Choosing an op-amp for your project
83 comments
·January 3, 2025dragontamer
kazinator
Amplification of the "common mode" means that if we tie the + and - inputs of the op-amp together, so that it's not doing its job (amplifying the difference), it's possible that the op-amp amplifies the voltage present on the combined + - node; the voltage between that and the ground reference. Ideally, the amplifier should not do that, but they do. The gain is not nearly as much as the differential gain, so I think the CMRR (common mode rejection ratio) is something like how many decibels down from the payload gain is this unwanted common mode gain.
Common mode gain can cause or contribute to an output offset. But if the amplifier circuit is biased such that + and - are kept very close to zero, that should not be the case. We normally care about the common mode rejection because common mode noise makes the + and - inputs swing together, causing the amplifier to amplify noise, like RFI/EMI.
In your current sense example, although we have a 10% error, it is a systematic error (a displacement), not 10% uncertainty in the value. It is an inaccuracy, not an imprecision.
You're assuming that the amplifier just buffers the output of the current sense resistor, so that the +/- 0.05 V across the resistor is translated to the output of the amp via unity gain.
We could configure the amplifier to amplify the signal, so that it becomes +/- 5V (or whatever). Then the Vos of 0.0005 will be a 0.1% inaccuracy.
auxym
Correct.
A 10% error can be calibrated out if it is constant (in practice it probably varies a bit with temperature).
But for measuring thermocouples or strain gauges, for example, 50 mV (your example) can be 100-500% of the signal, which becomes impractical to calibrate-out (due to maximum output levels, etc).
For these applications, Vos is one of the first things to look at. Another one is the temperature coefficient on the gain. High frequency noise metrics such as CMRR and PSRR are sometimes important if you're looking at high frequency signals, but most of the time mechanical phenomena don't have much interesting content above a few 100s of Hz, and high frequency PS or CM noise can be removed by a simple high pass filter.
kazinator
Power supply and common mode noise can be 60 Hz.
kragen
And its harmonics, if you're standing near a fluorescent lamp.
dragontamer
> A 10% error can be calibrated out if it is constant (in practice it probably varies a bit with temperature).
Unfortunately, Vos on cheaper general purpose OpAmps is the kind of thing that varies by... voltage. Ick.
> For these applications, Vos is one of the first things to look at. Another one is the temperature coefficient on the gain. High frequency noise metrics such as CMRR and PSRR are sometimes important if you're looking at high frequency signals, but most of the time mechanical phenomena don't have much interesting content above a few 100s of Hz, and high frequency PS or CM noise can be removed by a simple high pass filter.
No. CMRR is about DC in the applications I'm talking about. It's weird because CMRR is listed in decibels but it's absolutely a DC spec.
If you have a high side current sense circuit with common-mode voltages of 24V +/- 0.05V (ex: 24V power supply that dips to 23.95V at 5Amps), CMRR tells you how accurate you are here.
Your typical 60db (btw I need to kill the engineer who decided db measures DC noise/errors....) means that the 24V of common mode voltage (which is the 24V DC power supply in this case) leaks into your measurements.
Or in other words: 60db * 24V == 3 decades or 24mV of 'Noise' aka your +/-50mV signal/measurement got completely wiped out by your DC errors. Like 50% error bars on your signal now, good luck with that.
That's the real issue with OpAmps. There's surely an OpAmp out there that solves your problems. But it requires knowing the general tradeoffs and picking-and-choosing different parts for different purposes.
Secondly, the specs are not intuitive. 60db CMRR sounds like a high frequency issue but becomes DC in this case.
You could of course go full isolation (optoisolators) that allows you to shift voltages down to near zero (removing CMRR issues) but that's money and additional parts.
You could go low-side voltage sense but this doesn't work for all circuits (most circuits are fine with Vcc error, not Ground errors). So high side current sense is the most flexible and generic well engineered solution. So long as you choose the correct OpAmps.
