Rigol's "long memory"

[cyberfish]

I have decided to buy a DSO, and came down to Instek 1062A (seems like a very good deal), and Rigol 1052E (just because it can be hacked to 100mhz).

I'm leaning towards the Instek, because of the memory depth.

Then I saw on the Rigol DS1052E page that it has 1 Mpts "Long memory". What does that mean? Is it just marketing crap?

Also, are there other notable differences (except 60mhz/100mhz, and memory depth) between the 2 scopes?

Google doesn't seem to know anything about it, and the Rigol site requires registration to download the manual (what? really?).

Thanks

[alm]

I don't have the Rigol scope, so I'm not the best person to answer this, but I believe the Rigol has a special 'Long memory' mode to be able to use the 1Mpts, otherwise, you can only use 10kpoints or so. The long memory mode is not available at the fastest sweep speeds, not sure if there are any other draw backs (responsiveness?). If the Instek offers the 2Mpts at all sweep speeds by default, that would actually be a pro compared to Rigol.

I don't think anyone here has both scopes (and they didn't want Dave to review one, which is a con, although I don't think Rigol send him one either). The consensus based on specs and support seemed to be that Instek was somewhat better. The Rigol has a few weird features, eg. the measurements are based on the QVGA display, not on the raw data, it would be nice if Instek improved on that, but I don't know if that's the case.

[mikeselectricstuff]

Can't comment about the Rigol specifically, but in general long/deep memory is an extremely useful frature in a scope.
The most common way it helps is that it is much less necessary to get the triggering set just right, or even use triggering at all - you just grab a bunch of data and zoom in on the area of interest.
When debugging any sort of comms protocol, deep memory is pretty much essential to give a decent amount of traffic with sufficient resolution.

[cyberfish]

Thanks! I think I will go with the Instek then.

[DJPhil]

Thanks! I think I will go with the Instek then.

If you do, please keep us posted! Any info is good info, as there's not much out there for the Instek yet.
I was about halfway through saving up when I had to go to the ER on no insurance, so I'll be waiting til next year.

[cyberfish]

For sure. It won't be for another 2-3 weeks, though (applying for a USD credit card so I won't get charged insane exchange rate from my Canadian dollar card). No one in Canada (or anywhere for that matter) sells it at anywhere near tequipment's price.

I won't mind doing any tests that don't cost money/warranty/too much time.

[EEVblog]

User manual available here:
http://www.tequipment.net/RigolDS1052E.html

The 1Mpoint memory on the Rigol is only available in single channel mode at 500MS/s or below.
Dual channel it gets halved, and at 1GS/s it's only 16KB (which is a still fairly decent).

Dave.

[cyberfish]

Ah! That makes sense.

I guess they have 16KB SRAM and 1MB DRAM, and they can't manage to clock the DRAM fast enough for 1GB/s write.

If the Instek one can do 1GS dual channel for 2Mpts total, they must be using 2MB SRAM?

To do 2GB/s in DRAM would be very difficult (I only get 4GB/s on my desktop PC and that's with 4 sticks of memory with 8 chips each).

[mikeselectricstuff]

User manual available here:
http://www.tequipment.net/RigolDS1052E.html

The 1Mpoint memory on the Rigol is only available in single channel mode at 500MS/s or below.
Dual channel it gets halved, and at 1GS/s it's only 16KB (which is a still fairly decent).

Dave.

That's a reasonable compromise. IME you don't often need deep memory and high sample rate at the same time, and it costs a lot to provide deep memory at high speed.
It is however important the the scope's User Interface is snappy enough to be able to scroll & zoom through the memory quickly. Deep memory becomes a lot less useful it it's too sluggish to actually view it easily.

[rf-loop]

I recommend to read GW instek user manual and datasheet very carefully. And also after this can you exactly know how it works? Why?
 (Rigol datasheet and selling brochure is much more open mind and honest imho. There is clear table example about memory depth in different modes). Maybe it is "too difficult" fot GW to make this kind of clear table.

GW tell on the technical data only that:
"Record Length Maximum; 2M points (1 channel), 1M
points (2 channels)"

Compare to Rigol what give extremely clear table about sample rates and mem depth. Why GW want speak very high in every place about maximum 2M points. But then they do not speak loud about 4k points "normal" fast memory. (2x8k or 1x16k points in Rigol)
It seems that GW want littlebit hide truth but then becouse they have double slow memory they speak it so loud that peoples forget check how things are in real. So do not be fooled... If need more slow memory it is ok but if need more fast memory... Rigol have double.




