So what exactly do you need a 6-1/2 digit multimeter for again?

[dilbert]

For marketing purposes of course...

if you want to outsell competitors 5.1 model you need a 6.1 model.

If the competitor comes up with a 3D  multimeter you need a 4D multimeter to outsell it...

Your signature reminds me of an anecdote.
We were trying to run basic flow code with matlab.To do so,we simulated a curved wall,defined by a nicely-given function.
Check with the HP calcs,ok.Run the code...something strange on the outer wall,right in the final part where troubles can arise..but that problem shouldn't have come out.
Hand check the functions,the discretizations,calc and draw some flow lines by hand,everything ok.
It looked like that part of wall was and wasn't there.Open the variable table and check them all,everythings' fine.
Then i had a lightning.
The wall function was discretized in points,and matlab was reporting them as discretized values.
But the variable editor just showed the first 4 or 5 digits,not the exact value.
So 99.99999999 was displayed as 100.0000,which is nice for a cake recipe,but not for the boundary conditions checks who red "fluid" instead of "wall" on those points because 99.99999999<100.0000
So there you go,a true case when a 1E-8 can waste you an afternoon.I still have the hand drawings somewhere.

[HLA-27b]

99.99999999<100.0000

True

 BUT
 99.99999999...=100

Those three dots make all  the difference.

[poorchava]

It makes sense when building measurement equipment.

I had a need for such meter when building device for measuring C-V characteristics of semiconductor junctions and MOS structures (my master's thesis). I wanted to measure resistances of PCB tracks and such to nullify as many arrors as possible in software. Keithley 2001 was a really nice thing to have.

Shame I can't afford to buy it for private use

[free_electron]

The wall function was discretized in points,and matlab was reporting them as discretized values.
But the variable editor just showed the first 4 or 5 digits,not the exact value.
So 99.99999999 was displayed as 100.0000,which is nice for a cake recipe,but not for the boundary conditions checks who red "fluid" instead of "wall" on those points because 99.99999999<100.0000
So there you go,a true case when a 1E-8 can waste you an afternoon.I still have the hand drawings somewhere.

Which is why real precision mathematics needs to be done using packed bcd arithmetic. There precision is absolute.

Packed bcd is a format where you use 1 byte to store two digits. .  1 byte is two nibbles
If you draw the hex table then 0 to 9 represent the numbers. A represents the sign , b represents the comma , c represents infinity , d represents the sign of the exponent, e represents where the exponent begins , f represents 'j' so you can have imaginary numbers as well.
In case no imaginary numbers are needed you can use it as the fraction sign ( f = fraction , e = exponent ) so you can store things like 1/3

For example
-1.7e-10 is stored as a string
0xa1b7ed10

12e13 would be 0x012e13

And so on

The problem is speed hit. You need a decoder and encoder. Cpu's in general can do bcd arithmetic.
As long as you have storage space you can store numbers.
If you have a decimal number with 500 digits you need a string of 257 bytes to store it. But it will be correct without rounding error.
The ieee765 floating point system is the worst thing ever invented. Because it gives you errors.

Packed bcd libraries exist for 'big iron' like Cray and others.

[nukie]

When you have 6-1/2 digit multimeter you get an additional feature its called warm up time. If you don't need it then stick to lower digits meter.

tapatalk

[FenderBender]

Very insightful guys. Starting to see the point...but no way in hell I'm getting one, unless I get it for free. 

[rr100]

Which is why real precision mathematics needs to be done using packed bcd arithmetic. There precision is absolute.

Problem is it really works only with arithmetic, that is the 4 basic operations and at most up to fractions. Once you throw in radicals, logarithms, trigonometry (and even limits/series/integrals) you have the problem that you can't compare the results as the same number can be represented by different strings (and there is no guaranteed "canonical form" or way to get from one representation to another).

[anotherlin]

Which is why real precision mathematics needs to be done using packed bcd arithmetic. There precision is absolute.

If you want absolute precision, use "symbolic computation".
There is no way you can perfectly represent 1/3 (0.333) using 2 (binary) or 10 (decimal) base system.
You have also the problem of constants like pi and irrational numbers (like square roots).

Double precision floating points have 57-bit precision. Proper floating point usage can be tricky (and it is often overlooked in computer science courses).
But I don't see any point of using BCD arithmetic. Especially if you consider the fact there's quite a few libraries for arbitrary long integers (BigNum). GNU even offers an arbitrary precision floating-point library.

For our DMMs, if you have 6.5 digits, that can easily fit into 24-bits.

[analex]

2 words : DYNAMIC RANGE.
That's why you need so many digits.

Yes, more SNR could carry more information. 6.5 DMM gives more DR abilities to get more SNR and more informations.

As 6.5 DMM has 0.1m ohms resolution, it is easier to locate the short circuit point by using 2-Wire or 4-Wire(better way) OHM.

Some of 6.5 DMM have data plot function, and DMM have variable BW(change NPLC), and uV/nA/mohm level resolution,  is very easy to find out low freq characters of a "noisy" signal. BTW, DSO can't do that, as DR/resolution reason; SA can't do that, as it can't go down to low freq range.

6.5 DMM is faster than 5.5 etc. Most of time, I set my 6.5 DMM to DCV 1000V range, so I can measure form 1mV~1000V without waiting for auto range. As this I can find out over load immediately at the power up moment, and cut off power supply before smokes.

[saturation]

That would be nice.  One could be lucky today and get a used one in $1000-3000 in perfect order.  Its the calibration costs that are high, if you want the best it can deliver.

The 1989 designed 3458a 8.5DMM sells for about $10,000, about its new price in 1989, in could be called cheaper given the depreciation of the US $. 

However, with 1nV resolution and 4ppm stability, few labs can calibrate the 3458a properly.  It runs about $600 for the 4ppm calibration, but the 0.2ppm version is about $1500.

http://lvldstdslabagilent.blogspot.com/2005/01/faq-what-3458a-calibrations-are.html

http://www.febo.com/pipermail/volt-nuts/2011-August/000840.html

Only a Josephson Junction has the resolution to calibrate an 8.5 DMM these days, and even if it were made into a turn-key standard, the labor to maintain it is far more than buying a 732a and keeping it in an airconditioned room, so calibration costs won't drop soon.

Eventually we will all have to use 8-1/2d meters

[nukie]

I used mine as a coffee warmer today, it works quite well.

[ejeffrey]

The wall function was discretized in points,and matlab was reporting them as discretized values.
But the variable editor just showed the first 4 or 5 digits,not the exact value.
So 99.99999999 was displayed as 100.0000,which is nice for a cake recipe,but not for the boundary conditions checks who red "fluid" instead of "wall" on those points because 99.99999999<100.0000
So there you go,a true case when a 1E-8 can waste you an afternoon.I still have the hand drawings somewhere.

Which is why real precision mathematics needs to be done using packed bcd arithmetic. There precision is absolute.

No, the only reason to use BCD is if you need to represent exact decimal numbers, typically with currency. This works OK because there is a finite required precision, and you only have to be correct on addition and subtraction -- multiplications are allowed to be rounded before adding to the accumulator.  Physical quantities have no special preference for base 10, so BCD is rarely if every used for scientific or engineering numerical computing (it is used in CAD for geometry definition, but converted to floating point for things like FEA).  Infinite precision arithmetic, either binary or decimal, is only used in specialized circumstances, for any sort of general algorithm the required size grows exponentially with the number of steps in the computation, and in any case operations other than +,-,* are not even possible.

The real lesson here is 'don't compare floating point values for equality'.