-------
As far as when this could be useful: consider Maximum PowerPoint Tracking for solar. 0-24 V and 0-1 Amps. And a need to accurately measure Amps and Volts from this entire range. (A variable load + voltage converter like a switch-mode power supply + battery can search for the optimal Voltage/Current combination to maximize the Solar Panels power).
Yes the microcontroller will do the bulk of the math. But the initial multiplies and subtract is best handled by an OpAmp.
kazinator
The difference in brightness between two incandescent light bulbs running on DC could be expressed in decibels. Log10 of the ratio of wattages times ten.
Decibels are useful for any intensity scale that occurs in a context where something responds in a logarithmic way, regardless of whether the power phenomenon is steady or pulsating.
hn4352
> It's weird because CMRR is listed in decibels but it's absolutely a DC spec.
If you get the Franco book equation 5.27 (my edition is the 3rd) explains why they do that. Long story short: It's a convenient form when CMRR = dVcm/dVos due to the orders of magnitude involved.
hn4352
I was kinda shocked by the Vos comment as well.
On CMRR, in some mathematical treatments it's modeled as a change in offset voltage with respect to common mode, which indirectly effects output voltage of course so at the end of the day it's the same result. (See: https://www.google.com/books/edition/Design_With_Operational... highly recommended )
It's also odd that the 741 was dismissed, as it should be, but the TLV9301 was not recommended. This part is specifically called out on TI's 741 page as what to use instead in 2025. Not only does it perform better in basically every possible spec, it's also a drop in replacement for most, if not all, applications.
https://www.ti.com/product/LM741 https://www.ti.com/product/TLV9301
TLV9301 ($0.5) is also cheaper than a MCP6272 ($0.88)
nine_k
For current measurement, I'd rather use a Hall sensor. But to measure the signal from that sensor a sensitive opamp is needed, and your comment applies :)
blackguardx
Using Hall-effect sensors to measure current isn’t that common when currents are in the uA to mA range. I’ve also never seen them used even for Amp range measurements when the application is purely on-board power supply meaurement.
nine_k
The example uses 5A.
For mA range, of course not.
analog31
Or a high side current sense amplifier...
https://www.st.com/en/amplifiers-and-comparators/tsc1031.htm...
Hall sensors drift.
kragen
I was puzzled about that too, so I appreciate you confirming what I thought.
I'd think the main benefit of using lower-value current-sense resistors in this application would be that the resistor would heat up less, so its resistance would be more stable?
dragontamer
Accuracy is likely secondary. I expect that most applications are good with 1.5 digits (aka 95% accurate or so). You don't want to blow the entirety of this 5% allowable error on one micro-spec of one component, but I don't expect that most people especially need lots of accuracy here.
The issue is that any circuit with 1 to 5 amps of current is a serious amount of power, meaning power efficiency is likely one of the top priorities.
A 5-Amp circuit with a 0.01 Ohm sense resistor wastes 250mW on the resistor alone, likely more than the entirety of your microcontroller!! You can actually run an entire Linux capable microprocessor + Low-power DRAM off of that kind of power!!
Dropping down to 0.0001 Ohms uses 1/100th the power or 2.5mW. which is likely a more reasonable cost.
kragen
That depends on what "most applications" means. I remember a paint program I saw in high school that some kids had written to draw sprites and backgrounds for their video games. The documentation explained that it could only edit 320×200×256 images, but that that should be adequate for "most projects". Depends on the context!
In the contexts I'm thinking of, I would think that, if your load is drawing 5 amps of current at 3.3 volts, which is 16.5 watts, an extra 0.25 watts in the 10mΩ current shunt is not likely to be a big problem. And if it's 5 amps at 48 volts or 240 volts, it's even less of a problem, relatively speaking. I guess you're thinking of different contexts, contexts where the power-measurement system is paid for from a different budget than the load, but I can't figure out what they are.
