ASUS ROG Swift PG27UCDM
Specifications
| Brand: | ASUS |
|---|---|
| Model: | ROG Swift PG27UCDM |
| Size: | 26.5" |
| Resolution: | 3840x2160 |
| Panel type: | QD-OLED |
| Max. refresh rate: | 240 Hz |
| Panel: | ? (Samsung) |
| Backlight type: | n/a |
| Price (approx.): | USD 1200 |
| Monitor release date: | 2025-01 |
| This review's date: | 2025-12 |
Other Reviews
TFTcentral review of the ASUS ROG Swift PG27UCDM.
RTings.com review of the ASUS ROG Swift PG27UCDM.
At a glance
This is a 4K QD-OLED monitor capable of 240 Hz refresh rate at 10 bpc without signal compression (DSC), albeit only via DisplayPort. Note that 4K doubles the resolution of 2K only in terms of the overall number of pixels, not in terms of pixel density along X and Y. Still, this results in a pixel density of 166 ppi, which makes the color fringing artifacts caused by the triangular sub-pixel layout of a QD-OLED type of monitor, which were somewhat disturbing on a 27"/2K display, basically disappear (which which is totally subjective assessment, of course).
OLED technology still comes with the risk of burn-in, which means that too bright content presented for too long can leave permanent traces. To reduce this risk, the monitor will only be tested and later used in the SDR mode (Standard Dynamic luminance Range) and with the "Uniform Brightness" setting enabled, even though the monitor supports HDR (High Dynamic luminance Range). Moreover, we intend to use the monitor only at max. 100 cd/m2 and without BFI (Black Frame Insertion), all of which should help extending the monitor's usable lifetime. Nevertheless, BFI is still a very welcome option for improving motion clarity. The monitor supports BFI for input signals with 120 Hz, which is where the 240 Hz maximum refresh frequency is put to a use without requiring the PC to actually drive this high frequency. The BFI issues reported for the ASUS PG27AQDP (that are still waiting to be fixed by ASUS) have not been observed in the ASUS PG27UCDM.
In the previous reviews of other OLED monitors, a rather high noise level in the color processing was observed. Moreover, the QD-OLED monitor MSI MPG 271QRX exhibited low repeatability in those measurements. At least in part, this might reflect poor firmware implementation, so there was hope that a different manufacturer, in this case ASUS, would have done a better job. Unfortunately, this was not the case, which speaks for a more general problem with OLED panels. All of the tested panels, which is two QD-OLEDs and one W-OLED, have high color processing noise, and both QD-OLED monitors (one from MSI, one from ASUS) exhibit exceptionally low repeatability. Although this is not a deal-breaker, it currently makes W-OLED the preferred technology over QD-OLED.
That being said, regarding color processing noise, QD-OLED might have a slight edge over W-OLED when it comes to presenting gray-scale content, because QD-OLED has to use all its three sub-pixels (RGB) for presenting grays, whereas W-OLED uses only its white sub-pixels. This means that, with W-OLED, there is no averaging between sub-pixels going on, which would otherwise reduce - statistically - the effect of color processing noise. Obviously, presenting gray-scale content with only white sub-pixels has the advantage of removing color fringing completely, which might be the more important property anyway.
A QD-OLED monitor might be the wrong choice if it will be used in bright environments, because it tends to back-scatter light more strongly than W-OLED monitors, thereby reducing the effective maximal contrast. Moreover, external light is not back-scattered neutrally but with a purple tint. However, in only dimly lit rooms this not an issue.
Quirks
- The monitor arrived with firmware version MCM103, which was updated almost immediately to version MCM105 for this review. However, the update process was an ordeal, because updating via an USB stick did not work. The monitor just was not able to read the file from the stick, even though several different sticks were tried. The alternative update method, which requires to run a program on a Windows PC that is connected to the monitor via a USB cable, did work without further problems.
- It is currently unknown how to make the monitor's service menu available.
- DisplayPort 1.4 with UHBR20 (80 Gbps) should be good enough for driving 4K at 120 Hz with 8 bpc uncompressed (i.e., without DSC). However, out of the box, Windows would always switch to a lower spatial resolution when 120 Hz was selected. This could be fixed by overwriting the timing parameters with the CVT-RBv2 parameters (replacing the CVT-RB parameters in the EDID). See also this Online calculator.
