Im looking for knowledge. I would like know how the different aspects of overclocking relate to each other. What does the Multiplier do and how does it relate to the FSB, ram speed, cas latencies and the final cpu speed. Why do you lower the multiplier to get higher performance, and why do the cas latencies have to then go up to achieve stability? Does the FSB have to be at or below the ram speed inorder ot be stable or do they have nothing to do with each other. Why are the fx chips so much better for oc'ing than the other just because thier mulitpliers are unlocked both ways, dont you lower the multiplier to oc? thanx daniel
overclocking and how it works
- Thread starter newmenu
- Start date
Look at the post that with the sticky at the top of this forum. It is a well written guide to get you into OC without f'ing up your stuff.
it didnt really answer my question/s. im looking for a deeper understanding of whats going on in their realationships to each other and whats happening to each side when one is changed. I guess only an engineer would be able to elaborate.
Not really, if you understand the functions of the parts of your system, and read the OC guide in the top of the forum, most all of the questions included in your post will be answered. When you are done reading up, no doubt there will be more questions. That's where the rest of the community will be there for you.
Zebo's guide might be ok for someone who knows what he's doing.
Howevere, for a newbie, that guide isn't very helpfull, since its too technical, and too little elborated/explained there.
I know it 'cause I had the same problem with it.
Now, instead of being smart a$$es, and send the guy to read some thread that most of you didn't even bother reading, 'cause else you'll know it doesn't answer this guy's Q's., just answer his godamn questions...
Peace MOFO's
Howevere, for a newbie, that guide isn't very helpfull, since its too technical, and too little elborated/explained there.
I know it 'cause I had the same problem with it.
Now, instead of being smart a$$es, and send the guy to read some thread that most of you didn't even bother reading, 'cause else you'll know it doesn't answer this guy's Q's., just answer his godamn questions...
Peace MOFO's
You've got a sweet rig, newmenu. I'll handle the questions that I know clear answers to. I could answer all of them in person, but some of my thoughts don't go into writing well, so here goes:
FSB and ram speed don't need to be the same. That is where you'll hear people talk about "dividers." FSB (or HTT in the case of A64-interchangable nomenclature for this explaination) is how fast the CPU communicates with the memory. The rate that the CPU requests things from the memory can be different than the speed the memory runs at without a problem anymore. For instance, my HTT is 300, but my ram is running @ 200.
Lowering the multiplier doesn't help as much as it used to. The reason that it's useful is that by lowering it, you can reach higher FSB/HTT with the same overall clockspeed. Higher FSB/HTT is useful because CPU performance is often somewhat limited by the speed at which it gets its information from the ram and if it's working faster, then that widens a bottleneck.
FX chips are not that much better for overclocking. The nice thing about being able to increase the multipier is that it really simplifies the equation for OC'ing. You don't have to worry about ram speeds, HTT, AGP/PCI/PCI-e locks, HTT Multipliers, etc. You just have one setting to worry about. BTW, your chip is the same thing as an FX-53. Look it up, same speed, same cache-same chip, essentially.
By reading this, you know that you don't have to lower the multiplier to OC. If your goal is to get to 2800 (nice, round number-not necessarily possible), you could go 12x233 or 10x280. With older architectures, the 10x multipier would be much faster. It still would be a little faster, but less pronounced-but you don't necessarily have to change the mulipliers at all (otherwise Pentium 4's wouldn't be overclockable).
FSB and ram speed don't need to be the same. That is where you'll hear people talk about "dividers." FSB (or HTT in the case of A64-interchangable nomenclature for this explaination) is how fast the CPU communicates with the memory. The rate that the CPU requests things from the memory can be different than the speed the memory runs at without a problem anymore. For instance, my HTT is 300, but my ram is running @ 200.
Lowering the multiplier doesn't help as much as it used to. The reason that it's useful is that by lowering it, you can reach higher FSB/HTT with the same overall clockspeed. Higher FSB/HTT is useful because CPU performance is often somewhat limited by the speed at which it gets its information from the ram and if it's working faster, then that widens a bottleneck.
FX chips are not that much better for overclocking. The nice thing about being able to increase the multipier is that it really simplifies the equation for OC'ing. You don't have to worry about ram speeds, HTT, AGP/PCI/PCI-e locks, HTT Multipliers, etc. You just have one setting to worry about. BTW, your chip is the same thing as an FX-53. Look it up, same speed, same cache-same chip, essentially.