Tade0
Back in college I was told to not think too hard about Vos, as it's typically temperature dependent so you need to assume it's non-zero or will be in some circumstances and compensate with feedback anyway.
kragen
How do you compensate for Vos with feedback?
hn4352
You probably don't, but your opamp might. https://www.analog.com/en/resources/technical-articles/to-ch...
Technically you probably could do it externally in most cases but it would require a bunch of extra stuff, and be a pain, so usually it's best to use the stuff built into the amplifier itself.
kragen
I found this post extremely interesting and informative, well above my expectations even given the eminence of its author. I'm not sure that everything in it is covered in The Art of Electronics — it's the kind of stuff the book covers, but I learned things from this post I didn't learn from the book. (But possibly it's information that's in the book that I just failed to absorb the last few times.)
I was recently looking at opamps as alternatives to the LM324 and found some interesting-sounding parts, in particular for a poor man's SMU application (precision, low current, and voltage requirements, but not much bandwidth). Haven't tried any of them yet. Comments would be welcome.
- LM324B: TI's improved LM324, with half the input offset voltage and otherwise improved ratings, and just as cheap, but still bipolar.
- OPA4197 and family: three dollars but it's a quad RRIO 36V 10MHz opamp in a 14-SOIC with ±15nA input bias current, ±100μV max input offset voltage, and 120dB min CMRR. The datasheet makes it sound amazing for the price. The OPA177 seems like it would be better but pricier.
- OP4177ARUZ: a 16-dollar quad 36V 1.3MHz opamp with ±2nA input bias current, 75μV max input offset voltage (at ±15V power supply), and min 120dB CMRR
Then I decided I'd screwed up my design sketch by requiring one of the opamps to sink significant current very close to the negative rail, which is something even "rail-to-rail" opamps can't do; I was planning to use millivolts from ground to represent measured nanoamps. If you want to look at a simulation with idealized opamps, it's at https://tinyurl.com/2aomvpn5, but don't take it as exemplary in any sense; it's a novice design with novice mistakes (and I would be grateful in the unlikely case that someone took the trouble to point some of them out). I think I need to redesign the circuit as a bipolarity-supply circuit or something, or use a differential output for the current measurement, or rethink it entirely.
Cervisia
Nowadays, it is hard to recommend a general-purpose opamp. Just plug the desired parameters into the search function and sort what's left by price.
(Distributors like DigiKey and Mouser have somewhat adequate search functions; I usually have to go to manufacturers' web sites like https://www.ti.com/amplifier-circuit/op-amps/general-purpose... to be able to filter by all important parameters. I'm mentioning TI because they have a large selection and a good search; even when you do not end up selecting on of theirs, you see what is possible.)
___
If you need only a small negative supply and have nothing else, the LM7705 charge pump can generate −0.23 V. (This is designed to fit into the typically allowed 5.5 V range of a nominal 5 V opamp.)
I do not know what a "significant current" is for you, but there are opamps with strong outputs. (When comparing opamps, you usually have to estimate the drive strength from the short-circuit current.)
fsckboy
>the LM7705 charge pump can generate −0.23 V
fwiw I searched and found "A newer version of this product is available Same functionality with different pin-out to the compared device LM27761"
kragen
Thanks! This is very useful advice!
What I meant by "requiring one of the opamps to sink significant current very close to the negative rail" is that, if you look at the schematic, the differential-to-single-ended op-amp that measures the voltage across the current-sense shunt resistor is using 10kΩ resistors in its feedback path, and the inverting input to that feedback network might be close to the positive voltage rail, say 12V, while the single-ended output is ideally millivolts from ground. So you have 12 volts across 20kΩ, which works out to 600μA, which has to be sunk into that op-amp's output.