OLED care
OLED monitors are potentially susceptible to burn-in, meaning that static image content can alter pixel performance permanently when presented for a prolonged period of time. Manufacturers implement a number of counter-measures, but, luckily – in the case of the ASUS PG27UCDM –, all of these can be turned off in the on-screen menu. Another counter-measure is to not present bright image content, where "bright" is relative to the maximum luminance in HDR mode, which is around 1'000 cd/m2 for this monitor. Time will tell whether burn-in is still an issue when operating the monitor under more moderate conditions.
One of the counter-measures implemented by ASUS is called "Screen move". If activated, the entire screen content will be shifted from time to time. Besides "Off", there are three "Screen move" options available: "Light", "Middle", and "Strong". These levels differ in how often the content is moved (from once per 3 minutes to twice per minute), while the content is moved by one pixel at a time, mostly in diagonal directions within a range of 14x14 pixel positions (see Figure 1). The motion pattern has some weird randomness to it.
Motion blur reduction
Motion blur reduction, which is called ELMB in the monitor settings (Extreme Low Motion Blur), is available only for 120 and 240 Hz refresh rates and is implemented through BFI (Black Frame Insertion). The BFI on/off ratio is fixed to 50:50 for both refresh rates. Neither VRR (Variable Refresh Rate) nor HDR (High Dynamic Range) are available while ELMB is active. Unfortunately, the monitor does not synchronize all that well to the input signal when ELMB mode is active. It is not that there are visible tearing artifacts, but the monitor skips or repeats frames for adapting its internal refresh frequency to that of the input signal received from the PC. Moreover – and possibly more severely – the latency of the displayed image with respect to the input signal varies over time. The difference between the shortest and the longest possible latency for, say, the top-leftmost pixel, is one input frame cycle. How often frames are skipped or repeated and how fast the input lag is changing over time depends on how well the monitor's internal refresh rate happens to match the input refresh rate. The respective cycle time can easily be in the order of minutes, i.e., rather long.
By the way, at least with firmware version MCM102, ELMB gets automatically deactivated when (re-)booting the PC. This is likely due to the refresh rate falling temporarily back to 60 Hz during the boot process, thereby causing ELMB, which is not available for 60 Hz, to be deactivated.
Saturation, White
When it comes to color accuracy, one potential problem lies in the interactions between the primary color channels (red, green, blue) when rendering arbitrary colors, including white. Specifically for white, if everything was perfectly accurate, the luminances of the primary colors would add up to LW=LR+LG+LB(=LRGB for short). Ignore, for a moment, that white has its own sub-pixel in this monitor. Deviations from this perfect relationship can be quantified by the (normalized) error Δe=(LRGB−LW) / LRGB (see Figure 2). We can interpret this error also as a saturation error, a de-saturation coefficient, or a cross-talk coefficient. Note that, normally, the according measurements are also sensitive to de-saturation effects caused by the residual background illumination, which becomes overly dominant for dark shades. Besides dark shades being more difficult to measure accurately, this is why dark shades have been excluded from the graphs shown in Figure 2. Obviously, residual background illumination is not an issue with OLED monitors, because they don't have a backlight that could cause residual background illumination. Moreover, because this monitor has a WOLED panel, that is, an extra sub-pixel for white, this test does not measure what it was designed to measure in the first place, namely the interaction effects between the primary color channels when they are mixed to reproduce white. Still, a big Δe would indicate some sort of color accuracy issue, no matter where the big Δe comes from. The effects would be rather subtle though. For example, unlike with non-WOLED monitors, a big Δe would not mean that all colors appear washed out; it would rather mean that there are some color space distortions somewhere "deeper inside" the color gamut (loosely speaking).
Normally, the Δe presented here do somewhat correlate with the more familiar dE color accuracy value known from other review websites. And, indeed, the dE2000-value for this monitor is, with dE2000 = 0.641 (averaged over 400 colors), rather high as compared to 0.152 (Razer), 0.182 (MSI), and 0.428 (BenQ). Note that the white sub-pixel is not only making things more complicated on the monitor side but also on the color measurement side. Colorimeters often can be calibrated for the specific color spectra of the monitor at hand, which works best if there are only 3 color channels to be taken into account. Accounting for a fourth color channel, as for white here, requires compromises to be made for the colorimeter calibration. It is not clear yet what quantitative impact this might have on the dE2000 measurements though.
