Choosing an Oscilloscope in 2026

TL;DR: Decide your signal frequency first, channel count second, and resolution last. 100 MHz is comfortable for MCU work, switching-supply ripple, and I²C/SPI/CAN debugging. The headline "1 GSa/s" sample rate drops to 500 MSa/s when all four channels are on, and that caveat appears nowhere on the retail listing. Models with a built-in 25 MHz signal generator save you a separate function generator purchase. Some links below are affiliate links and do not change my conclusions.

A signal that changes over time cannot be judged from a single multimeter reading. I learned this the hard way while chasing ripple on a power board: the digits on the multimeter would not settle, so I put the probe across the output capacitor on the oscilloscope and the spikes and the switching frequency were obvious immediately. An oscilloscope is not a replacement for a multimeter, it covers the half of the job a multimeter structurally cannot do. Every number below comes from a vendor page or an official datasheet unless I say otherwise. Prices and stock status are as of publication.

Which kind of signal are you measuring

Three signal families lead to three different conclusions.

MCU digital side: I²C, SPI, UART, CAN. These protocols typically run from a few MHz to a few tens of MHz, so 100 MHz of bandwidth leaves generous headroom.

Switching supplies: switching frequency ranges from 50 kHz to 500 kHz, but what you actually care about is overshoot and ringing on the switching edge. The few hundred nanoseconds of rise time is the part that matters, so the sample rate has to keep up.

Sensors and analog front ends: thermocouples, strain gauges, photodiodes. These signals are in the millivolt range, and the real bottleneck is the noise floor and vertical sensitivity, not bandwidth.

If your work sits entirely in the first two categories, 100 MHz is the sweet spot, which is why all four scopes here share that ceiling.

Checking the four models against official specs

I pulled the spec page for each instrument, and one pattern showed up repeatedly: the sample rate on a spec sheet is almost always the single-channel number.

Siglent SDS1104X-E: four channels is its real limit

The grey 4-channel model on Amazon ships with one standard passive probe. Siglent's own page describes it as 100 MHz bandwidth, 4 channels, 1 GSa/s real-time sampling (across two ADCs), 14 Mpts of memory, a minimum vertical input range of 500 µV/div, a waveform capture rate of 100,000 frames/sec in standard mode (400,000 in sequence mode), a 7-inch display, and an 8-bit ADC.

The qualifying condition is stated just as clearly on the same page: when all channels are enabled, each channel drops to 500 MSa/s and a standard record length of 7 Mpts. You only get back to 1 GSa/s and 14 Mpts when a single channel per ADC is active.

So what does that mean in practice? At four channels the sample rate is exactly 5x the 100 MHz bandwidth, which is 5x oversampling. That is enough for signals with gentle edges, but if your goal is to see ringing on a 3 ns rising edge, single-channel mode is the only way to get the full detail. That is the first real limit of this scope.

The second limit is in the probe. The standard probe Siglent lists is a 100 MHz 1x/10x selectable model, rated at 6 MHz in the X1 position and 100 MHz in the X10 position, with maximum input of 300 V / 600 V depending on attenuation. Set the probe to 1x and the effective bandwidth of the instrument collapses to 6 MHz, regardless of the 100 MHz you paid for. This is a shared trap across every entry-level scope, and it is the thing I most want to warn you about.

The official accessory list offers an optional 16-channel logic probe and a 25 MHz AWG module, both sold separately.

Siglent SDS814X HD: resolution pushed to 12 bits

This is the only one of the four that states a 12-bit vertical resolution, and the four-channel 100 MHz standard version is listed on Amazon. The official noise floor is 70 µVrms at 200 MHz (typical), which against the 8-bit ADC above means roughly 16x lower quantisation noise. That difference is visible when you measure power rail ripple or small-amplitude signals.

The sampling and memory ladder is explicit too: 2 GSa/s and 100 Mpts on one channel, 1 GSa/s and 50 Mpts on two, 500 MSa/s and 25 Mpts on four. Waveform capture rate is 120,000 wfm/s in normal mode and 500,000 wfm/s in sequence mode. The display is a 7-inch capacitive touchscreen at 1024×600.

One thing deserves its own mention: serial bus trigger and decode is standard, requiring no licence, and covers I²C, SPI, UART/RS232, CAN, CAN FD, LIN, FlexRay and I²S. Power Analysis, by contrast, is an optional software licence listed at €239.00 in the official store. The Bode plot function is built in and needs no purchase.

