Fluke 8558A/8588A 8.5-digit DMMs

[splin]

But how are we to interpret the 0.1ppm difference between the highlighted values? They make no sense to me; it would be surprising if the factory calibration uncertainty was much better than 2ppm. Clearly Fluke have the capability for sub ppm 10V calibration, but in that case why not specify absolute uncertainties for periods less than 365 days?

If I were paying $17k+ I'd want it to be freshly calibrated, preferably within 48 hours - not from a distributer with potentially only a few weeks left of a 12 month calibration. I guess that would be an expensive option like the 3458A meteorological calibration option.

What is the problem with the relative specs? That is what the meter does and it is what HP also specified for the 3458A. If you want to have fully traceable specs you have to use absolute specs from Fluke or add the 2ppm for the 3458A. Or you RSS the traceability uncertainty with the rel specs like described in the Fluke datasheet. The nice thing on Fluke specs is, they give you absolute values which you can take without using your brain. But they cannot know how you calibrate the meter and therefore you have the rel specs.

Ah there is the answer - thanks. What I'd missed was that Fluke tell you to RSS the relative and absolute uncertainties so .7ppm uncertainty for 10V plus 2.7ppm relative gives 2.8ppm as you state. This is in contrast to the 3458A datasheet which tells you to use the algebraic sum. If Fluke had simply stated their uncertainty in the datasheet, like most other manufacturers do (or used to do), then there would have no uncertainty. 

It still does't exuse them because the 365 day spec is not consistent with the 2 year spec. - subtracting 1.2ppm from the 2 year +/- 5C absolute figure (TC is .3ppm/C) gives 6.8ppm which implies an absolute 10V uncertainty of 4.1ppm RRS'd with 5.4ppm. That can't be right so either I've missed something else or it's another error.

As far as I can see neither Keithley nor Keysight specify which method they use to specify the absolute uncertainty for the DMM7510/34470A, nor their uncertainty for 10V. It makes quite a bit of difference so it would be helpful if all meters were specified consistently. I guess you find out from the cal certificate after you've bought one?

As we ordered our 8508A is came directly from Fluke and was calibrated with 0.7ppm uncertainty for 10V. Which gives 2.8ppm for 10V exactly as the new meter. The new better absolute value is no improvemt of the meter, it is improvement of the calibration.

Is that standard for the 8508A or an extra cost option?

[e61_phil]

First of all: I'm not an metrology expert in terms of uncertainty calculations. Therefore, I might be wrong on some points here.

What I'd missed was that Fluke tell you to RSS the relative and absolute uncertainties so .7ppm uncertainty for 10V plus 2.7ppm relative gives 2.8ppm as you state. This is in contrast to the 3458A datasheet which tells you to use the algebraic sum.

Fluke did it in the extended spec datasheet of the 8508A. Perhaps they add it in the final version of the datasheet? I don't know.

I think one cannot the compare the 3458A datasheet. HP says one have to add 2ppm for traceability, but that is many years old. In older datasheets Fluke also adds uncertainties algebraic. Is that changed due to the GUM in the last years?

If Fluke had simply stated their uncertainty in the datasheet, like most other manufacturers do (or used to do), then there would have no uncertainty. 

Could you explain what do you mean by that? They should remove the absolute specs?

It still does't exuse them because the 365 day spec is not consistent with the 2 year spec. - subtracting 1.2ppm from the 2 year +/- 5C absolute figure (TC is .3ppm/C) gives 6.8ppm which implies an absolute 10V uncertainty of 4.1ppm RRS'd with 5.4ppm. That can't be right so either I've missed something else or it's another error.

I'm also confused by that. I also can't reproduce the exact values from the 8508A datasheet. My results are <0.5ppm lower.
What I can imagine is another (higher) calibration uncertainty if you order a 2year calibration.

These are things one could probably discuss in Kassel with Fluke.

As far as I can see neither Keithley nor Keysight specify which method they use to specify the absolute uncertainty for the DMM7510/34470A, nor their uncertainty for 10V. It makes quite a bit of difference so it would be helpful if all meters were specified consistently. I guess you find out from the cal certificate after you've bought one?

I think in contrast to the 3458A specs, the 34470A specs are absolute specs (directly traceable). But I can't find any uncertainty calculations in the service manual. They only suggest calibration equipment. In the 34401A service manual they talk about at least 1:4 TUR imho.
I can have a look into the calibration sheet from Keysight for the 34470As. If I remember correct they are tested against 24h specs with an 5720A. But I will check that.

