconds can you win by wasting some hours in settings, extra risks, problems and lost work?
:-)

[Mr.B]

How many milliseconds can you win by wasting some hours in settings, extra risks, problems and lost work?

+1
I am currently purchasing parts to build a new system.
I am purchasing top end components.
I want a fast and responsive system, but have no intention of suffering through the pain of overclocking faults.
Spend a bit extra and "it just works"...

[Simon]

I have disabled half the cause on both my home machine and my work machine with a marked increase in performance. This I am hoping saves power and I had considered overclocking my home machine slightly further given the lower heat output.

I think the jump in performance is due to the fact that less processes are now writing to the DDR memory. The problem with DDR memory is that you can only access a certain set of addresses at a time. The addresses are locked together so if one address is accessed the other seven addresses are also accessed automatically. If the other seven happened to have the information you need you are on a win and for once DDR technology actually works. However if you have 4+ cause writing data simultaneously into the RAM there is no guarantee that when one of those programs want some data back it will have any use for the other data or that the other programs will also be accessing that same data at the same time. This reduces the efficiency of the DDR memory. DDR technology in my opinion is mostly bollocks and in fact if you compare systems starting with SD RAM going through the various DDR technology levels you will find a reduction in efficiency in the bandwidth as you go up in DDR level. The advent of multicore processors has totally killed the advantages of DDR RAM albeit 2 cores seem to get decent efficiency but for or more is just asking for trouble unless of course each core has its own bank RAM but then that will have its own problems I'm sure. I am not certain if having multichannel memory means that addresses between the two different modules or more modules are locked together as well if they are this just makes life worse if they are not then multichannel is preferable over DDR crap.

[Simon]

I don't quite get the connection with your joke. There are planes engines are not processes. Each engine is designed to take a portion of the load of the aeroplane. In a multicore processor the chords do not necessarily take an equal load. Unless the software you are using is specifically written to take advantage of multiple cores a multicore processor will do absolutely nothing for you. The only advantage to a multicore processor is the fact that you have a core dedicated specifically to your task and all of the other background tasks can be done by other cores. To this end only a dual core processor actually do anything for you. The other 2 to 6 cores you spent all that money on actually probably doing you more harm. I don't understand the internal workings of a processor but as we are all aware between the processor and RAM there is a memory controller. In the case of DDR memory this is set up because of the technology so that addresses have to be addressed in blocks. If a particular address is accessed then the memory controller has to access another seven addresses in the case of DDR three. These addresses are all locked together so if you access one in the set you have to access all the others regardless of whether or not you actually want the data in the others. So imagine you have your heavy duty program like the speech recognition software I'm using right now to write this it is putting a heavy load on one core and that core is doing quite a lot of reading and writing to ram yet at the same time you have randomly another one to 7 programs in the background which are also having data written alongside the data from your heavy program that you care about. So if the processor writes data at eight programs at once when it comes to read that data back is it going to find the data in the exact pattern it needs it? Probably not and this is where the efficiency of the DDR controller comes in. Now that I have reduced my course from 4 to 2 there was much less conflict between cores trying to access RAM and there is more chance of the program that I actually care about finding the data it needs in the pattern the memory controller is able to supply it.

Yes I have noticed an increase in performance reduced from four cores to 2 obviously the overall computing power of the computer has reduced but as it actually happens I really could not give a damn about how much computing power every other background process has available when the program I care most about is only running on one core at a time and he's been denied RAM bandwidth by other programs I couldn't care less about which can be served just as well with one core rather than three. Obviously this is not a silver bullet solution but I have up my chances in getting better performance. As I say I don't know how multichannel RAM access works and if addresses between the two or more RAM modules are also locked together or if each RAM module can be independently addressed this would in fact help ease the problem otherwise it just compounds it.

[Simon]

Well my perception is that the machine is slightly faster and more responsive. I'm not saying that It's had a huge boost and can do loads more but if i don't need 4 cores and  2 are plenty and it gives me more edge with power saving i see it as an improvement. The need for more than 2 cores is rare, I see 4+ cores as just marketing unless you really have the software to use it.

My work machines iShit processor is now running 2 cores rather than 4 and hypercrap has been turned off, the machine is now much more responsive and it was bad before.

At the end of the day halving the cores is doubling the cache for each