I do not know exactly how it works if I read this from user manual (GDS-1000A series user manual):

"
Note: 2M point memory lengths are only available
for time bases slower than 10ns/div on a single
channel, and 1 M point memory lengths are only
available for time bases slower than 25ns/div on
two channels.

Who knows how it works if read this next?

"The memory depth is limited to 1 M points when
both channels are activated or 2M points when
only a single channel is activated. The signal must
be triggered /stopped to have access to the full
memory depth. Therefore when a signal is saved
the waveform will be automatically stopped if it is
not manually triggered /stopped first.
There are a number of conditions when all of the
available memory is not utilized due to a limited
number of different sample rates. This can be
caused by an un-triggered signal, or a time/div
setting that is too fast to display all the points on
screen."


Maybe this is clear:

"When in the Roll mode, an indicator appears at the
bottom of the display. When in roll mode the
record length is 2M (1 channel) or 1M (2 channel)."



I have not connection to Rigol... yes I have some Rigol also with many others (Tektronix and Hewlett - "Agilent" Packard)
GW is one Chinese name just as Rigol, Siglent (Atten), UNI-T and many many others.
(yes GW is made in Taiwan isle and there is only one China (PRC) what is internationally recognized as an independent state if US want or not.)

[cyberfish]

That is a very good point. I will try to contact GW Instek for clarification. If it's really like you said, though, I doubt they will reply.

In which case I will probably go for the Rigol.
(and by the way I am quite aware of the situation in Taiwan. I was born there and spent 14 years of my life there )

[cyberfish]

I found a review for the 150MHz version.

http://welecw2000a.sourceforge.net/docs/Hardware/GW_Instek_GDS-1152A.pdf

First thing to notice is, the sampling rate is 500MSa/s when using both channels, so the manual is a little deceiving, since I would think it's a really important thing to mention, and they didn't mention it at all.

The ADC configuration is the interesting part.

It uses the same ADC (AD9288), except the 100MSPS version, but only 4 of them (8 ADCs, since they are duals) instead of 5 in Rigol. Which one is better depends on what theory you believe in. If you believe speed grade is just marketing and doesn't matter, Rigol is better. If you believe speed grade matters, Instek is better (25% overclock vs 125% overclock).

The author, who sounds like he knows what he is talking about (he is the developer of an alternative open source firmware for Welec scope), thinks highly of the hardware design, especially the frontend. It would be great if we can compare it to Rigol's. Even a component listing can help. From a very superficial glance, I see quite a few aluminum polymer caps on the Instek, and all electrolytic on the Rigol.

I also read an informative thread on a Chinese (yes, I understand Chinese ) forum about this issue.
http://www.ourdev.cn/bbs/bbs_content.jsp?bbs_sn=3428440&bbs_page_no=1&bbs_id=3043
Most of what they talk about is the same as what we talked about here (whether speed grade matters, if it's ethical, lot's of angry posts, etc).

A few interesting notes -
1. Only 9 chips are sanded - the 5 overclocked ADCs, and 2 AD8370 (variable gain amplifier). The other 2 no one knows. DAC maybe? (for voltage offset, the Instek has it)

2. In a post, search for "?750??", someone tried running a 100MHz rated part similar to the AD9288 (AD9283BRS100) at 100MHz, 120MHz, and 150MHz, measuring a 1MHz active crystal oscillator with a 10:1 resistive divider without compensation. He noted that 100MHz and 120MHz graphs looked "normal", but 150MHz graph has significantly more noise, even on the unconnected second input (green). Unfortunately he didn't write much about the testing methodology.

Someone can conceivably do the same test with the chip in use, if you have time and an FPGA board lying around, and don't mind making a small PCB. The chips themselves don't cost that much.
--------------

My impression is the Instek's build quality is higher, but of course the Rigol has the 100MHz hack. I also have higher confidence in Taiwanese build quality vs China. How much is that worth is up to you.

They are both somewhat dishonest in my opinion - overclocking and sanding chips, and deceiving manual and specs. I guess it's lesser of the 2 evils, since we don't really have any other sensible choice at this price point.

