Wait, seriously?
That can't be right. Even my old MSO6000 series scope does 100K updates/sec.
If that's true, I can see why Keysight literally had to give those scopes away. Somebody dun goofed.
Wait, seriously?
That can't be right. Even my old MSO6000 series scope does 100K updates/sec.
If that's true, I can see why Keysight literally had to give those scopes away. Somebody dun goofed.
They are talking about lower timebases, and they don't mention memory depth or what mode is being used (e.g. FastAcq)
But the scopes they are comparing to are signal analysis scopes, they aren't designed as fast updating general use scopes. That's why Keysight have the X and S series. The X is the fast updating series.
Notice how they have these two big huge graphs for update rate for the Keysight and the Lecroy scope, but such graph of update rate in the R&S RTO2000 comparison, just a small mention in the bottom table.
Why? because the RTO2000 beats the crap out of the Tek in update rate.
Classic comparison document,m they leave out stuff that isn't to their advantage. No fault of Tek, all the manufacturers do it. Never take them at face value.
They are talking about lower timebases, and they don't mention memory depth or what mode is being used (e.g. FastAcq)
But the scopes they are comparing to are signal analysis scopes, they aren't designed as fast updating general use scopes.
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
It sounds like I'm a bit behind the times.
More likely, I'm misinterpreting how the newer Agilent/Keysight scopes actually work. Visual dead time, which is what's important ("probability of intercept" in frequency-domain terms) doesn't necessarily have to depend on the trigger rate, which is what Tek is presumably benchmarking for their brochure.
It's safe to say that Tek is using their FastAcq mode for a comparison of this nature. Their graph covers timebases from 200 ps to 0.1 second.
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
Correct, they are not designed for general use were update rate matters. It's all about signal integrity, analysis, and maximum memory depth at all times. Unlike Tek that like to promote their high update rate using FastAcq no doubt that has pathetic amount of memory. Well ok, nothing wrong with promoting that if that's what they want to do. It's up to the user to understand the difference and their requirements.
No one buys an S Series scope for day to day general lab use.
It sounds like I'm a bit behind the times. More likely, I'm misinterpreting how the newer Agilent/Keysight scopes actually work. Visual dead time, which is what's important ("probability of intercept" in frequency-domain terms) doesn't necessarily have to depend on the trigger rate, which is what Tek is presumably benchmarking for their brochure.
They won't tell you, but I'd bet their figures are only valid using FastAcq mode.
Would you like to back that up with sources and pricing? I'm working on this sort of thing day to day, appropriate screens are much more expensive than you might think, its not a high volume consumer product that buys out capacity for a few months they need a stable supply of screens for years to support T&M products.
If they used standard laptop screens then replacing is not a problem even years after. But they use mate screen which is very rare... But may be different coating is not a problem, I dunno (may be it's just a film?).
I'm not sure why you mentioned temperature, it doesn't look like a field scope, nor the temperature range is specified in MSO58 datasheet.
Seriously, take the entire quote:
Well, let's hope the developments behind this enterprise price level scope trickle down to sub $5000 range units.
Though not specific to this new high end Tek scope, my bet is that larger, touch screens will become the new norm even for the < $5k mid-range and eventually even entry level scopes. That will probably also mean fewer knobs and with the LED light coded vertical knob or some variation.
I'll be surprised if the next mid level scope from Rigol doesn't follow that trend and I'll bet the next update to Tek and Keysight's mid level scopes will have those features as well.
Screens (touch or not) are more expensive than physical controls, so unless there are some "big" advantages they wont replace physical controls in the low end.
No, not in 2017 they're really not. A handful of half decent quality rotary encoders, a large PCB and a load of buttons and LEDs costs -way- more than a decent high resolution capacitive touch screen.
Would you like to back that up with sources and pricing? I'm working on this sort of thing day to day, appropriate screens are much more expensive than you might think, its not a high volume consumer product that buys out capacity for a few months they need a stable supply of screens for years to support T&M products. Then you get other constraints like:
Operating Environment: 0 ºC to +50 ºC
A UI board with some encoders, membrane buttons, LEDs etc is much cheaper than screen space to replace it. If you just slap a touch digitiser on top of your existing screen then it would be cheaper than physical controls but thats not what people are talking about here.