The connectivity on this model is noticeably more complete: two USB 2.0 Host ports, USB 2.0 Device, 10/100M LAN, and an auxiliary output for trigger and pass/fail. If your bench needs to feed data into an automated flow, LAN is the dividing line between workable and awkward. One passive probe per channel is included, and it ships with a calibration certificate.

Hantek DSO2D15: 150 MHz bandwidth, connectivity is the weakness

Hantek's official page positions the DSO2D15 as a value option, and the Amazon listing is the 2-channel version with signal source: 150 MHz analog bandwidth, 1 GSa/s sampling, 8M memory depth, 2 channels, 8-bit vertical resolution, plus a built-in 1-channel 25 MHz waveform generator supporting arbitrary waveform output and modulated waveform output.

150 MHz is the highest of the four. It gives you more margin against high-frequency switching supplies or fast edges. There are 14 trigger modes, 32 automatic measurements with statistics, standard decode and protocol analysis for UART, LIN, CAN, SPI and I²C, and both a 3-digit digital voltmeter and a 6-digit hardware frequency counter. That feature density is unusual at this price.

The interface situation is where this one gives ground: USB Host and USB Device only, with no Ethernet. That is stated on the official spec sheet, not inferred. The trade is explicit: the price advantage from omitting the network module comes at the cost of not being able to plug into a remote acquisition system. If you intend to record waveforms long-term for regression comparison, know this before you buy.

One spec discrepancy is worth resolving. Some distributors list memory depth as "4 Mpts/ch", which looks like it contradicts the official "8M". The official manual gives the precise answer: 8M maximum on a single channel, 4M when two channels are used. So it is one figure that switches with channel count, not two conflicting numbers.

The official manual also states that bandwidth is limited to 6 MHz when using a 1x probe, identical to the two scopes above. Input impedance is 1 MΩ ±2% in parallel with 13 pF ±3pF, vertical sensitivity runs from 2 mV/div to 10 V/div, vertical gain accuracy is ±3% from 10 V/div to 10 mV/div and ±4% from 5 mV/div to 2 mV/div. The unit measures 318×110×150 mm, weighs about 1900 g, and draws under 15 W.

FNIRSI 1014D and 2C53T: three instruments in one

The 1014D is listed as a 2-in-1 dual-channel 100 MHz, 1 GS/s oscilloscope with signal generator. The 2C53T is a 3-in-1 with a multimeter and DDS signal generator at 50 MHz bandwidth, carrying a Best Seller badge in Lab Oscilloscopes.

Their appeal is integration and price. As for vertical noise, trigger jitter, and long-term calibration stability, the vendor publishes nothing comparable to the figures above. I have not tested these hands-on, so I will not judge them. If you care about long-term reliability and documentation, these are not the first choice. If you care about covering several instrument needs with a single purchase, they are worth a look.

One caveat: as of publication, amazon.com returns prices in different currencies depending on the access region, so I cannot give you a trustworthy dollar figure. Check the official store page directly.

How I would actually choose

The order I use goes like this.

First question: how many signals do I need to see at once? Two or fewer, and the 2-channel Hantek DSO2D15 is enough, with a 25 MHz signal generator included. Three or more, go straight to a four-channel model rather than discovering the shortfall after purchase.

Second question: what is the smallest amplitude I need to measure? At the 1 mV level, the 12-bit converter and 70 µVrms noise floor of the Siglent SDS814X HD make a real difference. Above tens of millivolts, the 500 µV/div sensitivity of an 8-bit model is enough.

Third question: does this signal need long-term recording or remote capture? If so, the LAN port on the SDS814X HD is the only one of these four that satisfies it directly. The other three route everything through USB.

Fourth question is bandwidth. By this point, if your signal is still under tens of MHz, 100 MHz will not be the bottleneck on any of them.

For how the tools fit together at the beginner stage, I wrote earlier about how to read multimeter CAT ratings and fuse voltage limits, which covers static level measurement and complements the time-domain view here. If your project starts from a development board, the single-board computer comparison can help you work out which signals you need to observe at all.

Real defects, model by model

Reviews that list only strengths are a genre problem, so here are the drawbacks in one place.

The SDS1104X-E halves both sample rate and memory depth with all four channels enabled, which is a consequence of the hardware architecture and not adjustable. It is an 8-bit ADC, so quantisation steps are visible on small ripple measurements. Only one probe is included, so a second channel needs another purchase.