I'm not sure if the 10ns/div limit on the long memory for the Instek really matters. 10ns/div means you can see 1 period / 2 divs on a 100MHz signal. Maybe it just means the long memory doesn't work with equivalent time sampling? The bandwidth is not even that high, and I doubt I will be working with that kind of high speed analog signals anyways. For digital stuff a logic analyzer would be more suitable. That said, I haven't got a reply from them yet. Maybe I should write them in Chinese, too.

By the way, Sparkfun's very interesting new product -
http://www.sparkfun.com/commerce/product_info.php?products_id=9857
$50 logic analyzer with open source software + firmware.

[scrat]

By the way, Sparkfun's very interesting new product -
http://www.sparkfun.com/commerce/product_info.php?products_id=9857
$50 logic analyzer with open source software + firmware.
[/quote]

I'm also in the choice between GDS-1062A and DS1052E. From what has been written above, there are no heavy reasons to buy an Instek, since some of the features they claim are at least not much clear.
However, anyone can see that two channels can become too few for some things, especially if you deal with mixed-signal. So, my idea was to add one day a board which could read digital signals, being triggered by the trigger output from the scope, so that it would be possible to have both data sent to a PC. This board from Sparfun seems a good candidate for reaching this purpose, but the problem is I can't understand if the Rigol and the Instek have a sort of trigger output (which is a typical feature). The Rigol has a strange BNC on its back, just take a look at the pic...

[joelby]

The Rigol has a strange BNC on its back, just take a look at the pic...

It's the pass/fail test output. Hook it up to a siren.

[alm]

However, anyone can see that two channels can become too few for some things, especially if you deal with mixed-signal. So, my idea was to add one day a board which could read digital signals, being triggered by the trigger output from the scope, so that it would be possible to have both data sent to a PC. This board from Sparfun seems a good candidate for reaching this purpose, but the problem is I can't understand if the Rigol and the Instek have a sort of trigger output (which is a typical feature). The Rigol has a strange BNC on its back, just take a look at the pic...

The Open Logic Sniffer (actually built by SeeedStudio) is a much better deal than the Rigol Logic Analyzer board (which adds like $300-400), although it's nice to have logic and analog signals on the same timebase. I think the software is still a bit rough around the edges, but is being worked on. I don't think any of them has a trigger out unless they state so in the manual. The BNC bus on the Rigol is labeled pass/fail output, so it's for use in an automated test setup, not much use outside that. The OLS might have the option for trigger out that you could connect to the trigger input on the scope.

[scrat]

I was looking for the opposite (trigger output on the scope, trigger input on the logic analyzer), since scope's trigger should be much better and much more flexible (and can be set on a further channel).
Although I thought it was a typical feature, I took a look in the lab at a Tektronix (TDS3032) which seems not to have a trig out, while an older Agilent (52622D, maybe an ancient of the Rigol) has it on its back.
If a k€ Tek doesn't have it, it's not fair to want it on a Rigol!

[alm]

I think a trigger out was a new feature on the TDS3000B series, together with superficial features like a USB port. Got to save something for the next version . Don't think the older Agilent series were made by Rigol, they probably didn't even exist back then. Connections like trigger out seem more common on older scopes, for some reason.

The scope triggering the logic analyzer or visa versa is both useful. The scope is better at triggering on analog parts (eg. level, rise time), but the logic analyzer is better at triggering on digital patterns. I just suggested reversing it because that might actually be possible without major surgery on the scope.

[scrat]

The scope triggering the logic analyzer or visa versa is both useful. The scope is better at triggering on analog parts (eg. level, rise time), but the logic analyzer is better at triggering on digital patterns. I just suggested reversing it because that might actually be possible without major surgery on the scope.

OK, this should work and will be the simplest way, but...
..at the cost of that useless BNC on the back, why didn't they put trig output on their scopes?  (couldn't the pass/fail test or calibration be done via a USB/serial signal?)
However, the Instek competitor doesn't have that output too.

[alm]

OK, this should work and will be the simplest way, but...
..at the cost of that useless BNC on the back, why didn't they put trig output on their scopes?  (couldn't the pass/fail test or calibration be done via a USB/serial signal?)

I agree, pass/fail is useless for hobbyists or R&D, although troubleshooting and production testing is probably a significant part of the market. I'd rather have a trigger out, too. Pass/fail could be done via USB/serial, but then you need a more intelligent to receive it. The BNC jack can be used to connect to simple things like a light or part of the conveyor system, without using a computer with custom software.

