CPU-Z v.1.38 has incorrect ram speed calculations!!!

[Swampdonkey]

Today's tip:

Do NOT the trust CPU-Z versions 1.35 and 1.38.

The ram speed calculations could be wrong.

Don't believe me? Here's the example that helped me discover this:

I have an MSI 7207 board (BIOS 3.0) paired up with a AMD San Diego 4000+ with two 1gig stick of Kingston HyperX pc3200 ram.

This processor has a 12x multiplier. I can turn up the Fsb bus speed to achieve my overclock. Of course, I cannot use a 1:1 FSB to Ram divider because my ram is not high end ram that can achieve higher speeds such as 230mhz and up, so I'll be using a 183 divider.

 

So let's do some math shall we...

We divide 183 into 200,

then multiply the cpu multi to determine our FSB to RAM divider:

200/183=1.092896 x 12 = 13.11475. This number ALWAYS has to be rounded up, so in this case it's 14. This is where CPU-Z version 1.35 and 1.38 is messing up. They are saying it should be rounded down.

 

For me, my target CPU speed is 2760mhz(12x230).

This was determine from earlier testing of the cpu using Prime95

So we need to divide 14 into 2760mhz to get the ram speed, right?

 

2760/14=197.14mhz ram speed

 

This is what CPU-Z is showing:

 

13 divider is incorrect. 212.4mhz ram speed is incorrect.

This is definitely leading builders astray by making them think they're maxing out their stuff when in reality, they're running their stuff at stock speed and in some cases below stock speed.

CPU-Z version 1.31 doesn't seem to have this miscalculation problem but one novel idea is to do the calculations yourself instead of relying on a program to do it for you.

 

Special thanks to renethx who has an excellent guide/thread on anandtech about this mobo and AG who has told us we should never rely on programs to do our overclocking for us. Now we know we can't even trust them to do our calculations for us either. lol

[Swampdonkey]

Just checked version 1.37. It has incorrect ram speed calculations too!

[jaz]

It gives me incorrect cpu voltages...very buggy. Not like. A64info much better.

[Guest r3d c0m3t]

I seem to have the exact same problem, ver. 1.38 tells me my memory is running at 269MHz, when in fact it's running at 245MHz.

 

200/216= 0.925925925 x 11= 10.18518519 (My CPU speed is at 2695MHz (245x11))

 

2695/11= 245MHz

 

Mind you, my RAM is supposed to running at speeds of 500MHz, thanks to the miscalculation of CPU-Z I'm running below spec - this is one of the reasons why I rarely use CPU-Z for anything other than to tell me my CPU speed.

[My Ugly PC]

CPU-Z has always that I can recall had issues reporting incorrect values for certain dividers. Try using no divider, or a 'simple' one like 100mhz or 150mhz, and you'll see that it can be perfectly accurate. To gauge whether CPU-Z deserves a public calling out, compare the reading with dividers that don't work, to another program's readings, say everest.

 

It gives me incorrect cpu voltages...very buggy. Not like. A64info much better.

 

No software, and I mean absolutely none can be trusted for voltage readings. I've not even met a BIOS reading that I would really call accurate. It's either a calibrated multimeter or someone else's best educated guess. Those are the only options.

 

In my humble opinion CPU-Z is a fine program, that at the very least is a way to show an absolute idiot which BIOS he has, or what mem he has, or what the main timings should be at teh rated frequency. Of those features and many more I've never seen an inaccuracy, and what fixable issues there are (that every single piece of software will have, let alone a free one) are pretty rapidly fixed with a new version release. The fact the the divider issue has not been fixed leads me to believe the issue is all about OC'er mobos offering us many tweener dividers that save for the enthusiast market, would never exist, and that somehow . up the process.

 

Ugly

 

EDIT: My simple point is that it won't(that I've seen) tell someone a falsehood about stock speeds ie no divider. So the only ones affected are those people(OC'ers) that should know exactly what Mhz every component is at before ever even booting windows. We're all aware enough to know, so some dude isn't going to think that his ram is maxing out at 213Mhz or whatever when at the last lower divider it was at 240Mhz. I wonder if this issue is acknowledged in any CPU-Z release notes.. maybe I check after I shovel some snow, its a freaking blizzard in Chicago.

