Sign in with
Sign up | Sign in
Your question

Voltage + ? = Processor Frequency

Last response: in CPUs
Share
February 1, 2006 6:34:21 PM

Hi!

Here's my 2 cents question:

You input a specific voltage into the processor, right? What happens then (at the die level) which states that "Aha! I must work at 2GHz!"? A clock? Where is it, then?


Anyone?!


Cheers!
a b à CPUs
February 1, 2006 7:10:12 PM

wrong...

It's FSB X multiplier! :-)
February 1, 2006 8:30:06 PM

Ok.

Let me put it another way: the 'multiplier' changes the FSB (in an Intel sys) input voltage, right? Now, even the FSB has to have a "clock" generator (just like my wristwatch has a piezo-electric crystal which sets its frequency...); and, you can change the processor's core voltage on its own. So, my point is: where's the «piezo-electric crystal» within the processor?


Just something i haven't been able to explain to myself... and probably, a simple & obvious... something.


Thanks for the attention.
Related resources
February 1, 2006 8:55:58 PM

Inside your chip, is what's known as a Crystal Clock (that's how I know it, that's not what it's really called), where it says "Hey, run at this speed!", and it does. This is what those programmers over at AMD and Intel spend time doing, I cannot say how they do it exactly, as I do not know, but that clock speed (200MHz on newer chips) is the set speed it runs at, and they introduce what's known as Clock Stepping (Multiplier) where it keeps pace the speed externally (Front Side Bus) and increases the speed internally (CPU). This was first introduced due to, I believe, thermal reasons (probably included others being this was years ago.)

The Multiplier changing does not change your voltage on its own, the multiplier simply increases the internal clock frequency of your chip (called overclocking ^_^). The part of the CPU where this Crystal Clock (as I call it) exists I do not know, but I'm sure if you ask nicely, somebody here will tell you ^_^. Not sure how much help this was, anybody else here please enlighten further if you know.

~~Mad Mod Mike, pimpin' the world 1 rig at a time.
February 1, 2006 9:12:33 PM

MadModMike is correct! :D 

Changing the FSB or multiplier does not change the voltage.

Changing your voltage is doable on some OC friendly boards but is not recommended because increasing the voltage not only generates a lot more heat it can also cause stability problems, errors and could even fry your CPU.

If you decide to change your voltage make sure you do it in TINY increments and NEVER set it to more than +- 10%
February 1, 2006 9:33:22 PM

Is there any part in modern CPUs that actually runs at frequency higher than Yx10^6Hz(1<Y<1000)?
February 1, 2006 9:33:56 PM

Quote:
MadModMike is correct! :D 

He's not the real MadModMike. :wink:
February 1, 2006 9:35:11 PM

Yes I am, I'm the fat guy with a jetpack on who goes around and pimps out computers with nvidia cards ;) .

~~Mad Mod Mike, pimpin' the world 1 rig at a time.
February 1, 2006 9:36:35 PM

Quote:
Is there any part in modern CPUs that actually runs at frequency higher than Yx10^6Hz(1<Y<1000)?


Umm.. What?

~~Mad Mod Mike, pimpin' the world 1 rig at a time.
February 1, 2006 10:07:13 PM

Ok, if you are overclocking, do this:
-Set CPU voltage to 1.5 since you probably have a venice core.
-Put hypertransport link at 3x, which is 3x the fsb and must equal at or under 1000. So for example, 5x200fsb would be safe, but 5x250 would not be stable.
-Remember to use memory dividers if necessary.
-Turn cool'n'quiet off.
-To get the desired clock speed, your FSB times your multiplier (in your case 10) will yield the ghz speed of the cpu, for example 200x10=2000mhz, or 2ghz.
By the way, 3200 venices will max out on air at 2.8. You probably won't even need to change the stock heatsink because they run so darn cool! I have one, and I bought an XP90 with a Vantec Tornado...needless to say it was overkill :p 
February 1, 2006 10:34:11 PM

Thanks.