The question was why wont sub $5000 scopes have
larger touch screens and fewer controls. The larger scopes have had mouse/keyboard and touch screen interfaces alongside knobs for a long time, and they slowly year on year see less physical controls and more touch interaction. But those are products that can absorb the costs of those screens and bury it in the noise of the other parts in the BOM, they also have less stringent environmental specifications making it easier to find suitable screens.
Screens (touch or not) are more expensive than physical controls, so unless there are some "big" advantages they wont replace physical controls in the low end.
No, not in 2017 they're really not. A handful of half decent quality rotary encoders, a large PCB and a load of buttons and LEDs costs -way- more than a decent high resolution capacitive touch screen.
That's just the hardware cost. Don't forget the cost to develop good software (UI) around the touch screen. Maybe that's a sunk cost though, since even if you have a small screen it's going to be touch.
Well, since I can buy an Android tablet with a 10.1" 1024 x 600 capacitive touch screen for $60, it seems hard to believe that the hardware cost is preventing these from being implemented on low to mid range scopes. Software development costs money of course, but any new scope will require that, large or small screen, touch or not.
[sarcasm]We all await your fantastic new scope, let us know who you pick for your screen supply contract so we can all access long lived cheap high quality LCD panels.[/sarcasm] Why not skip the embedded system entirely and just build a box which you connect the tablet to.....
https://www.lab-nation.comOpen source and ready for you to build upon.
Well, let's hope the developments behind this enterprise price level scope trickle down to sub $5000 range units.
Though not specific to this new high end Tek scope, my bet is that larger, touch screens will become the new norm even for the < $5k mid-range and eventually even entry level scopes. That will probably also mean fewer knobs and with the LED light coded vertical knob or some variation.
I'll be surprised if the next mid level scope from Rigol doesn't follow that trend and I'll bet the next update to Tek and Keysight's mid level scopes will have those features as well.
Screens (touch or not) are more expensive than physical controls, so unless there are some "big" advantages they wont replace physical controls in the low end.
High quality rotary encoders are very expensive, whereas 7", 10" and 13" Retina screens are commonplace now and will continue to be commonplace.
I'm thinking a screen about the size of an iPad mini would be perfect for a low end scope. And trust me, the extra resolution *does* make a difference, even from a distance.
Well, let's hope the developments behind this enterprise price level scope trickle down to sub $5000 range units.
Though not specific to this new high end Tek scope, my bet is that larger, touch screens will become the new norm even for the < $5k mid-range and eventually even entry level scopes. That will probably also mean fewer knobs and with the LED light coded vertical knob or some variation.
I'll be surprised if the next mid level scope from Rigol doesn't follow that trend and I'll bet the next update to Tek and Keysight's mid level scopes will have those features as well.
Screens (touch or not) are more expensive than physical controls, so unless there are some "big" advantages they wont replace physical controls in the low end.
High quality rotary encoders are very expensive, whereas 7", 10" and 13" Retina screens are commonplace now and will continue to be commonplace.
I'm thinking a screen about the size of an iPad mini would be perfect for a low end scope. And trust me, the extra resolution *does* make a difference, even from a distance.
Says you, I can look over the BOMs here and see that encoders are cheap compared to the screen space needed to reimplement their functions unless you want obnoxious popup menus which hide the displayed waveform. There is a balance of some touch and some physical controls in these products playing to the strengths of each.
Larger high DPI screens add a lot of BOM cost to a product, when you need to increase the processing to drive those pixels the BOM quickly balloons out. You'd either need to find things to remove from the BOM (such as the encoders) that would offset the cost but encoders in bulk are cheap and already have the needed long product lifespan. LCDs are a really awful commodity market where you can't just pick whoever is cheap and popular this month because they might lose a big contract and then discontinue the line leaving you without a supplier for a screen and having to redesign for a new part. Consumer and industrial products are worlds apart when it comes to procurement and system design, I've worked on both and see this stuff all the time.
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
Correct, they are not designed for general use were update rate matters. It's all about signal integrity, analysis, and maximum memory depth at all times. Unlike Tek that like to promote their high update rate using FastAcq no doubt that has pathetic amount of memory. Well ok, nothing wrong with promoting that if that's what they want to do. It's up to the user to understand the difference and their requirements.
No one buys an S Series scope for day to day general lab use.