Is that standard for the 8508A or an extra cost option?

That was no option. I can imagine that they improved their uncertainties over the years and no one has changed the 8508A datasheet.

[TiN]

If I remember correct they are tested against 24h specs with an 5720A. But I will check that.

I doubt any commercial calibration lab maintain and characterizes 5720A every 24 hours for customer with 34470A. 

[e61_phil]

If I remember correct they are tested against 24h specs with an 5720A. But I will check that.

I doubt any commercial calibration lab maintain and characterizes 5720A every 24 hours for customer with 34470A. 

The 5720 was used to check the 34470A 24h specs.

There shouldn't be any need to characterize the 5720A

[TiN]

Perhaps it's possible to see data points from report to make sure of that? 

[e61_phil]

Perhaps it's possible to see data points from report to make sure of that? 

I will search for that paper tomorrow. But I can't see any problem here. Even the 1 year 10V spec of the 5720A delievers a TUR of nearly 1:2 for 34470A 24h specs.

And it isn't a big deal to bring the 5720A daily in 24h specs with artifact calibration.

[e61_phil]

The certificate of calibration lists a 5720A and a 5725A with their trace numbers and cal dates. But no further uncertainty values. The report shows readings for severeal ranges and the error and how much that error is in percentage of the 1year spec. Biggest error in that list is 1.1% for 10V ACV. As there is no spec which is in 1 year 100x worse than for 24h the meter was in 24h specs. But you don't know that for sure, because there are no uncertainty statements.
It might be is not fair to compare a manufacturing cal report of a 34470A against the report delivered with the 8508A, but if you do, the Fluke one looks much more like metrology

[TiN]

What I expected. To be exact testing against 24 hour specifications would be only valid right after adjusting all functions and ranges on the meter, within same time interval.
It's just bit strange how you say "not a big deal to bring 5720A into 24h due to ACAL", but keep saying that 3458A's ACAL is drawback and does more harm than good. You don't have to use 3458A's ACAL in which case you get same behaviour as 8508A and use external full calibration for all ranges/points every time. 
But I agree, this is bit offtopic in this thread.

[e61_phil]

What I expected. To be exact testing against 24 hour specifications would be only valid right after adjusting all functions and ranges on the meter, within same time interval.
It's just bit strange how you say "not a big deal to bring 5720A into 24h due to ACAL", but keep saying that 3458A's ACAL is drawback and does more harm than good. You don't have to use 3458A's ACAL in which case you get same behaviour as 8508A and use external full calibration for all ranges/points every time. 
But I agree, this is bit offtopic in this thread.

You are right, if one would argue completely straight I should also don't like the artifact cal of the 57xx. My point was just, that you are able to calibrate anything with a daily adjusted 57xx if you like. It isn't a big deal, even for smaller labs.

I doubt that a 3458A behaves like a 8508A if you don't use ACAL. And you are not able to adjust every single range on a 3458A. But, I think we should discuss that in another thread, as you already suggested .
I modified my logging scripts for the continous monitoring of severeal 10V standards against the 3458As. It logs all CAL? values until now. That should show the stability of a 3458A without ACAL in a couple of months. If you are right and a 3458A is as stable as a 8508A the CAL? values shouldn't move.

[splin]

First of all: I'm not an metrology expert in terms of uncertainty calculations. Therefore, I might be wrong on some points here.

I'm certainly not either, but I guess that's pretty obvious!

Fluke did it in the extended spec datasheet of the 8508A. Perhaps they add it in the final version of the datasheet? I don't know.

I think one cannot the compare the 3458A datasheet. HP says one have to add 2ppm for traceability, but that is many years old. In older datasheets Fluke also adds uncertainties algebraic. Is that changed due to the GUM in the last years?

I don't know, but since it gives better marketing numbers using RSS I guess all manufacturers use it for their more recent products.

If Fluke had simply stated their uncertainty in the datasheet, like most other manufacturers do (or used to do), then there would have no uncertainty. 

Could you explain what do you mean by that? They should remove the absolute specs?

The latter part was just a (poor) joke which should have read "... then there would have been no uncertainty". What I meant was that Fluke (and all manufacturers) should include in their datasheets their 10V uncertainty to traceable standards, ie. the 0.7ppm they use for calibrating these meters. That would eliminate the guessing and allow you to use the most appropriate method for calculating the absolute uncertainty for your purposes.