It is the part you don't know i would like to know!
Again, inside my wristwatch sits a tiny piezo-electric crystal, powered by an also tiny battery. This battery sends a [tiny!] voltage to the crystal which makes it 'tick'; each 'tick' is a cycle (measured in Hertz, Hz) per units of time (seconds, s). [More exactly, a 'tick-tack' is a cycle...]. There's one in each MB as well, called Real Time Clock (RTC), which does - exactly - the same thing as my wristwatch: measures time through crystal pulses. Everything's fine, up to here.
Now, is there something like this, inside the processor? I'm sure there is (otherwise...); but, WHAT?

[As for the 'Clock Stepping' (or 'Multiplier'), if you do increase it, you're also increasing the voltage supplied to it; the same with the FSB (on Intel sys...). It translates into more cycles/s = more Hz/s; the processor just 'tries' to keep in sync with the bus, through some... multiple.]

Still, my doubt remains.

Thanks, anyway.
February 1, 2006 11:02:29 PM

errr.. the clock generator? I think that's where the multiplier is set as well as the other frequencies like the internal frequency (the FSB) and ...i don't know i'm just guessing from pieced together bits of research from way back.
February 1, 2006 11:04:29 PM

Quote:
wrong...

It's FSB X multiplier! :-)


Humm so, since there is no FSB on AMD64, how do they work??? 8O
February 1, 2006 11:12:37 PM

It is core clock X multiplier :D 
February 1, 2006 11:16:38 PM

Quote:
It is core clock X multiplier :D 


Ok.. so where does the core take its clock?
February 1, 2006 11:17:24 PM

Apparently they have a thing called a HTT, which the same thing. Not sure what's up with overclocking those as I have had no experience with A64, sad I know.
February 1, 2006 11:19:01 PM

From its original frequency. and when you mess with the original core frequency you mess with people's minds :?
February 1, 2006 11:19:14 PM

Quote:
Apparently they have a thing called a HTT, which the same thing. Not sure what's up with overclocking those as I have had no experience with A64, sad I know.


no its not the same thing
February 1, 2006 11:23:08 PM

Could someone please explain how the A64's work?
February 1, 2006 11:29:12 PM

Simple. since the memory controller is intyegrated to the CPU, there is no need for FSB. Since it integrated in the CPU, the MC(memory controller) is connected directly to the memory from the CPU. With other system, the cpu has to take the "front" bus to the cchipset which contain the MC and the "side bus to the memory.

But the MC has to retain a connection to the rest of the system, so it needs to be connected to the chipset. this is the role of the Hypertransport link.

Understand that?
February 1, 2006 11:42:31 PM

Joset, I think you're not understanding something. On a processor, there is a clock setting inside the CPU programmed into it by a AMD or Intel programmer. It does not matter where this physically resides because you aint gonna edit it. On Intel and AMD cpu's the clock is set to 200MHz (excluding Extreme Edition and older CPU's). On Intel CPU's, they use a standard Front Side Bus where that 200MHz is sent 4 times per second (using all 4 sides of the Square Wave, also known as Quad Data Rate). On AMD64 CPU's, they use HyperTransport Links, which replace traditional Front Side Bus's and they also have the Memory Controller on the chip, vs. in the Northbridge on Intel CPU's.

The clock signal that runs at 200MHz is a pre-defined speed that is how fast your computer is running. Before there was no Multiplier (Clock Stepping), that was how fast your CPU ran, at w/e the speed was it was set to by the programmer (in this case, it's 200MHz). But because 200MHz is obviously not enough speed, and setting a crystal clock of 2GHz would melt the CPU, they introduce Clock Stepping where they are able to increase the times the 200MHz crystal pulses per second and also control the pulses of where and how it communicates to the Northbridge w/o destroying the CPU.

1 important thing to note is the multiplier inside the CPU that makes it operate at 3GHz has nothing to do with the multiplier on HyperTransport or the Quad Data Rate on Intel FSB, Clock Stepping is the increase in the speed of the internal clock (being the 3GHz in this case) and the FSB or HyperTransport is independent itself from the Clock Stepping inside the CPU, but only to the extent where the higher the multiplier, doesn't mean increase in speed to HyperTransport or Front Side Bus.

The HyperTransport links operate on either 200MHz with a 4x Multiplier (800MHz sent each way) or 5x Multiplier (1000MHz sent each way). Also, the traditional Front Side Bus is actually made up of 3 parts, a Data Bus, Address Bus, and a Command Bus. 1 thing to note is, only the Data Bus is sent 4x 200MHz for 800MHz FSB, the Address & Command Bus only operate at 1x 200MHz speed. On HyperTransport, all the information in those 3 bus's is stuffed into 64-bit packets sent on 16-bit wide bus's (1 each way, sent at the same time, vs Intel FSB where it's sent either direction at 1 time, not both at same time).

The Memory from the Memory Controller on Intel CPU's is accessed at 200MHz Double Data Rate (400MHz Effective) where it accesses RAM at 400MHz and receives at 400MHz, UniDirectional (not at the same time). On AMD64 CPU's, it accesses information at 1GHz and receives at 400MHz BiDirectional (at the same time). This allows less latency and combined with HyperTransport, the effiency and bandwidth of an AMD64 system vs. Intel P4 system, is greater on the AMD64. To sum it up, Intel CPU's have a Front Side Bus, AMD64 systems do not have a Front Side Bus, it has been replaced with HyperTransport.

On Intel systems, the Front Side Bus accesses the Northbridge which than accesses the Memory Controller inside the Northbridge and tells it to get what it needs from Memory, after it receives the information, it than takes the Front Side Bus back into the CPU to process the information. Whereas on AMD64 systems, the CPU directly accesses the Memory w/o needed to wait for the Northbridge to get the information from the Memory Controller, because the Memory Controller is already on the CPU in AMD64.

And for the record, Joset, Volts and Multiplier have no correlation to each other, other than the obvious of more power is needed for a higher frequency. If you are looking for a more indepth definition of how a computer processor works physically, you're going to have to look elsewhere, as you're hardpressed to find somebody here who knows that, unless they looked it up.

~~Mad Mod Mike, pimpin' the world 1 rig at a time.
February 2, 2006 12:07:03 AM

:trophy: :trophy:

Go MadModMike :D 

Well said!

Some of us did go 2 school ;-) but some didn't. Some people don't know what an ALU or a decoder is.
February 2, 2006 12:11:03 AM

Thanks Linux_0, what I'm studying now is CPU Registers and more about the physical archtectures of them, I find the hardware to be the most interesting part of computers, and I seem to enjoy building PC's more than using them (lol weird eh?).

~~Mad Mod Mike, pimpin' the world 1 rig at a time.
February 2, 2006 12:12:01 AM

As Pat said and simply put, AMD implemented Hypertransport (HTT, not A64) as a dedicated, serial point-to-point, bi-dir interconnect bus (as a remark, AMD & partners made it 'open', meaning it can be further improved & adopted by those who adhere to the HTT Protocol...); it works much like PCIe and replaces the MCH (Memory Controller Hub, also known as the "Northbridge") in Intel-based systems, which use a parallel, shared bus (another note: Intel will, most probably, use the Common High-Speed System Interconnect (CSI), within two years from now...). HTT links memory directly to the Memory Controller, which resides in the chip's die.

For a light glimpse on HTT: http://www.theinquirer.net/?article=27317

A simplistic view on HTT... and my topic remains unresolved.


Cheers!
February 2, 2006 12:14:12 AM

you deserve my admiration for taking the time to type all this.. I was going to suggest Google to search for CPU and Motherboard clock.....

I was too lazy to explain..
February 2, 2006 12:26:57 AM

lol :D 

I do both. I build them and use / admin them... I also break them ;-) :lol: 

You ever try Linux or BSD?
February 2, 2006 12:28:34 AM

Joset, thank you for cramming my 6 paragraphs into 1, but those exact words came from a website, I know this because I read that same thing a few months ago, thanks for the refresher though, I had forgotten what Intel called their memory controller.

Linux: We use many variations of Linux in my Microsoft Certified Systems Administrator program (weird, eh?) and I prefer Windows, only because of software support, but I do believe Windows to be as stable as a Prescott CPU under 30c (Hell musta froze over if that happened, lol). Only thing that sucks is that we have nearly 40 PC's in our lab in which it feels like we reinstall Server 2003 every single day, due to some mishap with DHCP or a configuration gone wrong.

BTW, I suck at crimping RJ-45 cables, but I am the master of ghettoniss. Can't beat making the wires 5" too long and than crimping the connectors on and taping the plastic shielding back on the wires to cover 'em up, I'm so cool 8).

~~Mad Mod Mike, pimpin' the world 1 rig at a time
February 2, 2006 12:36:57 AM

Quote:
The Memory from the Memory Controller on Intel CPU's is accessed at 200MHz Double Data Rate (400MHz Effective) where it accesses RAM at 400MHz and receives at 400MHz, UniDirectional (not at the same time). On AMD64 CPU's, it accesses information at 1GHz and receives at 400MHz BiDirectional (at the same time).

I don't mean to be picky, but I believe you got the buses mixed up. Intel's FSB is bidirectional, while HyperTransport is unidirectional. The difference is that the FSB design is half duplex, while HyperTransport is fully duplex. This means that a single FSB can send data in either direction, but only in one direction at any one time, while HyperTransport uses two buses, each one permanently sending data in a given direction.

http://www.hypertransport.org/tech/index.cfm

I also thought that the memory to memory controller speed on Intel systems was 266MHz, 333MHz, and 400MHz for 533MHz, 667MHz, and 800MHz in Double Data Rate Mode versus 200MHz for 400MHz DDR in AMD due to the use of DDR2 in Intel systems. Intel systems have the memory bandwidth advantage, while AMD systems have the latency advantage. This fits into the Netburst philosophy of raw power versus the K8's efficiency.
February 2, 2006 12:38:32 AM

LOL.... with Linux you have to screw the system up REALLY badly to have to reinstall it.

Once a day??? Geez... that's a lot
February 2, 2006 12:41:40 AM

When I said "BiDirectional" and "UniDirectional", I was meaning Full Duplex and Half Duplex, but I should have been more thorough, thank you for pointing that out. Another correction I need to make, is the packets are 32-bit, but if the "job" requires it, HyperTransport can send 64-bit packets.

Linux: It's because other people are idiots. "Hey, here's a .SYS file, let's delete it!". But we use BCCD and Knoppix mostly, with BCCD we clustered a few computers, but couldn't find software to use for benching or performance testing, if you know of any, that'd be nice to know.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
February 2, 2006 12:53:23 AM

Although you replied to Pat, i must congratulate you for your thread. I would say that - apart the small 'slice' which concerns my topic - it's a wonderful piece of info. Thanks for the time taken.

I didn't ask for an explanation (although enlightening as it was) as how a CPU works; I know the "hows". My quest has - only - to do with «On a processor, there is a clock setting inside the CPU programmed into it by a AMD or Intel programmer. It does not matter where this physically resides because you aint gonna edit it.» you mentioned. Certainly, it matters what & where this "unit" is and how it works. It's the only piece i'm interested in right now and, despite my search, i came out empty handed.
I'm also sure that programmers (either Intel's or AMD's) have something to do with it though i believe it's more a matter of design & engineering...
You would agree that a programmer would be useless programming a piece of hardware he doesn't know what or where it is, wouldn't you?

I don't intend to edit it. Just out of curiosity...
Maybe, it's harder a question than i thought. I'll keep searching, for sure.

Thanks, anyway.


Cheers!
February 2, 2006 12:56:30 AM

When you first posted, you sounded like you were going to try and do something with it (beyond me how you would). And yes, the designers of the CPU know where it is, but with each chip it is physically different, only way to really find out that indepth of information is to talk to a person in the company, or get lucky and read a post from somebody who really knows the answer to where it lies in the chip, good luck either way.

Linux: Thanks, gonna check themz out :) .

~~Mad Mod Mike, pimpin' the world 1 rig at a time
February 2, 2006 2:54:28 AM

Amazing info! Thanks a lot, you really cleared out a lot of doubts I had! :D 
!