It sounds like I'm a bit behind the times. More likely, I'm misinterpreting how the newer Agilent/Keysight scopes actually work. Visual dead time, which is what's important ("probability of intercept" in frequency-domain terms) doesn't necessarily have to depend on the trigger rate, which is what Tek is presumably benchmarking for their brochure.
They won't tell you, but I'd bet their figures are only valid using FastAcq mode.
Yes you are correct Dave, the graphs are using FastAcq mode. The data was collected from all the scopes in its class when they're in their peak speed at each horizontal scale. Some are easy to get there, some require lots of work (like the R&S RTO) that you need to turn on dots mode and manually change the RL. The improved FastAcq on the 5 Series MSO brings a variety of enhancements and optimizes the whole system at the press of a button. Technically you could leave it on all the time and get measurements, decodes, etc. and all really fast. FastAcq is designed around low record length, but you can manually change that. As typical horizontal scaling models will typically increase up to a larger RL up to a limit, then will start decreasing the sample rate. Really no dis-advantage being in FastAcq, its just an optimized state of acquisition.
[sarcasm]We all await your fantastic new scope, let us know who you pick for your screen supply contract so we can all access long lived cheap high quality LCD panels.[/sarcasm] Why not skip the embedded system entirely and just build a box which you connect the tablet to.....
https://www.lab-nation.com
Open source and ready for you to build upon.
High quality rotary encoders are very expensive, whereas 7", 10" and 13" Retina screens are commonplace now and will continue to be commonplace.
I'm thinking a screen about the size of an iPad mini would be perfect for a low end scope. And trust me, the extra resolution *does* make a difference, even from a distance.
I agree. After the experience of going from my Rigol DS2072 to the R&S RTB2004 it just seems obvious to me that larger capacitive touch screens are the next step for mid and eventually low end scopes. I knew that moving to a higher resolution 10" screen would be welcomed but was not sure how much the touch screen would be appreciated. Now, I can't imagine using a digital scope without one.
Looking at how the market is now flooded with inexpensive tablets - it seems obvious to me that the screen hardware is not the cost bottleneck. I'll bet it's not long before one of the Chinese manufacturers offers a sub $1000 4 channel traditional form factor scope with a 10" touch screen. (I think Micsig has already done this though in a tablet form factor and apparently with immature software -for now),
As I argued earlier in this thread, I do think that the touch screen does not and should not fully replace all knobs. Tactile feedback is important for quick, precise adjustments. But I have to admit, I'm finding myself using the knobs less than I thought I would on the RTB2004
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
Correct, they are not designed for general use were update rate matters. It's all about signal integrity, analysis, and maximum memory depth at all times. Unlike Tek that like to promote their high update rate using FastAcq no doubt that has pathetic amount of memory. Well ok, nothing wrong with promoting that if that's what they want to do. It's up to the user to understand the difference and their requirements.
No one buys an S Series scope for day to day general lab use.
It sounds like I'm a bit behind the times. More likely, I'm misinterpreting how the newer Agilent/Keysight scopes actually work. Visual dead time, which is what's important ("probability of intercept" in frequency-domain terms) doesn't necessarily have to depend on the trigger rate, which is what Tek is presumably benchmarking for their brochure.
They won't tell you, but I'd bet their figures are only valid using FastAcq mode.
Yes you are correct Dave, the graphs are using FastAcq mode. The data was collected from all the scopes in its class when they're in their peak speed at each horizontal scale. Some are easy to get there, some require lots of work (like the R&S RTO) that you need to turn on dots mode and manually change the RL. The improved FastAcq on the 5 Series MSO brings a variety of enhancements and optimizes the whole system at the press of a button. Technically you could leave it on all the time and get measurements, decodes, etc. and all really fast. FastAcq is designed around low record length, but you can manually change that. As typical horizontal scaling models will typically increase up to a larger RL up to a limit, then will start decreasing the sample rate. Really no dis-advantage being in FastAcq, its just an optimized state of acquisition.
One other note, if some scopes are designed for 'signal integrity', whats the opposite? If you optimize for speed are you willing to give up measurement accuracy?
All Tek scopes perform measurements on waveform data (high signal integrity) and most of them provide FastAcq, from the MDO and higher models allowing >250,000wfm/sec. This allows high signal integrity and fast waveform capture rate. Only scopes i'm aware of with over 250,000 wfms/sec and do measurements on waveform data are from Tek and R&S. Just my 2 cents.