[My Ugly PC]

Allright, damn snow..

 

Anyway, I must admit that I didn't even look at any of the calculations in this thread, I was speaking to the many times CPU-Z has really just been odd about certain dividers in my experience. But like I said now that I look at the formulas being used, they are incorrect. First and fore most, RAM Mhz goes off of the FSB or HTT or whatever you feel like calling it(heretofore reffered to as FSB), the only relation to the CPU clock is that they both multiply or divide off of the same FSB and that in SDRAM the cycles are synchronous with the CPU. CPU clock speed will never be part of the equation for exacting actual ram frequency from a divider enhanced setting. It goes like this:

 

We will use the OP's settings as an example, so we will find out the real ram frequency for 230FSB with the 183Mhz or 183/200 divider.

 

1. divider the BIOS divider by the default FSB -

We will use an example of 183/200 for the 183 divider so 183/200 = 0.915. This number will always be less than 1, because we are using a divider not a multiplier. If you end up with 1.xxxx you inverted 183/200 for 200/183.

 

2. multiply the decimal equvilant of the divider (explained above, see step #1) by the actual FSB

So we came up with a decimal equivilant of 0.915 for the fractional divider of 183Mhz divider (183/200). Now we multiply that by the current FSB for the real RAM frequency; 0.915(the 183Mhz divider in decimal form) x 230(example FSB). This equals 210.45Mhz actual ram speed, given that the FSB can be an inexact number (example 230.6Mhz) we may be off by a few Mhz, but if our app tells us within IDK 2-4Mhz I'd say that is close enough, because mind you we can't fine tune the ram Mhz off of a divider, only get it as high as possible without bringing down the CPU Mhz.

 

Full equation: 183/200x230 = 210.45 or BIOS listed divider / default FSB x actual FSB

 

Note: A different BIOS may list a divider as 330/400 or 333Mhz otherwise in that format, in this case just break the idiot number in half to get more or less the 166Mhz or 166/200 divider.

 

Now you know where you are, I wish I could tell you if CPU-Z is as inaccurate as I remember, but as it stands you can now tell yourself.

 

Ugly

 

P.S. Please someone tell me if my equation is off, It'd be damn funny if I messed it up trying to correct someone else. A scenario not at all above my spastic nature, lol.

[CPDMF]

CPUZ uses 9/10 as the divisor which of course is not correct. I hate software for reporting anything. It just isn't accurate or repeatable. I use math for mem speeds, a multimeter for voltages and an infrared temp gun for temps.

[Swampdonkey]

Ugly,

I know CPU-Z is free and I applaud cpuid.com for that, but I really think that they are aware of this problem. Surely after putting out three versions (1.35,1.37 and 1.38) this has been noticed by scores of people beside me.

I was using this program to check my cpu frequency and ram speed.

Why was I using it?

1. Simple (I'm not a braniac)

2. Accurate (Something everybody should appreciate)

3. Free (I am however a cheapskate)

Now that I know that it's accurancy is questionable, I will probably start looking for another product. If I continue to use it, it will be with a grain of salt.

 

I just wanted everybody on the street to not trust CPU-Z as gospel as far as your ram speed is concerned.

 

"Misinformation -- Sometimes we think we know something, but we don't really. If our "knowledge" unintentionally is mistaken, incorrect, untrue or misleading, then we are misinformed. The facts have been corrupted by unintended errors somewhere along communication lines. From the standpoint of actions affecting our life support system, misinformation is more dangerous than ignorance. Why? --Because we are more likely to act on wrong information, satisfied in the belief that we are knowledgeable, than if we realize that we are ignorant and need some facts. True misinformation usually cannot persist long, before someone forces correction, and accurate information takes its place."

[Swampdonkey]

version 1.39 just released. It's showing incorrect ram speeds also.