[cyberfish]

Got a reply from Instek (actually "TRADEPORT ELECTRONICS GROUP", which I am assuming is their distributor/representative in Canada, since I selected Canada on the contact form).

Hi Matthew,

Thank you for your interest in Instek test and measurement products.


The GDS-1000A series has 2 M long memory which is not dependent on sampling speed.  The 1, 2, and 4 that you mention refers to the save feature on the SD card.  The memory will be 2M long if capturing and saving one channel but will split to 1 M if capturing two channels. It has a fast save 4K feature which will capture short events.

The 1Gsa/s is applicable for one channel but is split to 500Msa/s for two channels.

So it seems like 2M is available at 1GSa/s. Guess we will find out soon enough! I have pretty much decided on the Instek (talk about switching back and forth). The only advantage of the Rigol is the 100MHz BW hack, but I don't think I will need that any time soon, so memory depth is more important, and build quality appears to be better.

[rf-loop]

I found a review for the 150MHz version.

http://welecw2000a.sourceforge.net/docs/Hardware/GW_Instek_GDS-1152A.pdf

First thing to notice is, the sampling rate is 500MSa/s when using both channels, so the manual is a little deceiving, since I would think it's a really important thing to mention, and they didn't mention it at all.

The ADC configuration is the interesting part.

It uses the same ADC (AD9288), except the 100MSPS version, but only 4 of them (8 ADCs, since they are duals) instead of 5 in Rigol. Which one is better depends on what theory you believe in. If you believe speed grade is just marketing and doesn't matter, Rigol is better. If you believe speed grade matters, Instek is better (25% overclock vs 125% overclock).

Thank you this pdf. In my own needs I am happy now that I select Rigol. (for some special use, becouse most time I use old Tek and HP)

About ADC. Rigol and Instek use exactly same ADC. (chip itself is exactly same... only speed stamp is different. It means that -100 chip reach ALL datasheet limits with this speed and -40 with its speed (quaranteed).  But first, oscilloscope do not need all these specs, they need only these things what are meaningful in this aplication. If we set different limits for some parameter we can drive it other than specified speed.

Rigol use exactly same chip with 100MHz and Instek use these 125MHz. (chip architecture is same!)

Rigol use -40 stamped and Instek use -100 stamped (specially selected chips). Maybe Instek need do this if they have no capacity to do they own chip tests for individual chips. (If Rigol buy chips and they can accept 50% of chips with they own aplication specified specs there is nothing wrong. It tell only that there are some very clever design engineer who understand how fundamental physic go in chips. Maybe he is not just after school.
If I run chip with all spec limits borders  and then with speed I go over... then I can maybe tell that I am overclocking.

(look datasheet max parameters and remember that specs means they all can be true in same time and chip works... all timing and temp etc things. If you find some max,min,quaranteed limit you do not need you can maybe find that you can use chip with different speed or some other parameter you can go over limits. If you need broke some other limit you maybe need adjust some other limit. You can try with some cheap components....

Example in old times I can use speed classified (memory) chips just as faster classified same chips becouse my temperature do not go at all this area where is datasheet max (also datasheets have some other min/max parameters what I did not need... if you know you do not need some you maybe can go over some other parameter becouse they affect each others. Did I "overclock". No. I only know how datasheet need read and how fundamental physic in chip go and then I make lot of tests so that I can proof it works.  Nobody tell that "oh you are overclocking"... they tell that oh you find very clever road to save lot of money... nice please... how you do it. And some "brown letter" I get later.

Think what is behind this -100 or -40 stamp on chip. How this classification after product is made.

Also if can not sell any -40 chip I think they do not stamp and package any.
Same for -100. If nobody never buy these they can sell all with -40 stamped and save some money from separating and logistik.

[cyberfish]

About ADC. Rigol and Instek use exactly same ADC. (chip itself is exactly same... only speed stamp is different. It means that -100 chip reach ALL datasheet limits with this speed and -40 with its speed (quaranteed).  But first, oscilloscope do not need all these specs, they need only these things what are meaningful in this aplication. If we set different limits for some parameter we can drive it other than specified speed.

That's one theory. Of course ADI won't agree with you, and we have no way of finding out if that's true.