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
One other note, if some scopes are designed for 'signal integrity', whats the opposite? If you optimize for speed are you willing to give up measurement accuracy?
They are not consistently labelled even by one manufacturer so you have to pay attention to the specifications. Some applications like semiconductor characterization and signal integrity have repeating patterns so there is no disadvantage to using equivalent time sampling or having low acquisition rates and long record lengths are not necessary. Higher than 8 bit resolution is common though.
I wasn't aware that "signal analysis scopes" were a category all their own. If that's true, what they (Keysight and LeCroy) are building is a very expensive DSO that can't replace an analog scope.
Correct, they are not designed for general use were update rate matters. It's all about signal integrity, analysis, and maximum memory depth at all times. Unlike Tek that like to promote their high update rate using FastAcq no doubt that has pathetic amount of memory. Well ok, nothing wrong with promoting that if that's what they want to do. It's up to the user to understand the difference and their requirements.
No one buys an S Series scope for day to day general lab use.
It sounds like I'm a bit behind the times. More likely, I'm misinterpreting how the newer Agilent/Keysight scopes actually work. Visual dead time, which is what's important ("probability of intercept" in frequency-domain terms) doesn't necessarily have to depend on the trigger rate, which is what Tek is presumably benchmarking for their brochure.
They won't tell you, but I'd bet their figures are only valid using FastAcq mode.
Yes you are correct Dave, the graphs are using FastAcq mode. The data was collected from all the scopes in its class when they're in their peak speed at each horizontal scale. Some are easy to get there, some require lots of work (like the R&S RTO) that you need to turn on dots mode and manually change the RL. The improved FastAcq on the 5 Series MSO brings a variety of enhancements and optimizes the whole system at the press of a button. Technically you could leave it on all the time and get measurements, decodes, etc. and all really fast. FastAcq is designed around low record length, but you can manually change that. As typical horizontal scaling models will typically increase up to a larger RL up to a limit, then will start decreasing the sample rate. Really no dis-advantage being in FastAcq, its just an optimized state of acquisition.
Hi Brian
Is anything disabled or crippled when using fast acquisition mode such as measurements or memory depth or is this mode exclusively used for glitch detection etc ?
cheers
High quality rotary encoders are very expensive, whereas 7", 10" and 13" Retina screens are commonplace now and will continue to be commonplace.
I'm thinking a screen about the size of an iPad mini would be perfect for a low end scope. And trust me, the extra resolution *does* make a difference, even from a distance.
I agree. After the experience of going from my Rigol DS2072 to the R&S RTB2004 it just seems obvious to me that larger capacitive touch screens are the next step for mid and eventually low end scopes. I knew that moving to a higher resolution 10" screen would be welcomed but was not sure how much the touch screen would be appreciated. Now, I can't imagine using a digital scope without one.
Looking at how the market is now flooded with inexpensive tablets - it seems obvious to me that the screen hardware is not the cost bottleneck. I'll bet it's not long before one of the Chinese manufacturers offers a sub $1000 4 channel traditional form factor scope with a 10" touch screen. (I think Micsig has already done this though in a tablet form factor and apparently with immature software -for now),
As I argued earlier in this thread, I do think that the touch screen does not and should not fully replace all knobs. Tactile feedback is important for quick, precise adjustments. But I have to admit, I'm finding myself using the knobs less than I thought I would on the RTB2004
I also agree - I was expecting to not use the touch functionality much, but there are certain things it makes _much_ quicker. A good example is configuring anything complex such as decoding setup or numeric entry - having a large UI with directly accessible controls and numeric entry is hugely faster than being limited to rotary knobs and controls on the edge of the screen next to soft buttons. Other things I could take or leave - I'm still reaching for the knobs for horizontal or vertical controls (even with the awful detents on the position knobs).
Big screen is also a huge plus - there is a point where extra resolution is wasted, but that point is not at 7-9" 800x480 screens, which seem very poor compared to >10" 720-1080p class kit. Even once you start not being able to see individual pixels, the extra resolution makes UI elements much nicer to look at and allow for more data to be displayed (e.g. lots of digital channels, horizontal zoomed view, stacked channels) without making a complete mess on the screen. I look forward to decent screens moving out of >$5k territory down to hobbyist level kit - even with the added space and cost for a larger screen I'm surprised there aren't more moves into this market already.
