ASUS ROG Rampage Formula: Why we were wrong about the Intel X48
by Kris Boughton on January 25, 2008 4:30 AM EST- Posted in
- Motherboards
The Rules of Working with tRD: What's Allowed and What Isn't
We mentioned earlier that there are a few rules pertaining to the modification of the default tRD value for a particular memory subsystem configuration. These rules are actually more of a set of equations that we have derived in order to assist the user in pre-determining whether or not a system will POST and operate given the settings in question. They can also be used to explain after the fact why certain configurations refuse to function.
In actuality, there is only one requirement that must be satisfied when setting tRD: The MCH must have a minimum amount of time to complete the clock crossing procedure for each data bit translated from one clocking domain to the other. We have shown the equation below for those that care to know. Unfortunately it does not lend itself well to being solved as the input variables ("t0" and "tckxss") are not readily obtainable.
"t0" is the total time it takes data being returned to the CPU as read in memory to cross from the memory bus domain to the system data bus domain. This is dependent on a number of variables, including memory burst length (4 or 8 cycles), Command Rate (1N or 2N), the memory divider in use, CAS (tCL) timing, and the current FSB. "tckxss" is a little more esoteric in nature and is outside the scope of this discussion; the value is generally very small when compared to "t0". The difference in these times, rounded to the lowest integer plus one, bounds the lowest potential tRD setting necessary for data to be properly transferred from one bus to the other.
Because the first expression is too difficult to warrant everyday use, we spent many hours populating a large test matrix table created for recording the POST (Power On Self Test) results of every strap/divider/FSB/CAS setting combination that our test bench was physically capable of supporting. Using this data, we then developed the method and equation you see above, which can determine whether or not a desired memory subsystem configuration will work. It is possible, and rather probable, that there is another step discontinuity in the logic for FSB speeds in excess of the high value in our test range (466MHz). We will leave the discovery of some such value up to others - unless Intel is kind enough to send us additional 45nm dual-core processors, in which case further testing on our part might be justified.
Entering arguments for the use of the "POST Test Equation" are as follows: tRD, in clocks; tCL (CAS), in clocks; FSB, in megahertz (MHz); N, the memory divider in use, expressed in fractional form (i.e. 3:2 would be 3/2); and "x", which should be chosen from the options provided, depending on the FSB in use. Evaluate the left and right side of the equation separately. The expression is satisfied if the left (actual margin) is greater than but not equal to the right (the minimum required margin).
There are no units purposefully associated with these numbers, as this equation is intended as nothing more than a test to determine whether a system will POST using the desired parameters. If the expression is false, the configuration/system will fail to boot; if it is true then the configuration is allowed and the POST event should at least occur. Keep in mind that this equation provides absolutely no assurance that the system will be stable at the settings provided - just because you want to run your memory at DDR2-1200 CAS 3 and the equation says this is possible, does not mean that your wish will be fulfilled. Let's go through a quick exercise of what we have learned regarding the proper use of the "POST Test Equation" with a few practical examples.
If you can follow these examples then you are ready to move on to the next step - determining optimal system performance points and then validating your results. There are many choices when it comes to deciding how to configure a system for the best possible experience. Some choices are clearly better while other decisions may come to down to personal preference. For instance, some users may be willing to subject their expensive hardware components to higher voltages, creating an environment of accelerated wear and earlier failure. Others may be far less concerned with the consequences of their choices; in either case the trade-offs will be clear. We will now take what we have learned and provide our rationale for why we would feel one overclocking approach to be superior to another. After all, overclocking should always be based on an intelligent decision making process and not the clumsy application of brute-force.
Facebook
Twitter
RSS
73 Comments
View All Comments
poohbear - Friday, January 25, 2008 - link
one thing i liked about some of the recent high end mobo releases was the inclusion of an onboard wi-fi chip on a desktop mobo, but this mobo seems to be lacking that. i mean, they threw in everything but the kitchen sink, why not include wi-fi?:(TheDoc9 - Friday, January 25, 2008 - link
One of the best I've read here, definitely one of the best on over clocking I've ever read. It takes it to the next level, reminded me of how a body builder friend of mine schedules and calculates his workouts, calories, and entire life to be the best he can be. Hope to see more like this one in the future.jimru22 - Friday, January 25, 2008 - link
The article references the use of an Intel Extreme processor with adjustable multiplier. I'm planning on building a system hopefully anchored by the Asus Rampage Formula and a Intel Q9450 with locked 8X multiplier. Based in the charts, it seems to me that in order to run the Q9450 (333 MHZ) at 3.6 MHZ a 450 MHZ FSB is required. Therefore in this case, a tRD of 6 / Trd 13.3ns is the optimum value. Is this correct?kjboughton - Friday, January 25, 2008 - link
You would be correct. Processors with lower maximum multipliers present somewhat of a challenge when selecting the best memory configuration. In this case the 8x multiplier forces a higher than normally desired FSB, which is one of the many benefits of owning an Extreme processor (no such limitation). As such, the next best option, and the first choice for you, would be to go to 450MHz FSB and set a tRD of 6. Although this might not be completely ideal (we like to stick with 400MHz) your results will without a doubt be within a few percent of real-world performance at 400MHz FSB and a tRD of 5. Yet another reason why the Extreme line of processors are worth their price.Odeen - Saturday, January 26, 2008 - link
I'd like to differ on that.. As someone who first discovered overclocking during the Celeron 300A days, where a budget chip could run at 50-60% faster than its stock speed, and deliver higher performance than a $400 (at release time) Pentium III 450MHz, all without overstressing the rest of the platform (i.e. with bog-standard FSB and memory speed) I view overclocking as two ratios:Maximum attainable clock speed / original clock speed. 3:2 is the minimum ratio that isn't depressing to see booting up.
Cost of equivalent performance from a processor w/o overclocking / cost of actual processor. In the case the ratio was 4:1. Some of the best-case scenarios (like the very last 300A's being 100% overclockable to 600mhz), the ratio can be 6-7:1.
The Black Edition CPU's fail both value tests tests, because they are typically ONLY available at the fastest speed grades. Therefore, they are unlikely to reach a 30% overlock, never mind the requisite 50. And, being the most expensive SKU in the class combined with the lackluster overclock potential means that they are unlikely to outperform a processor that costs 4x as much (even an imaginary SKU that fits on the price-performance regression line of the class).
That said, if the Wolfdale E8190 is $130 and Intel somehow offers an "enthusiast edition" of it for $180 (that is, an edition for true enthusiasts, who want to extract the maximum bang for their buck), I would get one - the unlocked multiplier would make overclocking less of a "platform" issue (i.e. "how fast will the chip go until my motherboard peters out") and more of "how fast will this particular chip go period". I can definitely get behind that.
jimru22 - Friday, January 25, 2008 - link
Thank you Kris for the outstanding article as well as your response.Kind regards,
Jim
Orthogonal - Friday, January 25, 2008 - link
What are the chances someone could whip up an Excel Macro to incorporate all these inputs, equations and graphs for easy computation of optimal settings for a given CPU and Memory configuration.kjboughton - Friday, January 25, 2008 - link
Already exists, although you'll have to sweet-talk me into releasing the file. Seriously though, the Excel spreadsheet makes choosing the right settings downright simple.Orthogonal - Friday, January 25, 2008 - link
Fair enough, pretty please!Well maybe there could atleast be a web applet on the site or something of the sort. That would be killer.
LoneWolf15 - Friday, January 25, 2008 - link
Just one thought...IMO, no "Board Layout" portion of a review is complete without a picture of the port cluster on the back of the board.