The Tcal +/-1C relative specs are the most important for understanding the meter's stability and I believe they should be provided for all high end meters - eg. you can use them to compare the drift of other instruments rather than the absolute errors (especially if you can control the temperature to +/-1 C). Fluke is very good in this respect and dissapointing that Keithley only provide 24hr relative specs for the DMM7510. Keysight presumably don't consider the 34470A a metrological instrument and point to the 3458A if you are that concerned with accuracy.

Absolute numbers are helpful of course but you can't necessarily calculate the relative figures from them, even if you know the cal standard uncertainty, as they don't tell you how they calculated the absolute numbers. For example, they could use a lower TC when calculating the Tcal +/- 5C absolute specs than the TC for the wider temperature range they specify in the datasheet. They might not do this of course but it could allow them to publish better numbers.

It still does't exuse them because the 365 day spec is not consistent with the 2 year spec. - subtracting 1.2ppm from the 2 year +/- 5C absolute figure (TC is .3ppm/C) gives 6.8ppm which implies an absolute 10V uncertainty of 4.1ppm RRS'd with 5.4ppm. That can't be right so either I've missed something else or it's another error.

I'm also confused by that. I also can't reproduce the exact values from the 8508A datasheet. My results are <0.5ppm lower.

It would be interesting to see how you calculated that.

What I can imagine is another (higher) calibration uncertainty if you order a 2year calibration.

Seems a bit of a stretch - I would expect a meter to be within the 2 year specifications after 2 years of ownership from new.

These are things one could probably discuss in Kassel with Fluke.

Right; it's unreasonable to criticse Fluke for specifications they haven't officially released yet. If I were spending $17k+ on a meter then I would certainly be talking to Fluke to resolve my uncertainties about the uncertainties in their specs!

I am unlikely to buy one but it is interesting (for me) to analyse the numbers anyway to help my understanding of how the specifications are derived and what they really mean.

I think in contrast to the 3458A specs, the 34470A specs are absolute specs (directly traceable). But I can't find any uncertainty calculations in the service manual. They only suggest calibration equipment. In the 34401A service manual they talk about at least 1:4 TUR imho.
I can have a look into the calibration sheet from Keysight for the 34470As. If I remember correct they are tested against 24h specs with an 5720A. But I will check that.

Is that standard for the 8508A or an extra cost option?

That was no option. I can imagine that they improved their uncertainties over the years and no one has changed the 8508A datasheet.

Ok, thanks. A bit surprising as marketing are generally very keen that any potential purchasers know about any improvements thay make.

[TiN]

A bit surprising as marketing are generally very keen that any potential purchasers know about any improvements thay make.

I'm sure they will spend few pages on benefits of color TFT screen and power sensor support. As for DC/LF metrology and performance improvements over 8508A (if any exist) - they will keep it all shut to avoid giving hints for competitors. After all this is very niche market and you can count remaining players by fingers on one hand.

[e61_phil]

The new meter looked really nice and the linearity is improved over the 8508A. (short check with a 10V and 1V source whose ratio I later compared on a 3458A)

They also said, that the transfer specs are valid for the whole range not only within 10%

[TiN]

linearity is improved over the 8508A.

Sorry, I find it very hard to believe just on words. I'd like to see linearity specified or measured, just like it was on 3458A.
Call me Fluke hater or whatnot, but from such expensive meter with "reference" marketing the bar must be set accordingly.
Fluke has ability to do such measurements, they don't need to go NIST today, like HP had back in 1989.

Again, I don't say linearity itself is bad on 8508A/8588A, but completely omitting this important specification is bad.
So saying that unspecified parameter on new meter is improved is no different to saying that unicorns have rainbow hair. 

[e61_phil]

@Tin: Linearity was briefly tested on a 8588A and the measured 1V to 10V ratio was within 0.05ppm of the ratio measured by my 3458A. They said there are a lot of mistakes in the leaked documents. Therefore, let's wait until the final version is published and start the discussion again.

@splin: I only saw updated specs for the HP 3456A after some time in the market. I think the 3458A is also much better than specified and there is much experience with the unit. However, nobody has updated the spec sheet.
I calculated it simply as shown in the 8508A datasheet (there is an example for the temperature range).

[TheSteve]

How can it be a "leaked document" if it was on the Fluke website for all to see?

[e61_phil]

How can it be a "leaked document" if it was on the Fluke website for all to see?

I can only repeat what they said today: "wasn't finished, should have been set to private and so on". I don't know why they published that and if the story is "true".
I'm looking forward for a final version. There were also some different numbers in his presentation.