One key point though is that everyone is used to really smooth and fast touch UIs from their use of smartphones, tablets etc, and will (somewhat unfairly given the enormous investment in phone UIs) expect similar performance from everything with a touchscreen. Using the touchscreen on my R&S scope seems a little disappointing in certain circumstances (particularly scrolling) because of this, though it's still perfectly usable and is priced way below the Tek. Still hugely worthwhile having the touch functionality though - can always go back to buttons for most functions if needed. Maybe one solution is to fork android and leverage all the existing highly optimised code to built the UI around, allowing the manufacturer to concentrate on all the scope bits rather than the UI (is Micsig already starting to do this? Am pretty sure than some automotive manufacturers are for their in-car entertainment systems).
One key point though is that everyone is used to really smooth and fast touch UIs from their use of smartphones, tablets etc, and will (somewhat unfairly given the enormous investment in phone UIs) expect similar performance from everything with a touchscreen. Using the touchscreen on my R&S scope seems a little disappointing in certain circumstances (particularly scrolling) because of this, though it's still perfectly usable and is priced way below the Tek.
I don't think it's unfair at all. Phones and tablets cost a few hundred dollars to build, at most. They have to run for hours on small batteries. On the other hand, cars are famous for crappy, sluggish infotainment UIs, and they sell for tens of thousands of dollars and have tens or even hundreds of kilowatts of power to play with. Oscilloscope manufacturers sell small AC-powered boxes for more than some of those cars cost.
So these products have fewer excuses for poor UI performance, not more.
Their engineers also don't need to be especially talented at optimization, unlike the iOS and Android architects. They can and should use commodity GUI frameworks, and just throw hardware at them until they perform acceptably.
I am surprised how long proprietary all-in-one test gear has hung on as being viable.
I would have expected the UI's including displays, buttons, lights, etc to be modular from the actual test circuits. We see all-in-one dedicate units and we have blind USB boxes without only a mouse and traditional keyboard inside a PC architecture. Both have limits. It would be nice to see these things setup in modules that can be all-in-one or broken up into pieces where you have the front end close to the DUT with all of it high-end, high-speed architecture, a display at a reasonable distance that can be whatever size the user chooses, and a control surface with the knobs and buttons to control it. In an ideal world - the display and control surface could be setup to function with all sorts of gear from SA's, VNA's, to DMMs. Kinda like a user-friendly version of National Instruments and targeted toward general purpose bench use.
It would put everything you need where you need it and allow an easy upgrade path.
Newer DSOs seem to have worse user interface performance than older ones and this carries over to other devices so I predict that touch screen performance will be horrible. The same applies to desktop operating systems, calculators, and cars and has nothing to do with the hardware performance or power limitations.
I'm quite sure existing products are proving your prediction is wrong.
Newer DSOs seem to have worse user interface performance than older ones and this carries over to other devices so I predict that touch screen performance will be horrible. The same applies to desktop operating systems, calculators, and cars and has nothing to do with the hardware performance or power limitations.
Historically both automakers and T&M vendors have had to be pretty conservative with their computing platforms. Industrial and automotive single-board PCs are designed for longer-term availability, and that necessarily puts them behind the curve at a time when user interfaces are becoming more CPU- and GPU-intensive.
So yes, subjective performance is getting worse, and the manufacturers' excuses aren't
entirely invalid, but still... given the margins in this business, their excuses aren't entirely valid either. 500 million not-so-red Chinese don't give a hoot about 10-year availability guarantees on their SBC platforms, and this is going to catch up to the major vendors big-time.
I'm quite sure existing products are proving your prediction is wrong.
Existing products already confirmed my prediction. Every modern DSO I have tested has had performance (and functional) problems with its user interface despite having massive computing resources available.
Historically both automakers and T&M vendors have had to be pretty conservative with their computing platforms. Industrial and automotive single-board PCs are designed for longer-term availability, and that necessarily puts them behind the curve at a time when user interfaces are becoming more CPU- and GPU-intensive.
My complaint has nothing to do with hardware performance; it is a software design problem. A couple of times I have tracked this down to use of modern languages which rely on non-deterministic garbage collection but there are other causes as well. I think there is an assumption that faster hardware performance will fix it but that has not been the case for decades so why would it change now?
Historically both automakers and T&M vendors have had to be pretty conservative with their computing platforms. Industrial and automotive single-board PCs are designed for longer-term availability, and that necessarily puts them behind the curve at a time when user interfaces are becoming more CPU- and GPU-intensive.
My complaint has nothing to do with hardware performance; it is a software design problem. A couple of times I have tracked this down to use of modern languages which rely on non-deterministic garbage collection but there are other causes as well. I think there is an assumption that faster hardware performance will fix it but that has not been the case for decades so why would it change now?
If the problem is really just sluggish response, and the UI thread(s) have been adequately decoupled from everything else on the CPU, then it seems likely that excessive bean-counting is to blame. Either that, or a higher-performance computing platform that met all of the other requirements simply wasn't available. One of these two cases is
certainly true of the system in my car. That particular manufacturer certainly had the funds available to provide good UI performance. They just didn't.
Imagine my dread the first time I powered up a TDS 694C:
Yet the controls on these scopes are as snappy as a 547.
That's why I'm not inclined to cut manufacturers any breaks 20 years later...
My complaint has nothing to do with hardware performance; it is a software design problem. A couple of times I have tracked this down to use of modern languages which rely on non-deterministic garbage collection but there are other causes as well. I think there is an assumption that faster hardware performance will fix it but that has not been the case for decades so why would it change now?
It doesn't help that as time goes on, software developers unnecessarily pile abstraction on top of abstraction, resulting in call stacks that are many tens of levels deep (sometimes approaching 100 levels!). The amount of code that is executed to do even relatively simple things is now horribly excessive. You know that things have gotten out of hand when you depend on the computer to figure out the relationships in the code for you, because they are too complex and too numerous to figure out "by hand".
It would be one thing if that which is being accomplished were that much more complex than what a simpler architecture could accomplish, but from what I've seen in the software world (which is the world I live in), with
very few exceptions, it just isn't.
It seems that, instead of powerful tools being used in order to better handle previously existing complexity, the complexity is instead later scaled to match the power of the tools that are used, just because it can be (and, more importantly, because it's erroneously perceived as being easier and cheaper than intentionally designing and building minimal complexity -- a false tradeoff if there ever was one).
Newer DSOs seem to have worse user interface performance than older ones and this carries over to other devices so I predict that touch screen performance will be horrible. The same applies to desktop operating systems, calculators, and cars and has nothing to do with the hardware performance or power limitations.
I like my 3000 T and I like the touch screen on it. It's fast and responsive.
What I like about Dave's reviews is that his reviews are from the perspective of a user not only what's written on the data sheet. Today it's less about the hardware and more about the usability & intuitiveness that ultimately make or break products.
Newer DSOs seem to have worse user interface performance than older ones and this carries over to other devices so I predict that touch screen performance will be horrible. The same applies to desktop operating systems, calculators, and cars and has nothing to do with the hardware performance or power limitations.
I like my 3000 T and I like the touch screen on it. It's fast and responsive.
What I like about Dave's reviews is that his reviews are from the perspective of a user not only what's written on the data sheet. Today it's less about the hardware and more about the usability & intuitiveness that ultimately make or break products.
I find theres a small but tangible increase in sluggishness going from the vxWorks based Agilent scopes (where there's never any sluggishness) to those based on Windows CE. It's a key reason why I still use an MSO7000 is the first one to be powered up on the bench rather than an MSOX3000. I should note that both UIs are still very snappy compared to some I could mention.
My complaint has nothing to do with hardware performance; it is a software design problem. A couple of times I have tracked this down to use of modern languages which rely on non-deterministic garbage collection but there are other causes as well. I think there is an assumption that faster hardware performance will fix it but that has not been the case for decades so why would it change now?
It doesn't help that as time goes on, software developers unnecessarily pile abstraction on top of abstraction, resulting in call stacks that are many tens of levels deep (sometimes approaching 100 levels!). The amount of code that is executed to do even relatively simple things is now horribly excessive.
^ All of these.
Other reasons for non-deterministic behaviours are JIT compilation and unnecessary decoupling (e.g. going through the network stack on a standalone unit).