600μA doesn't sound like a lot, and it certainly isn't going to strain the drive strength of any op-amp IC, but in this context we're hoping for millivolt precision down near the negative rail. The OPA4197 datasheet https://www.ti.com/lit/ds/symlink/opa4197.pdf figure 14, "Output Voltage Swing from Negative Power Supply vs Output Current (Maximum Supply)", shows what you might call a gently nonlinear output impedance roughly in the 40–80Ω range depending on temperature (2–4V at 50mA), which means 0.6mA of output current works out to tens of millivolts (24–48mV using those nominal impedances). Worse, even under no-load conditions, it's rated to swing only down to as much as 25mV from the negative rail (§6.7, "Electrical Characteristics: VS = ±4 V to ±18 V (VS = 8 V to 36 V) (continued)", p. 8, "Vₒ: Voltage output swing from rail, Negative rail").
In retrospect, it seems obvious that the op-amp's output isn't going to be able to reach beyond the input rails (unless it integrates a charge pump like the LM7705 internally) and is going to have trouble getting too close to them when it's sinking any current (for the negative rail, or sourcing for the positive). Because where is that current being sunk to? You need some voltage drop to get the electrons and holes to move in the desired direction through the silicon. A small negative supply might be the right solution. Or a differential output, which would be easy.
dragontamer
A general purpose OpAmp is just that, your general purpose first choice.
If you know more specific information about your circuit or it's application, the. You can specialize. But general purpose OpAmps are jack of all trades with specific known weaknesses to avoid.
In most cases, you calculate the error bars and none of the errors matter, so sticking with a cheap general purpose amp is best engineering.
shadowpho
>Nowadays, it is hard to recommend a general-purpose opamp. Just plug the desired parameters into the search function and sort what's left by price.
This 100%. If you need a comparator get a comparator not an op amp. Current measuring? There are specialized chips for that as well, etc.
kragen
In this strong form, this is excellent advice for someone who is not me and is not doing what I am doing.
I live in a third-world country where importing chips from abroad is expensive, unreliable, slow, and sometimes dangerous. There are circuits I cannot build because I cannot get the very specialized parts they need. Obviously a linear power supply that can measure how much current it's supplying is not such a circuit, unless you have very stringent precision requirements.
It would be to my benefit to figure out a relatively small set of parts I can buy, ahead of time, in bulk, to cover a wide range of possible circuits. Better still if they're so popular that local distributors have them in stock. An analog comparator probably needs to be in that set. A chip specialized for current measuring probably does not.
If you're designing a product for mass production that needs to be competitive in the market, you can't do it that way. Super-specialized parts will always have better performance, and usually better price/performance than overpowered general-purpose parts. (Also, you need to live in Shenzhen.) But hobbyists have other priorities.
dragontamer
> Then I decided I'd screwed up my design sketch by requiring one of the opamps to sink significant current very close to the negative rail, which is something even "rail-to-rail" opamps can't do; I was planning to use millivolts from ground to represent measured nanoamps.
Get an OpAmp specifically designed for current sense applications.
OpAmps for current sense applications have high accuracy near 0 Volts and Vos measured in single digit microvolts.
Oh, you WILL lose bandwidth with these designs. So make sure you are allowed to be much much slower.
kragen
Thanks! This is probably excellent advice, and I wish I could follow it.
Being slower is not a big problem, but needing specialized parts might be, due to supply-chain issues.
(More detail in https://news.ycombinator.com/item?id=42627042.)
cushychicken
Oh boy you need to take another look at ToEv3 if you think this has better coverage on opamp topics. It has so much about opamps in it that it’s hard to tell you where to find it all - it’s so ubiquitous and spread out in the text.
Recommend Ch5 for Precision Design and Ch8 (or 9? Can’t remember) on noise.
kragen
I didn't mean that this post was "better" overall, but rather that it contained some information I didn't see in AoE (or didn't retain). Clearly AoE's presentation of opamps is far more comprehensive.
peteforde
I've watched the videos and agree that the OPA2323 does appear to sound amazing by subjective comparison to the TL071.