Rigol use -40 stamped and Instek use -100 stamped (specially selected chips). Maybe Instek need do this if they have no capacity to do they own chip tests for individual chips. (If Rigol buy chips and they can accept 50% of chips with they own aplication specified specs there is nothing wrong. It tell only that there are some very clever design engineer who understand how fundamental physic go in chips. Maybe he is not just after school.
If I run chip with all spec limits borders  and then with speed I go over... then I can maybe tell that I am overclocking.

Yes, of course engineers start running chips out of specification when they get more experience... wait what?!

In a bag of 10% tolerance resistors you won't find any within 1% tolerance. Because those would be in the 1% tolerance bag.

It's the same with speed binning. Some ADCs can satisfy the noise specifications at 100MHz, and they get marked as 100MHz parts. Some can only do it at 60MHz, and they get marked as 60MHz. Some can only do it at 40MHz. Some can't even do 40MHz and they get thrown out of the window.

This is what the process is supposed to be like. And in this case, 100MHz parts will be much different from 40MHz parts.

Of course, it's also entirely possible that their manufacturing process have matured so much that they now produce more 100MHz parts than they need, and are down marking them to fill the lower market segments.

Which theory is true I have no idea and have no way of finding out, so I left it as an open question.

It sounds like you have special insider information from ADI that tells you the second theory is true?

Just because chips are made the same doesn't mean they will perform the same. That's what speed binning is for.

Example in old times I can use speed classified (memory) chips just as faster classified same chips becouse my temperature do not go at all this area where is datasheet max (also datasheets have some other min/max parameters what I did not need... if you know you do not need some you maybe can go over some other parameter becouse they affect each others. Did I "overclock". No. I only know how datasheet need read and how fundamental physic in chip go and then I make lot of tests so that I can proof it works.  Nobody tell that "oh you are overclocking"... they tell that oh you find very clever road to save lot of money... nice please... how you do it. And some "brown letter" I get later.

Yes, you are overclocking.

The datasheet says, it will run at 40MHz at -25C to 125C or whatever. It DOESN'T say it will run at 100MHz at 25C. You are GUESSING it will work and making that number up.

I'm fairly certain Rigol engineers don't have access to the ADCs design.

And I'm also fairly certain Instek has the ability to test ADCs if they have the ability to make oscilloscopes.

[rf-loop]

Rigol make oscilloscope and in this aplication they use many components. Rigol give specification for scope.

AD give specification for single ADC and with specified parameters. Chip desingn is 100Ms/s ADC. There is not any kind of overclocking if drive it with 100M. But AD do not quarantee that it can do it with all specs limits.

AD tell that maximum speed is minimum 40M, 80M or 100M so that they guarantee it for full specified temp range and with Vdd = 3.0V and Vd = 3.0V. (chip Vdd and Vd range is 2.7 - 3.6V abs max 4.0V) and they guarantee this minimum speed in whole temperature range up to +85 celsius.

AD tell maximum conversion rate 40, 80 or 100 minimum. They do not tell maximum. (if you find it can do it in your aplication and aplication specs so it is ok.) If manufacturer tell that maximum temperature is example 100 celsius and you use this kind of component 110 celsius we can tell that you run it overheating. manufacturer do NOT give maximum conversion limit! Not even typical. They tell only guaranteed minimum what reach all other parameters also with this speed and also they tell that it is only just with they testing methods. (who know what are test machine impedances and so on... )

AD tell also minimum conversion speed. It is 1MSPS. Oh my god... they are underclocking?
Now I can ask... why nobody tell that they are underclocking? I want big mouth talking all over internet that Rigol is underclocking  ADC's
(yes but in real they are not... AD give: Minimum conversion speed is 1MSPS maximum.

Of course SINAD, effective bits of conversion, second harmonic, signal to noise can differ what AD quarantee. But agen Rigol oscilloscope specifications are not AD's ADC specifications. Rigol give they aplication specifications - Not AD.

I can see that term "overclocking" have lot of misunderstoods... maybe becouse PC "overckocking".

Who make aplication is full responsible how it works or not and what is aplication specs.
There are lot of things out from datasheet. With different limits, different temp, different voltage etc. If I find that datasheet tell "some thing" 20ns maximum at 75 celsius. And if my lab test for componet tell that 15ns at 55 celsius and if I know my aplication need maximum 40 celsius. I use it in my aplication as 15ns component. What I am "overspeeding"... exactly nothing. I only use more cheap chip becouse I do not need all these things what are specified for more expensive chip becouse my aplication do not use these limits what are for datasheet specs.