I have no contract with Fluke

[Magnificent Bastard]

{...snip...}
@splin: I only saw updated specs for the HP 3456A after some time in the market. I think the 3458A is also much better than specified and there is much experience with the unit. However, nobody has updated the spec sheet.
{...snip...}

And THAT will never happen.  What they COULD do, is release a 3458B (same as the 'A' model with some firmware mods to identify as a 'B')-- and then release an updated spec sheet for THAT.

The reason they could release updated specs for the 3456A, is that the new specs applied to all known 3456A's in the wild--- which is definitely NOT the case with the 3458A.

[TiN]

[Krampmeier]

The noise level / speed of the 8588A was quite impressive. The Fluke rep claimed that the accuracy at 1 second aperture would the same as with the 8508 at 26 seconds. Indeed, it did look quite stable at 1 s aperture, but default was 10 s for 8.5 digits.

The 8505 had ranges like "20 V". The ranges of the new meters are now slightly wider, but they are now called differently, e.g. "10 V", even though the full scale value is acutally more than double of that. When the measurement uncertainty is calculated, the _nominal_ range must be used, so the "ppm of range" numbers in the specifications must now be divided by 2.02 when comparing with the 8508 specs. Quarks, did you take this into account in your graphs?

At 5 MS/s, the digitizer should still provide 16 effective bits.

We got to play with the UI a bit and it felt a bit outdated to me. You really have to learn which options hide behind which button and soft key. The graphing feature does not feel very well thought out, for example you cannot easily delete the buffer, or change between "graph only" and "graph + statistics". People who have used the new Keithley meters will be disappointed by the UI implementation.

However, we saw an engineering sample and everything written above may change.

The official introduction of the new meters will be on March 26.

We got some information about the price point of the meters already:
8588: 15 - 16 k€
8558: 10 k€

They did not comment on the strange limitation in the safety notes of the leaked specs, which says that voltage sources must not be able to provide more than 200 mA. Probably an error or requested by lawyers?

[dr.diesel]

It's gonna be tough to out the legacy of the mighty 3458A, but glad to see another option!

Keysight is going to be forced at some point in the near future to revamp the 3458A, if they still have the engineering resources.

We need a good reliable evaluation of Fluke's new contender, maybe we need to group buy one for TiN.

[ScoobyDoo]

It's gonna be tough to out the legacy of the mighty 3458A, but glad to see another option!
Keysight is going to be forced at some point in the near future to revamp the 3458A, if they still have the engineering resources.

Keysight Germany has commissioned the design of the successor of the 3458A to a German engineering house - three years ago -
their new unit will be launched soon - in term of weeks but no longer ... several sources inside Keysight told me such.
Let's cross fingers that their new unit lines up with the mighty 8558A/8588A duo ...

Herzliche Grüße/Meilleures salutations/Best regards

ScoobyDoo

[e61_phil]

The reason they could release updated specs for the 3456A, is that the new specs applied to all known 3456A's in the wild--- which is definitely NOT the case with the 3458A.

Could you explain your statement in a bit more details? Do you think not all 3458A are much better than specified?

[e61_phil]

You don't have to use 3458A's ACAL in which case you get same behaviour as 8508A and use external full calibration for all ranges/points every time. 
But I agree, this is bit offtopic in this thread.

I found a few full lists of CAL constants for our 3458As. Not much data until now, but one can see a 3458A doesn't behave like a 8508A if you don't use ACAL. I will go on with collecting data..

[HighVoltage]

I found a few full lists of CAL constants for our 3458As. Not much data until now, but one can see a 3458A doesn't behave like a 8508A if you don't use ACAL. I will go on with collecting data..

What behavior of the 8508A do you mean?
I am using ACAL on my 3458A once every day and it has shown to be working perfectly, based on resistance measurements.

[e61_phil]

What behavior of the 8508A do you mean?
I am using ACAL on my 3458A once every day and it has shown to be working perfectly, based on resistance measurements.

I meant the stability of the 8508A. The 8508A is stable without daily ACALs. The fact that the CAL? constants move quite a bit and the reading is stable on a 3458A (if you use ACAL) means that the internals of the 3458A drift by the same amount as the "constants" drift.
If you don't use ACAL anymore, the 3458A will drift like the shown constants (which are constant without running ACAL). My point was only that a 3458A isn't nearly as stable as a 8508A. The 3458A (at least ours) really need ACAL to compensate for drifting components.