However, it's not enough to say that it costs ~8.7x as much in local tokens because that doesn't reflect how much it costs as a percentage of the other components around it, many of which have a more profound impact on the outcome. This wouldn't stand as a good argument except that in many devices, you have a ceiling for how much your total BOM can cost before you've priced yourself out of competition.
In the end, a 8.7x cost bump is a lot to swallow for a feature that most consumers are physically incapable of distinguishing between. Every time I've raised the possibility of using the fancy new chips, vastly more experienced engineers than me have come out of the woodwork to tell me that in almost all scenarios, the tradeoff between price and quality isn't worth it.
Of course, if budget is not an issue, use the OPA2323. It really does sound great. Or more accurately, the degree to which it destroys good sound is as low as we can currently achieve.
(This comment originally stated a 12x factor, but I was terrible at math.)
Majromax
Is the cost delta really 12×?
Looking at Mouser Canada, it looks like the cheapest 071 is the TL071CDR, at $0.138/each (Canadian) in quantities of 5k. The OPA2323IDDFR is $0.49 in the same quantity.
> In the end, a 12x cost bump is a lot to swallow for a feature that most consumers are physically incapable of distinguishing between.
I think that the performance of an op-amp should very rarely have user-visible effects. The more interesting question is whether the more expensive chip can make for a simpler design elsewhere. For example, can a rail-to-rail amplifier save the extra cost of needing charge pumps and split-rail design elsewhere?
Also, not all domains should be cost-optimized. Hobbyist or prototyping work might best benefit from using a more expensive but more capable amplifier as a first choice, saving on the number of components that might need to be stocked in the home lab.
stephen_g
Yes, and if you are actually buying in that kind of quantity you should be able to do even better. TI’s budgetary pricing estimate is US$0.252 per 1K for the OPA2323IDDFR and US$0.067 per 1K for the TL071CDR.
peteforde
All good points! And I am no expert.
FWIW, here is what I was going off of, price wise:
https://www.digikey.ca/en/products/detail/texas-instruments/...
https://www.digikey.ca/en/products/detail/texas-instruments/...
In other words, the cheapest TL071 variant and the cheapest OPA2323 variant on Digikey Canada, in a quantity of 1 (ie wildly expensive). $0.31 vs $2.70 means that I shouldn't attempt math before coffee; 8.7x is still a big bump, although I acknowledge it's not the 12x I disinformationed earlier, with apologies to anyone reading.
rasz
If you are making a one-off $3 is laughable non issue, you will spend more for a lunch drink. If you are manufacturing something in the thousands the difference goes down to 0.252 vs 0.067 so merely $0.2 BOM bump. Also a no brainer if performance is on the line.
kragen
I think Zalewski's post is aimed at electronics hobbyists rather than high-volume producers or competitors. In that context, an OPA2323 might cost 98.99¢ https://www.digikey.com/en/products/detail/texas-instruments... vs. 7.958¢ for the TL071 https://www.digikey.com/en/products/detail/texas-instruments... but the Saturday afternoon you spend building and debugging your circuit costs hundreds of dollars in foregone income—if it's just an afternoon and not every weekend for a month.
So the BOM cost may not be a significant consideration for hobbyists, especially when weighed against things like familiarity or being able to keep a stock of a smaller number of less specialized components, as Majromax points out.
I'm surprised to see you say that a US$1 opamp is as good as we can currently achieve. Presumably there are Analog Devices chips that are better than the OPA2323 even for audio? Even if you can't hear the difference, you ought to be able to measure it.
peteforde
Quite possibly! I learn new things every day.
elcritch
Great write up. And definitely skip the 60 year old models.
Also valuable to look for is ESD protection. Many op amps have good 2kV static shock protection. Really sucks to have cascading failures due to touching your circuit and blowing an op amp.