By reading this, you know that you don't have to lower the multiplier to OC. If your goal is to get to 2800 (nice, round number-not necessarily possible), you could go 12x233 or 10x280. With older architectures, the 10x multipier would be much faster. It still would be a little faster, but less pronounced-but you don't necessarily have to change the mulipliers at all (otherwise Pentium 4's wouldn't be overclockable).
thank you for your response superkdog. I did spend about 2 hours reading through the oc link and learned alot, but lack the insight to deeply understand it; how it really works and what's actually happenening in the lines and circuits and how real world performance will be affected by these adjustments. Not sure if Im right brained or left brained, but im opposite that of the liks author. The only thing Im still curious of is ram timings. Assume i dont know how to test stability short of system shutdown and want to get a core speed of 2.6 while retaining 2-2-2-10 ram timings.
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
Total memory speed is measured as bandwidth, Sandra memory bandwidth benchmark is the accepted standard for measuring this. Memory Bandwith is the combination of bus speed and timings(lower timings = more bandwidth). Ram performance is determined by the give and take between speed and timings to produce the maximum bandwidth, much like the relationship between HTT and CPU multi determine total CPU mhz.
Real world performance wise, overclocking the CPU always provides a greater increase than overclocking your ram on A64 systems. Most all the top o/c's on A64's are acheived using memory dividers which lower the memory bus speed to less than HTT.
Bottom line, run your highest CPU overclock, and use a divider to run your ram at the highest speed that you can maintain reasonably tight(low) timings
Real world performance wise, overclocking the CPU always provides a greater increase than overclocking your ram on A64 systems. Most all the top o/c's on A64's are acheived using memory dividers which lower the memory bus speed to less than HTT.
Bottom line, run your highest CPU overclock, and use a divider to run your ram at the highest speed that you can maintain reasonably tight(low) timings
ok, im going to ask this carefully. So if oc'ing your cpu by dropping your multiplier and raising your HTT your getting a higher clock speed and wider bandwidth. But, by oc'ing your cpu using a higher HTT/FSB your conversly oc'ing your ram, because its performance is determined by the HTT/FAB and timings(2-2-2-10), right? This K8n Neo2 Plat mobo has this D.O.T. core cell chip that apperantly oc's your sytem automatically and tweaks it some as loads change, optimal settings idle, oc settings under load. Would i really be better off using this auto oc for dummies DOT or what.
Ok, so your effective ram speed is FSB/HTT doubled(DDR Double Data Rate) but the actual speed of the ram itself is still 200, buy increasing your HTT/FSB your increasing the effective performance of the ram and not the actual speed of the chip right?. FSB/HTT of 200 will mean your DDR ram is running at DDR400. With intel your ram is DDR2 so its speed is your FSB/HTT times four right. and thats how they say their chips can run at 3.8Ghz huh, because of the DDR2 qaud pumping data instead of duel pumping. I read the reviews for my ocz plat rev2 and the tests were set up on an intel system giving me little to work with. So Im thinking Im gonna set the Multiplier to 11 x 233 HTT/FSB and set my ocz plat rev2's to 2-2-2-10 1T 2.8v and I should be set huh. or does 10 x 260 HTT with 2-2-2-10 1T at 2.8v sound better, lastly would 12 x 216 HTT with 2-2-2-10 1T at 2.8v be best? The cpu voltage does not change anywhere in here does it. again thanx for all the feedback and help
In an AMD64 system,
HTT -- bandwidth between CPU and RAM(s)
RAMs effective bandwidth i.e. 400MHz -- bandwidth it communicates to the rest of the component
Lowest CAS setting RAMs -- equivalent to quickest reaction/acceleration
Highest FSB (including RAMs) -- equivalent to the top speed the system can reach, modified by the CPU's multiplier.
The actual performance is bound by all the components in the system, from the CPU's architecture to the type of harddirk used.
HTT -- bandwidth between CPU and RAM(s)
RAMs effective bandwidth i.e. 400MHz -- bandwidth it communicates to the rest of the component
Lowest CAS setting RAMs -- equivalent to quickest reaction/acceleration
Highest FSB (including RAMs) -- equivalent to the top speed the system can reach, modified by the CPU's multiplier.
The actual performance is bound by all the components in the system, from the CPU's architecture to the type of harddirk used.
RichUK
Lifer
- Feb 14, 2005
- 10,341
- 678
- 126
DDR - just means that it effectivly sends double the amount of data per one cycle instead of just one single amount of data like --> (SDR - Single Data Rate the old 168pin ram), if you look at an oscilloscope, you will see an analogue signal, the side ways S is one cycle, so when running at 200Mhz (200 million cycles per second) DDR will send data on the up curve and the down curve (+ and - curve), so this is effectively 400Mhz aka DDR400 pc3200 (PC3200 = 3,200 MB/s, or if you run dual channel that is affectively 6,400 MB/s - using 2x64bit mem controllers aka AMD's 128bit on die mem contoller), SDR will only send data on the up curve, so when you see say SDR pc100 this is running at 100Mhz effective, but that is old hat now, im just using to help explain...