It is totally different case also if manufacturer give direct minimum and maximum values example like this (AMD):



If you go under min limit you are underclocking and if you go over limit you are overclocking. (still it may also work if adjust some other limit different as datasheet. Voltage, temperature etc... or if can accept more errors.. (typically we do not accept processor errors becouse system crash... but A/D conversion is different...

AD have not tell maximum... it have only tell minimum! Where is so overclocking. Where is manufacturer set limit what you go over? Guaranteed minimum is not maximum value what you can use if you aplication accept it. And they have test it... they have sell some amount of these... have you any data that something is wrong? It do just as they promise. (there is not also any problem with higher speed becouse chip itself have designed for 100MSPS.

Very interesting small detail is that -40 classified chip have less power dissipation with 100MSPS  test run if compare to -100 labeled chips in 100MSPS test. (source also AD).

If some have measured ADC9288 Vdd and Vd voltage in Rigol scope, what you have find?

[cyberfish]

Chip desingn is 100Ms/s ADC. There is not any kind of overclocking if drive it with 100M.

Of course there is.

If you have an Intel CPU designed for 4GHz, but tested to only run at 2GHz, and sold as a 2GHz chip, and you run it at 4GHz, is that not overclocking?

I understand what you are saying about the ADC MAY be able to run at higher speed if you don't need the full temp range. The problem is, we don't know by how much, or even at all. Why 100MHz? Why not 200MHz?

The correct engineering practice is to adhere to the 40MHz guaranteed speed. If you run it higher because you GUESS it can run higher at your smaller temperature range, you are trying your luck. The manufacturer (ADI) doesn't guarantee anything at this speed. This batch can work, next batch may not. Or it can have a shorter life, or whatever. It's your problem.

It's not possible to test all ADCs. At best, they picked a few samples, and found they work fine. "Ah well, that will do". For the cost of testing all ADCs themselves, they could've just gotten 100MHz parts.

If it's such an honourable practice, why are they sanding their chips to prevent people from finding out? (and doing such a poor sanding job that people still found out)

Very interesting small detail is that -40 classified chip have less power dissipation with 100MSPS  test run if compare to -100 labeled chips in 100MSPS test. (source also AD).

That makes perfect sense. Higher power dissipation = lower impedance = better silicon. For example, it can allow them to charge up the charge and hold capacitor more quickly, allowing higher sampling rate at high accuracy. This just proved the point that there are actually differences between the 40 and 100 chips.

From Digikey,
AD9288BSTZ-40 = $6.32
AD9288BSTZ-80 = $10.09
AD9288BSTZ-100 = $17.09

The total cost of the 4 ADCs Instek is using (17.09 x 4 = 68.36) is much higher than Rigol's 5 ADCs (6.32 x 5 = 31.6). This is for single quantities, but the ratio should be similar at high quantity. The 100MHz parts cost almost 3 times as much as the 40MHz parts.

Wouldn't you think there is a reason for them to use 100MHz parts?

I have read quite a bit about Rigol in the last few days on Chinese forums (it's probably the most popular oscilloscope brand there), including other products like a 6 digits multimeter that can only do 4 digits, and a signal generator with fly wires everywhere, etc.

And the build quality of the scope -
http://home.comcast.net/~ajawam1/rigol/RIGOL_DS1102E_GUTS.html
compared to what's shown in photos in the PDF.

To be honest, I have very little confidence in Rigol as a company.

In my opinion, build quality says a lot about the company. If they can't even do the soldering and sanding right, do you really trust them to make a good oscilloscope? And they are cutting cost on one of the most important components, the ADCs, and sanding them to prevent people from finding out.

If the Rigol scope costs $100 less or something it can be a sensible choice.

At the same price, it would be hard to choose the Rigol.

Of course, Tektronix or Agilent would be even better. But they cost about twice as much.

[rf-loop]

But Agilent = Rigol... sometimes.

This is Agilent "quality. 

One point is that I do not at all understand why Rigol grind ADC labels off. This is just "stupid". Maybe they do it just for persons who do not understand these speed things. (have you see any Rigol DS1000E where 100MSPS/ADC fails? Do you know why? )

BTW as AD datasheet read. They test 100% (for this parameter), not only random samples. Also -40 chips. It all read in datasheet.