For some uses it’s also good to checkout input current bias.
elcritch
Another bit I discovered is the quality of resistors. Pretty much any op amp circuit will use resistors. Often the resistors limit the static offsets.
Don’t stick with 1% for analog circuits as 0.1% precision resistors are cheap and come with low temp drift as well.
f1shy
To add to this discussion is useful the video or eevlog: https://www.youtube.com/watch?v=uq1DMWtjL2U
where Dave discusses the "Jelly bean" op amps
buescher
Be aware in a microcontroller design, an lm324-type op amp might outperform the on-chip ADC you’re using anyway. Even the cheapest op amps from major players are better characterized and specified on their data sheets than most microcontroller ADCs. There’s no virtue in spending other people’s money and supply chain risk to use a “better” part just because it wins a bench race or it’s newer or whatever.
pythonguython
What does it mean for an op amp to outperform an ADC?
buescher
It can have wider bandwidth, lower noise+distortion, smaller Vos, maybe other things I'm not recalling off the top of my head. In practical terms, you might not be able to run any test with the ADC that can find the limits of the op amp.
omgJustTest
"Frequency response: The most important AC parameter of an op-amp is known as bandwidth gain product (fGBP or fGBWP). Standard, fully-compensated devices are designed to have an internal gain that rolls off proportionally to signal frequency. At fGBP, this gain is reduced to one:"
"Stuff not to worry about:
Despite what content-farmed articles imply, most of the other parameters in the spec can be usually glanced over. For example, the exact value of open-loop gain (AOL) is almost never of real consequence; the same goes for input offset voltage (VOS) — even in high-precision instruments, the absolute value is less important than drift over time. In any case, the parameters are usually only eyeballed in most specs, so if you’re building sensitive instrumentation, you will still need to calibrate the readings using a known reference."
I rarely think about fGBP and AOL as separate ideas, the author could make the point more directly by saying they are related and that one is easier to select from a single decimal value. AOL or fGBP is a primary design consideration.
squarefoot
Any comments on the LM4562? As per data sheet it appears to be really good for audio wrt noise and distortion, yet I see it rarely mentioned. https://www.ti.com/lit/ds/symlink/lm4562.pdf
formerly_proven
It and the LME49720 (same part) were discontinued a couple years ago along with the entire LME Overture series.
squarefoot
You sure about the LM4562? I've just checked at TME, Mouser and Digikey, and all of them have many in stock. I's also marked as "nopb" so it could likely be a newer lead free version.
formerly_proven
It appears this particular part was un-discontinued in 2021(?). They certainly were all discontinued almost 10 years ago.
https://e2e.ti.com/support/audio-group/audio/f/audio-forum/4... (this list only contains the op-amps, not stuff like the also-discontinued LME49810)
ckocagil
Potential replacements for ultra low THD would be: OPA1612, OPA1656 and OPA1642 with bipolar, cmos and jfet inputs respectively.
magic_smoke_ee
There's a massive (old) list of them in Horowitz & Hill TAOE 3e Table 8.3a p. 522. I'm sure you can just go to Digikey or Octopart and do a parameter search for high-voltage, low-noise, BJT-input op-amps too. If one wanted to "use a Ferrari to go the grocery store", they could always use a $20 AD797 or LT1115 for audio applications. :o)
perriera
>f you’re designing a new op-amp circuit, here are some decent, all-around alternatives made in the 21st century
None of the listed opamps would be an alternative to where an LM324 or TL071 are typically used - low cost, 20V-30V supply, infinite/guaranteed availability. They are 5V only and cost 10x as much.
hn4352
Try the https://www.ti.com/product/TLV9301 $0.5 @ 1s of units w/ modern specs and a 40V supply range.