DDR2 - this RAM is the same as DDR (meaning that is still sends double the amount of data per cycle) but it uses different technologies (different modulated carrier waves) to be able to transmit data at higher frequencies i.e. DDR2 667 i think its at now ... AMD dont require the use of DDR2 yet and trust me DDR2 is not that amazing at the mo, since to run those speeds it requires really loose timings of like 5-5-5-12 or somthing like that, when DDR2 matures (new tecnnologies to run tighter timings AMD will be looking to adopt... AMD will possibily use this on their future chips, or as roumer has it they might skip to DDR3...
Quad pumping by Intel is theoretical, they use a technology to manipulate the carrier wave (analogue cycle), to be able to send 4 bits of data per cycle, so when looking at DDR being able to do 1 bit of data on the up curve and one bit of data on the down curve (resulting in an effective double data rate), Intel manipulates the carrier wave to send an extra bit of data on both the + and the - curve (up and the down curve), so that results in 2 bits of data on the up curve and 2 bits of data on the down curve, 4 bits of data in total on one cycle, so 4 x 200Mhz (effective) = 800FSB which you will see stamped next to all of the new intel P4's as there FSB rating ... Intel use technologies like this to ofcourse increase the bandwdith to the processors, but the FSB rating is set into play by jedec which state 200FSB (effective) is the highest FSB that can be used ... although i think the 1066 Intel EE chips where using 250FSB but done quote me on that ...
HTT - This is AMD's version of increasing the bandwidth from the stock 200Mhz HTT, i dont know a great deal about it, but i do know that it works as a link, and is known as a HTT-LINK, the basic way this works, is in the BIOS you have in affect a Multi for the HT/LT, the chip you have can handle 1000Mhz HT or a 1Ghz HyperTransport, so with the standard 200Mhz it has a 5x multi which equates to 5x200Mhz = 1000Mhz, you will also see advertised 2000mt/s, this is 2000 million transfers a second, which i dont know how true this is but i suppose mathmatically and theoretically this is the capability of the 1Ghz HyperTransport, (dont get this figure mixed up with anything to do with OCing because it is irelevent), when OCing you must make sure you keeps this at least below 1050Mhz (some people are able to run this higher but it really doesn't make much of a difference since AMD chips do not use all of this bandwidth anyway), you will have to work it out to make sure it is stable, for instance with me i am running 253HTT x 4HT/LT which is 1012Mhz total, which is exceptable..
With the relation of the CPU, RAM and FSB/HTT, you have the RAM working at 200Mhz which you will see in the BIOS as default (this will always be the default FSB/HTT on modern PC's using atleast pc3200 RAM), and then you will have the CPU Multi, the FSB/HTT is always 200Mhz by default (this is actual speeds not taking DDR speed into it) the multi just multiplies the FSB/HTT to equal the CPU clock, which is then controlled by the internal clock crystal.. For instance:
AMD 3000+ = 1.8Ghz -> 200Mhz x 9 (using the 9x multi)
AMD 3200+ = 2.0Ghz -> 200Mhz x 10 (using the 10x multi)
AMD 3500+ = 2.2Ghz -> 200Mhz x 11 (using the 11x multi)
AMD 4000+ = 2.4Ghz -> 200Mhz x 12 (using the 12x mutli) + 1Mb L2 cache
(the AMD 4000+ is a rebadged FX53 with a locked 12x multi)
AMD have locked the Multi so that you cant increase it, becasue then you can just have a 3000+ then just up the multi by 1 or 2 so that you can gain faster speeds aka 3200+ or 3500+ spec speeds. So all you pay for is a higher multi, because for instance the winchesters 90nm, are all the same just with a different Multi's (and ofcourse they all clock different, you just have to cross you fingers and hope for gooden)
However, the Multi can be lowered to allow you to increase the FSB/HTT to be able to get more precise CPU clock, so instead of just going up in 200Mhz increments you are able to go up in small increments to find its absolute braking point (highest clock)....
One thing you must keep in mind with RAM, is that you can not keeps the same tight timing as they were when they were default, trust me if you run 260HTT with 2-2-2-10 or even at 225HTT you will have to clear the CMOS staright away because it wont even POST (just for your info it is at the bottom of your MOBO under the orange PCI slot, you have to switch the jumper) ... i have the same RAM as you can see by my sig i am running pc4000 spec 253Mhz effective at timings of 3-4-3-10.. the CPU is max stable at that speed and those timings.
I wont bother explaining how you can OC your rig or what steps that you should take, becasue Zebo did an excellent guide which is stickied at the top of the forum ...
With your ram timings i would try the following (keeping a 1T command rate)
--The timings below are with regards to OCZ Platinum Rev2 (2-2-2-5)--
200 - 215 HTT = 2-2-2-10
215 - 240 HTT = 2.5-3-3-10
240 - 260 HTT = 3-3-3-10
260 - 280 HTT = 3-4-3-10
Anything over and you will have to use a 2T command rate at 3-4-4-10 untill the RAM tops out.