On the SDS814X HD, the most expensive line item is licensing. Power Analysis is a separate €239.00 software licence, and while the Bode function is built in, it depends on an SDG X or SAG1021 function generator, which is another expense. The 16-channel MSO logic probe is likewise an option.

The Hantek DSO2D15 has no Ethernet port, which is a genuine missing feature rather than a preference. It has only 2 channels, and memory depth drops to 4M with both in use. The 7-inch 800×480 screen is less comfortable for long reading sessions than a touchscreen. Its advertised 150 MHz bandwidth only holds with a 10x probe.

Neither FNIRSI model publishes vertical noise or trigger jitter figures. For any precision-critical scenario, that is a documented unknown.

The trap all four share: the 1x probe position limits bandwidth to 6 MHz. Hantek's manual says so explicitly, and Siglent's site labels the probe 6 MHz (X1) / 100 MHz (X10). The first thing to do with a new scope is set the probe to 10x, then confirm the channel attenuation indicator on screen reads 10X.

Troubleshooting

The top of my waveform is flat. Is the scope broken?

No. The input is exceeding the current vertical range and getting clipped. Increase V/div by one step and look again. If your signal is a square wave, a flat top means you are slicing the signal itself and your range is too small.

The trace is covered in noise spikes. Bad probe or bad scope?

Switch to 10x first to rule out probe attenuation, since the 1x position only gives you 6 MHz of bandwidth and visibly amplifies high-frequency noise. Then check that the ground clip is genuinely connected to the circuit under test rather than left floating. In the 10x position the probe's own input capacitance and lead inductance resonate with the test point, and routing the probe lead away from high-frequency traces improves this noticeably.

Serial bus decoding keeps showing errors even though the waveform looks correct

Usually the trigger point is not landing on the start of a protocol frame. Manually change the trigger type to the Frame or DATA mode for that protocol and set the level to the midpoint between the bus high and low levels. Decoding reliability improves noticeably when you use Single mode triggered on the clock rising edge.

The computer does not see the device over USB

Some models need the vendor virtual serial port driver installed before they enumerate as a USB device; Hantek's SCPI control depends on exactly this. Also confirm you have a data cable rather than a charge-only cable, because the two are frequently mixed in the box.

With four channels on, the sample rate reads 500 MSa/s. Did something break?

No, this is normal behaviour for four-channel scopes and I explained the reason in the official spec check above. Reaching 1 GSa/s requires reducing the number of simultaneously active channels.

Common questions

Can 100 MHz of bandwidth measure current-generation microcontrollers?

In the overwhelming majority of cases, yes. Mainstream MCU clocks run from tens to hundreds of MHz, but what you normally examine is not the clock itself: it is setup and hold timing on the rising edge, the shape of the SCL line on I²C, or supply ripple. Those frequency components sit in the few-MHz range. The only situation demanding more bandwidth is observing GHz-level clocks or high-speed serial links directly, which is beyond all four models here.

Should I buy a scope with a built-in signal generator?

Models with an integrated generator, such as the Hantek DSO2D15 and the FNIRSI 1014D, let you run loop response tests and power-up sequencing checks without buying a separate function generator. If you already own one, this feature is duplicated spend.

How much better is 12-bit resolution in practice?

Twelve bits reduces quantisation noise by roughly 16x against eight bits. On ripple and small-sensor-signal measurements you can directly see detail that quantisation steps hide on an 8-bit machine. On large-amplitude signals the gap is close to invisible. Choose based on need rather than assuming higher is always better.

Can I just compare prices and ignore the specs?

I would not. Among desktop instruments, oscilloscopes are the most prone to headline numbers that only hold under conditions. Sample rate, bandwidth and memory depth can all be simultaneously true only in a specific mode, and the 1 GSa/s figure stops being true the moment four channels are on. Reading the qualifying conditions on the vendor spec page tells you more than any review conclusion.

If you are just setting up a bench, the single-board computer comparison is a reasonable starting point for working out which signals you actually need to watch. I verified all four product pages were live at the time of writing. Prices and stock are as of publication, so treat the official pages as authoritative.

👉 Join Xiaomi MiMo Platform: Leading AI model platform with cost-effective inference

👉 Join Aliyun AI: Top AI products with exclusive coupons for business innovation

📌 This article was AI-assisted generated and human-reviewed | TechPassive — An AI-driven content testing site focused on real tool reviews