bratwurst3000
i build and repair audio electronics and the tl071 is used in abundance in audio circuits and is a bad choice imho. But good opamps for audio are at least a few dollars
emchammer
I have an M-Audio Delta Audiophile 2496 PCI sound card somewhere. The circuit board has an 8-pin socket with a replaceable op-amp. Why does a stereo sound card have only a single operational amplifier? This sound card already sounds very nice, but what kind of op-amp would I replace it with if I wanted to upgrade it?
hananova
If it's an 8-pin socket, that's perhaps for a dual op-amp.
bratwurst3000
thats nice and rare. Some modern dac have also this feature to please audiophiles. It seems to be an dualopamp. look up your opamp and if you are not happy with it you could replace it with a burr brown opa from texas instruments. I dont have your specific opamp so I cant give a specific recommendation but those audio opamps from burrbrown are fine.
Archit3ch
If you want the mirror the OpAmp distortion from the original units for authenticity/character, it makes sense to use the TL071, no?
shermantanktop
Based on my experience with vintage music electronics, it wouldn’t be good enough to use a component that just sounds bad like a TL071; it needs to be a TL071 so that the marketing copy can claim “vintage-correct parts.”
At that point you are engineering with a completely different set of tradeoffs than would be expected for a modern ground-up design effort.
bratwurst3000
thats true. I forgot to mention i mean electronics from mid 90s onward. Even in higher class electronics. for example many of those 2000 euro upwards dj mixer use tl071 and even high end dac and soundcards.
I replaced some of them with good opas and burrbrown opamps and for me it made , even if not big, audible changes to the better. But this could be my imagination.
whatever i prefere some opamps in the range from 2-4 bucks. even if its only for the feeling…didnt measure it.
spamatica
Any suggestion for a (reasonably) good and cheap one for new designs?
nicolaslem
The book "Small Signal Audio Design" by Douglas Self has an entire 50 pages chapter on selecting opamps for audio. The chapter roughly ends with this quote:
> One thing is obvious — the 5532 is still one of the great opamp bargains of all time.
bratwurst3000
i like the opa2134 pa , or in general the bur brown opa, on the expensive site and on the cheap one the ne5534.
bnetd
Come on fellas we're getting sloppy with our shitposts. Get it together.
CTRL+F "Op Amps For Everyone" (0 results)
https://web.mit.edu/6.101/www/reference/op_amps_everyone.pdf
bsder
I'm really disappointed that the article didn't mention what is probably the single most important characteristic in an opamp when being chosen by an amateur:
Unity gain stable
Sure, the other characteristics are important, but a whole bunch of circuits that beginners are likely to use rely on opamps being unity gain stable. If they're not unity gain stable, the circuit will do very weird things, and a beginner won't know why.
Of course, debugging issues like that are how you eventually become an expert.
Vos doesn't matter until it does. I find it surprising to see Vos mentioned as likely unimportant right next to CMRR, as CMRR and Vos are innately related in some circuits.
In particular, any low voltage current sense circuit is going to require very precise Vos. Let's say you have a 0.01 Ohm current-sense resistor on a 5-Amp or so circuit.
Your current-sense is now in the range of +/- 0.05Volts (!!!!). So a Vos of 0.005 would represent a 10% error. Likely too much for most applications.
In effect, CMRR has become hyper-sensitive to Vos in this particular use case, to the point that Vos is suddenly the most important statistic.
Fortunately, there are specially designed low-offset chopper or auto-zeroing OpAmps like MCP6V26 or whatever out there.
MCP6V26 has Vos of 2uV, or in relatable terms... 0.000002 Volts (!!!!). Meaning it is more than sufficient at reliably making this current-sense application. Indeed, you can drop down to 0.0001Ohm resistance and still have high accuracy (and power savings compared to the earlier assumption).
------
Alas, nothing is ever free in life. Chopper Amps have noise issues and other designs have very very low bandwidth (which is truly an important statistic for most circuits).
Choosing a chopper amp specifically is making a Vos tradeoff with Bandwidth. So only choose if you know what you are doing (aka, dealing with very low voltages and needing the precise zero).