As Jeff7181 has kindly stated (below) with different RAM (different chips and PCB i.e. totaly different DIMM) results in different tolerances which result in different timings so like stated the timings above are a guideline for OCZ Platinum Rev2's.
I beleive the best DIMM chips at the moment are Samsung TCCD and BH5 (cant remember the manufacturer for BH5) ... When purchasing RAM with the intent to OC just a Note: purchase from a well known and reputable manufacturer, because although they may use the same chips, the PCB (printed circuit board) is very different and you will find that the better makes are able to keep down electronic noise (with the design of PCB) when running high volts and frequency, which inturn equals stability and quality of signal. Also another thing even with identical sets of DIMM's they dont always behave the same, take mine for example they should be able to run 3-3-3-10, but i am unable to i have to run 3-4-3-10 which the CPU likes, if i wont to run 3-3-3-10 i would have to lower the CPU clock but i dont want to do that because CPU speed is the ultimate goal!
You do not need to run dividers because you have good spec RAM and you will easily be able to run 1:1, or 200 in you BIOS, but in the case that you did then again you can refer to Zebo's guide for instructions...
I would recommend that you get the latest official BIOS either BIOS rev 1.4 or 1.5 or you can try one of the modded ones at you own risk, i am soon to try BIOS mod v1.4 to see if this will help with my OC, as i have heard that this is the best modded BIOS for the NEO2 plat,
I would also recommend that you use the full 2.85v for the dimms as they will except 2.9/3.0 under warranty, and also you will wont to use some software to monitor and and test.. i would recommend
Everest for complete info on rig (not that you need it for info, but it is a good tool OCed mem specs etc)
MBM5 to monitor case and CPU temps
CPUZ for accurate mem and CPU specs
I use Stress Prime 2004 to bench for CPU/RAM stability, but most people use prime 95 (these apps are the same just SP2004 has much better GUI)
Anyways i hope this helps on your OC venture, and i hope that the information that i have provided is helpfull and fills all the gaps you dont understand.
Good Luck :thumbsup:
EDIT: i forgot to mention that with you chip, you asked the question on cpu voltage, if you check Zebo's guide you will see that 130nm chips like yours can handle upto 1.65v on air with still being safe, depending that temps do not regularly exceed 60 degrees, as this will shorten the life of you cpu and potentialy damage your cpu, at this point you will have to think about adequate cooling to combat this.
With regards to RAM timings, CAS (Column Address Strobe) is the most important timing (this is the timing that you will be able to adjust between 2, 2.5, and 3), the difference from running you ram at CAS 3 rather then CAS 2 is 33% worse, but running a higher FSB, for instance like me i have gained over 25% for the extra 53Mhz on top of the standard 200Mhz, so it is easily weighed up if you get the timings right...
Jeff7181
Lifer
- Aug 21, 2002
- 18,368
- 11
- 81
Nice, RichUK. The only thing I'd change is your list of RAM timings. That may work for a specific type of RAM, but it is by no means applicable to all, or even most RAM. The best way to figure out what timings to start at the timings the RAM is rated for, and then adjust from there. All RAM is not created equal, even the same model with the same chips and the same PCB.
this has been extremely helpful. thank you all. now that ive the system up and running, and very well i would say, Im haveing issues getting cirtain periferalls to be installed, i have a topic in tech support and help would be even more appreciated. I cant get my Audiology2 value to take its drivers, HP printer is doing the same thing, and my mouse freezes up occasionally, yet unplugging and replugging temporarily fixes that. Please help if you can, Such a dang shame to have such nice stuff and no sound, printer, working mouse to go with!!!!
RichUK, for good pamphlet thingy.... good job. One thing I saw though... SDR mean Sychnrounous Dynamic Random (the last two in the name are Access Memory) and DDR is actually DDR SDRAM. By your acronym that would mean DDR is actually Double Data Rate Single Date Rate Access Memory.
im sorry to nit-pick it's just that everything else looked very good and I don't think that fit in there
im sorry to nit-pick it's just that everything else looked very good and I don't think that fit in there
RichUK
Lifer
- Feb 14, 2005
- 10,341
- 678
- 126
Yes you are true silly mistake really ... thanks for correction
... I had a boring day at work when i did this so i just kept writing ... the reason i wrote DDR and SDR was because many of the UK sites list the different types of RAM like this ..FYI .. as i know it SDRAM and SDR terminolgies are different things
Sychnrounous Dynamic Random Access Memory - as i can remember it just synchronizes itself with the CPU's bus as i am shore you are aware
SDR - is just what the OP would recognise the RAM as when browsing retail websites i presumed..
Although i stand corrected and should have stuck to correct terminology as i did with the rest of the post
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter