The USB bottleneck myth debunked

Post any questions you may have about our current range of oscilloscopes
Post Reply
alan
Site Admin
Site Admin
Posts: 195
Joined: Thu May 25, 2006 8:02 am

The USB bottleneck myth debunked

Post by alan »

PicoScope oscilloscopes are well known for offering far deeper memories than competing benchtop or PC oscilloscopes. In fact for most of the last 20 years different generations of PicoScopes have offered the deepest memories available on any oscilloscope at any price.

Due to the fact our oscilloscopes offer deep memory and connect to the PC via USB we get quite a few questions about whether the USB link is a bottleneck restricting performance. Typically the questions are similar to the ones below:

1. I am trying to choose between a PicoScope and a traditional benchtop scope. Whilst the PicoScope has much more memory than the benchtop, the sales person for the benchtop scope manufacturer told me that you can not really use the deep memory because the USB connection to the PC causes a bottleneck that slows the screen update rate down. Is this true?

2. I have been researching which oscilloscope to buy and had decided on a PicoScope 3000. Someone knowledgeable on an electronics forum however told me that as soon as you enable deep memory on a PC oscilloscope it grounds to a halt and the controls become unresponsive. Do I need to reconsider my choice as I am keen to use deep memory for serial decoding?

Rather than keep answering such questions individually this post is an attempt at a generic answer.

Some oscilloscopes (both benchtop and PC based) DO have poor waveform update rates and unresponsive controls making their deep memory of limited use / something only to enable when essential. This however is not the case with PicoScope products - you can enable deep memory all the time without penalty.

One way of showing that the USB link is not a bottleneck is to benchmark a PicoScope against a traditional benchtop scope.

Benchtop oscilloscope. Tektronix DPO5054B
4 channels
500 MHz bandwidth
5 GS/s sampling rate
25 M point memory
Price $15,300

PC oscilloscope. PicoScope 6404C
4 channels
500 MHz bandwidth
5 GS/s sampling rate
1 G point memory
Price $6,600

At 500us/div both oscilloscopes sample at 5GS/s and collect 25 million samples per screen update. The measured screen update rates are as follows:

Tektronix DPO5054B 9 waveforms in 10 seconds
PicoScope 6404C 180 waveforms in 10 seconds

The Tek scope is collecting and displaying 22.5 million samples per second whilst the PicoScope is collecting and displaying 450 million samples per second. Of course the PicoScope can capture far more than 25 million samples. For example the PicoScope 6404D can sample at 5GS/s for timebases as long as 20ms/div. It is collecting 1 BILLION samples per waveform / screen update. Under these conditions it can still update the screen 2.5 times per second (bear in mind 20ms/div is 200ms across the screen so the scope is collecting data for 50% of the time and processing /displaying it for 50% of the time).

In the above example the PicoScope 6404D is collecting and displaying 2.5 BILLION samples per second. To put it another way it is collecting and displaying a DVDs worth of data every 2 seconds. We believe this capture and display is faster than any other oscilloscope on the market today.

Compare the 2.5 GB/s data rate with the theoretical maximum available USB bandwidth.

USB2.0 60 MB/s (480 Mbit/s)
USB3.0 625 MB/s (5 Gbit/s)

In other words the oscilloscope is collecting and displaying data far faster than its USB 3.0 port can transfer data. The reason behind this is simple - all PicoScopes with over 1 MS of memory use dedicated hardware inside the oscilloscope to intelligently form the image to be displayed on the PC screen without any loss of high frequency data. By only transferring the image to be displayed across the USB link any concerns about either a USB or PC CPU bottleneck are eliminated.

With each new generation of PicoScope we have improved our hardware acceleration engines to the point where they can process samples at least as fast as the oscilloscope can collect the data (5 billion samples per second in the case of the PicoScope 6000s). This is approx 2 orders of magnitude faster than the samples can be processed in a PC CPU.

As well as accelerating deep memory captures the hardware acceleration engine is also responsible for the following abilities:

To collect a burst of waveforms at speeds of up to 1 million waveforms per second
To continuously collect and display 10000s of waveforms per second to help find intermittent glitches. (For more on finding intermittent glitches see also http://www.picotech.com/support/topic13657.html )
Segment the memory into a circular buffer to automatically store 100s or 1000s of previous waveforms.

In short the USB bottleneck is a myth, PicoScope oscilloscopes cope better with deep memory captures than competing oscilloscopes be they PC based or benchtop.

blachanc
Newbie
Posts: 1
Joined: Fri Feb 01, 2013 2:49 pm

Re: The USB bottleneck myth debunked

Post by blachanc »

Hi Alan,

Thank you for your post. It does clarify something that was not clear in my mind.

You picked a very high end scope for your analysis.
Would the story differ in any way if you would compare a lower end scope?
Ex: 3206MSO
I guess the answer is no, but would like to get a formal answer if possible.


Thank you,

Ben

alan
Site Admin
Site Admin
Posts: 195
Joined: Thu May 25, 2006 8:02 am

Re: The USB bottleneck myth debunked

Post by alan »

Hi,

The answer is the same with all the deep memory PicoScope - there is no USB bottleneck with the 3206MSO. In this case the max sampling rate is 500 MS/s rather than 5 GS/s.

For example at 10 ms/div timebase you can collect 50 million samples per waveform. The hardware acceleration engine is always at least as fast as the max sampling rate so it will process these 50 million samples in less than 100ms. As such you should get around 5 updates per second.

As above we are talking here about waveform update rates with very deep memory / longer timebases. At short timebases you can collect 1000s of waveform per second with color / intensity grading.

Regards
Alan

blachanc
Newbie
Posts: 1
Joined: Fri Feb 01, 2013 2:49 pm

Re: The USB bottleneck myth debunked

Post by blachanc »

Hi Alan,

Thank you for your detailed explanation!

Benoit

EPAIII
Newbie
Posts: 0
Joined: Thu May 08, 2014 5:27 pm

Re: The USB bottleneck myth debunked

Post by EPAIII »

I am not sure that I completely understand this. Basically, the USB port is too slow to allow all the SAMPLES to be passed to the computer as actual samples. So instead of storing the samples, you are storing the shape of the displayed waveform. And this is done without loss of the fine (high frequency) details of that waveform.

So, if I understand properly, you are storing a compressed version of the actual data, not the actual data. Should I assume that this waveform shape that is stored in the computer is actually stored using the current display settings? That would mean that any fine, high frequency information that is not shown by the current display settings would not be stored. So, if the user were to retrieve a stored waveform and then magnify it in the time domain then there would be little or no further information that is available in the fine details of that waveform. It would just become "blurred" as it is magnified.

This would limit the usefulness of the stored information. But, of course, it would not make it completely useless as you could examine it for any events that wold have been visible at the current display settings.

alan
Site Admin
Site Admin
Posts: 195
Joined: Thu May 25, 2006 8:02 am

Re: The USB bottleneck myth debunked

Post by alan »

Hi EPAIII,

The raw data / samples are always stored in the oscilloscope. If we sample at 5 GS/s with a timebase of 2ms/div (20 ms total) then 100 million samples will be stored in the scope. As described above the scope uses the hardware acceleration to quickly form and transfer the image to be displayed to the PC.

If you now want to zoom on a small section of the waveform - say just 1ms out of the middle of the waveform then the new zoomed waveform (formed from 5 million samples) is again created in the oscilloscope and sent back to the PC. The raw data is kept in the oscilloscope - you can "un-zoom" to see all 100 million samples or zoom right in to see just say 100 samples on the screen. This re-processing of the data as you select areas to zoom is so fast that the data is displayed as soon as you release the mouse button.

I hope that makes sense / answers your question - if not just let me know.

Regards
Alan

EPAIII
Newbie
Posts: 0
Joined: Thu May 08, 2014 5:27 pm

Re: The USB bottleneck myth debunked

Post by EPAIII »

OK, I got it. Actual, raw data is in the scope. Only the shape of the displayed waveform is sent to the computer. This avoids the "bottle neck". Simple and elegant. And, with a gig of data storage in the scope, you have a fairly good solution.

I, and probably many others have labored under the impression that the raw data was sent to the computer. Perhaps some other brands do this, I don't know.

Eva zhou
Newbie
Posts: 0
Joined: Mon Jul 21, 2014 8:29 am

Re: The USB bottleneck myth debunked

Post by Eva zhou »

Hi alan,

As described above the scope uses the hardware acceleration to quickly form and transfer the image to be displayed to the PC. Can we use the hardware accerleration with deep memory PicoScope when writing our own application? if so, which function can we use in SDK like 5244? is it ps5000aGetValue in downsampling mode?

Thanks
Eva

Martyn
Site Admin
Site Admin
Posts: 4491
Joined: Fri Jun 10, 2011 8:15 am
Location: St. Neots

Re: The USB bottleneck myth debunked

Post by Martyn »

Yes, you pull out just the data you need for the display using ps5000aGetValue, you can use downsampling as required although for very high zoom this would not be necessary.
Martyn
Technical Support Manager

Pontisteve
Newbie
Posts: 0
Joined: Wed Mar 30, 2016 1:00 am

Re: The USB bottleneck myth debunked

Post by Pontisteve »

So is USB 2.0 a bottleneck then, to my Pico 4423?

Martyn
Site Admin
Site Admin
Posts: 4491
Joined: Fri Jun 10, 2011 8:15 am
Location: St. Neots

Re: The USB bottleneck myth debunked

Post by Martyn »

No, the USB bus is not a problem, are there any tests you have run that makes you think it would be ?
Martyn
Technical Support Manager

Mmm22
User
User
Posts: 2
Joined: Mon Jun 15, 2020 9:21 am

Re: The USB bottleneck myth debunked

Post by Mmm22 »

It is ok to downsample if only the screen representation is needed. But what if I need to do some measurements; is only the downsampled data used for the calculation?

oasis2016
User
User
Posts: 2
Joined: Fri Jan 06, 2017 4:06 am

Re: The USB bottleneck myth debunked

Post by oasis2016 »

But, why hasn't my question been answered satisfactorily yet?

topic27841.html?&p=94071#p94071

No offense.
I think, since picoscope's waveform refresh rate is so high, why can't it achieve the pattern I want to see?
Can the picoscope 6000 series meet my needs?
Because I think that the pattern similar to the analog oscilloscope can be displayed normally, which proves that the waveform refresh speed of the oscilloscope is fast.

Best regards!

Martyn
Site Admin
Site Admin
Posts: 4491
Joined: Fri Jun 10, 2011 8:15 am
Location: St. Neots

Re: The USB bottleneck myth debunked

Post by Martyn »

Please see the original thread.
Martyn
Technical Support Manager

oasis2016
User
User
Posts: 2
Joined: Fri Jan 06, 2017 4:06 am

Re: The USB bottleneck myth debunked

Post by oasis2016 »

:D
Thank you Martyn!
I have read your reply.

Post Reply