Infinitely adjustable HTT/Dividers/Multi

[charlie]

I've been thinking... why can't mobo mfg. have an option that allows "infinitely adjustable HTT/Dividers/Multi??

What am I thinking?
It goes something like this:
My rig has tccd and has a sweet spot around 290-305HTT, 1:1..... but at 10 multi it's 2900mHz-3000mHz....my chip will do 3150mHz....and if I use 11 multi, it drops me out of my ideal "range".... (examples of course)

Why can't we CHOOSE our own HTT/Mem speed.... precisely.

A motherboard should be able to have a CVT, like a constant velocity transmission in a car. It should match the divider to the multi to the mem speed.
So if yo have 265max on your BH-5 and your CPU will do 3150, you can set in BIOS 3150/265 = 11.89multi :D

Now about dividers... if you have a 3200+ with a 10x multi, and your CPU and board are good to 320HTT, you have to choose the 5:4 divider...that knocks your memory speed down to 256mHz, much too LOW for your shiney NEW G.Skill, right? You should be able to specify your divider, EXACTLY. Set in BIOS, Asynch, 320>295 always keep your mem at it's MAX, manipulating the Multi/Divider......

I'm not really good at explaining this stuff.... whatcha' think? Maybe CC to Oskar Wu, eh?

C

[P_1]

ive thought of this too, why cant they just auto set the dividers by having seperate controls for htt, and ram fsb. But for cpus theres no way u can get infinitly variable multis because amd sets the number of multis and what multis are availible.

[kiwi]

Yeah, would be nice :D
Maybe board manufacturers have something similar in their laboratories, who knows.

[STEvil]

Its all generated by PLL's I think, so the more multipliers you want the more complex the PLL's get.

[esdee]

charlie

394x8 gives the CPU 3154MHz
the 150div on the mems gives the mems 295
isn't something like this enought?

there are so many options with dividers to play with. and include the .5 multis!

oh inf div's would be cool ;)

[Ref]

I would at least to be able to select a specific CPU/RAM divider. for example i'm at 11x200, and i will select 9 divider to be able to have 244 MHz on RAM.
Or for example I'm at 3000 MHz (300x10) and i will select divider 11 because my ram maxes out at 275 2-2-2-5.


I hate those 100/133/166/200 options, why there is not an option to select specific divider ? :rolleyes:

[trans am]

this is why I think everyone should have a set of good tccd and a good set of 2-2-2 ram. (ex. vx)
just gives you more sweet spot options.

but I agree with prince charles, infinite options would rule! has anyone heard about the future clockless asynchronous cpu? what does this mean for overclockers?

http://www.arm.com/news/6936.html

[jikdoc]

wow charlie,

is your fx on air still at 3150? or did you get it under phase/di/water?

[felinusz]

This would be really great for overclockers everywhere.

However, the question manufacturers ask, is "What's in it for us?".

We are such a small part of the market, that there is no reason for manufacturers to make super-overclocker friendly parts, overhauling all the existing standards, for parts that would cost way more to produce en-masse.

We're lucky enough to have enthusiast-geared companies like OCZ and DFI in the first place :)

[Jack]

We need some sort automatic divider, so we can set htt or fsb and mem clocks independently and the mobo will give you the required divider :D

[charlie]

Maybe DFI could make a board with thses features:
They already have boards with MEMTEST built into the BIOS.... have the board automatically run a memtest to establish max mem speed capability, then you type in a mHz goal, say 3000mHz, and the board adjusts everything to run at memory peak :D

C

[wampir]

Or maybe mobo with voice interface with command "overclock" or just go "regular" :D

[trans am]

Or maybe mobo with voice interface with command "overclock" or just go "regular" :D

HEHE rofl I literally laughed out loud when I read that. hehehe

[DudeMiester]

Arn't the dividers handled inside the CPU for the CPU and RAM clocks? Seems to me like you should be bugging AMD about this. Then again, I don't know if a floating point devider will go over nicely with the hardware, it seems like it would take a very complicated chip design to do this.

[blinky]

i had this idea a couple months ago, lol

if it could work it would reeeeally be helpful

what they need is something that can emulate any combination of PLLs

in bios you could enter the fsb you want, and seperately the mem speed you want, so you could run your mem at max speed whatever fsb/htt :)

[Rabbi_NZ]

this is why I think everyone should have a set of good tccd and a good set of 2-2-2 ram. (ex. vx)
just gives you more sweet spot options.

but I agree with prince charles, infinite options would rule! has anyone heard about the future clockless asynchronous cpu? what does this mean for overclockers?

http://www.arm.com/news/6936.html
If that's what I think it is it CANT be overclocked... well, it can be supercooled I guess so the speed will be dependant entirely on how fast current can flow through the core... very similar to Artificial Neural Network design I think

[Aphex_Tom_9]

clockless processor...hmm, it would have to be rated somehow, maybe by efficiency...
it would still be composed of semiconductors, so their individual switching rate would determine the overall "speed" and the apparent "speed" would vary with application...

[_Eduard_]

I use a calculator when in the bios to calculate the best divider ratio's etc. Works quite well actually :D

[blinky]

I use a calculator when in the bios to calculate the best divider ratio's etc. Works quite well actually :Di do all the calculations in my head

... *1337*

[Aphex_Tom_9]

I use a calculator when in the bios to calculate the best divider ratio's etc. Works quite well actually :D
same, i keep a calc near my comp at all times lol.

[[XC] Oj101]

Sorry for the thread resurrection, but well done Charlie! Did anyone ever give him credit when the nVidia chipsets started allowing unlinked RAM?

[saaya]

why cant mobo makers do it? cause its inside the cpu...
why dont cpu makers do it? cause it costs extra transistors, makes debugging and evaluation a h3ll (think of how great all the dividers on nvidia chipsets turned out... NOT :D)
and it doesnt really result in a perf boost that would justify messing with it...

[knopflerbruce]

An option that would give you the possibility to set the "x" directly in the cpu/x=memspeed equation, instead of by ratios should be possible I think:)

[saaya]

An option that would give you the possibility to set the "x" directly in the cpu/x=memspeed equation, instead of by ratios should be possible I think:)this isnt software we are talking about here, its hardware :P

[Raja@ASUS]

You'd need to add extra logic sections for each and every divider option (or have some kind of buffer). Some of this would add additional system latency so the HW vendors (Intel/AMD) choose what they deem to be a set of divider options that co-incide with available DRAM (or the capabilites of any other bus). This has to be done at the chipset level in HW first, if the additonal logic gates are'nt there, there's nothing the board vendors can do. An ideal platform for overclocking enthusiasts would be one using the fewest possible divider combinations thus allowing for low system latency. 1:1 bus operation is generally the fastest electrical transfer path. In async operation, there has to be an hold time for data to be transfered over from one domain to the other. The wider the gap between two busses, the lower the efficiency (more non transferable clock cycles and wasted power).

[massman]

couldn't you just start messing with the PLL?

[Raja@ASUS]

What point is there screwing around with the PLL if there's no downstream logic driven by the selected operating frequency?

This is kinda like setting a 27X core multiplier on a 920 CPU - it does not exist.

[saaya]

well, bad example, cause 920s have all the hardware in place to run a 27x multi, its just locked :D
but yeah, like raja said, the more fine grained multipliers you want, the more logic you need to add.
afaik for multiplier steps from 1-4 you only need some very basic logic, going higher you need more logic, and the next step at some point is adding half multipliers and .25 multipliers like amd did etc...
if you adjust the pll you dont adjust the ratio of clocks of different logic to each other. you either make everything run faster or slower.