Anyway, assuming that these measurements indeed indicate that QD-OLED monitors are better than WOLED monitors in terms of color accuracy, this would be in line with the widespread opinion that QD-LED monitors provide better colors than WOLED monitors. However, "better colors" might not necessarily mean "better color accuracy" but, rather, more saturated colors. Indeed, QD-OLED monitors might show stronger color saturation, especially when it comes to red, but this does not mean that color saturation in WOLED monitors is in any way bad.
Color processing noise
Ideally, the monitor processes incoming pixel values so that the according output luminance follows a smooth transfer function. This processing usually takes place in the digital domain and aims for some favorable Gamma characteristic while taking other parameters into account, like the Contrast setting, the RGB channel gains, and also spatial or spatio-temporal dithering. One way of quantifying how consistently this is done across the pixel value range, without needing to know the intended Gamma characteristic, is to measure deviations from a smooth transfer function. This means that we do not focus on how well the transfer function is described by a simple Gamma function but how well the measured data points can be described by a reasonable smooth transfer function. We only measure the green channel here, because it is the brightest of the color channels normally available and, thereby, provides the best signal-to-noise ratio. Of course, here, where we have also a sub-pixel for white, we could also have measured the transfer function for white. The lowest (x<16) and the highest pixel values (x>240) are not taken into account here for several technical reasons related to the measurement method and data analysis.
Figure 3 shows the results for the ASUS PG27UCDM and, for comparison, the BenQ XL2540. For this comparison, the ASUS was operated at 8 bpc in order to match the capabilities of the BenQ which accepts only 8 bpc inputs. Clearly, the ASUS is doing much worse in this test than the BenQ (SD=22.3% vs. SD=6.3%; smaller standard deviations are better).
This measure should closely correspond to the Gradient score given in the RTings.com review, which is 9.8 of the possible 10.0. Obviously, this is not at all in agreement with our findings. It is probably because RTings.com scores the gradient subjectively and with a rather poor score resolution. The vast majority of the monitors tested by RTings.com are scoring between 9.5 and 9.9, which is only 5 steps for basically the entire relevant range.
The poor color processing also shows at the very low end of the gamma curve, as shown in Figure 4, which compares the gamma curve of the BenQ (for the green channel) to the gamma curve for the ASUS (for the white channel). For this comparison, both monitors were adjusted to reach approximately max. 150 cd/m2 for the respective channels (i.e., green for the BenQ, white of the ASUS). Note that, although the low OLED dark-gray luminances push colorimeters to their sensitivity limits, the notably large luminance steps shown in Figure 4 (e.g., at pixel values 15 and 23) are real and were also quite obvious when just looking at the monitor with the naked eye.
Although the ASUS, being an HDR-capable monitor, allows input signals with 10 bpc color depth, this does not make a difference for this test (Figure 5). If the results were different at all, we would expect the 10 bpc results to be actually worse than the 8 bpc results, as explained in more detail in the Razer Raptor 27 165Hz review. For the ASUS, however, the 8 bpc and 10 bpc results are absolutely identical (apart from measurement noise).
Settling behavior
The following measurements were made with a photo diode PDA36A (Thorlabs), the gain of which was set to 60 dB resulting in a bandwidth of 37.5 kHz and a minimal rise/fall time (10%-90%) of about 9 µs. The photodiode was placed at 4.5 cm from the screen surface at a straight angle. Ambient light was kept from the measured area by a rubber sleeve of 3 cm diameter which also limited the maximal incident angle to about ±20°.
The vertical extent of the measured screen area was not only limited by the rubber sleeve but also by the stimulus being a horizontal stripe covering just 5% of the screen height. Note that the OLED pixels are updated sequentially from the top of the screen to the bottom, which results in different delays for the luminance curves, depending on the vertical measurement position. By limiting the measurement to only 5% of the full vertical screen size, the smear effect caused by averaging over differently delayed luminance signals is limited to a well defined value. For example, at a refresh frequency of 120 Hz, the screen is updated within around 8 ms, so that a theoretically instant luminance onset on a single-pixel level would result in a measured luminance curve with a 0% to 100% transition ramp over a duration of 5%·8 ms = 0.4 ms. This smear effect is what limits the bandwidth of the measurement and determines, for example, the shortest rise/fall times that can be accurately measured this way. Normally, i.e., when measuring LCDs, this is not the overall-limiting factor, because LCD pixels switch considerably slower anyway. But with OLEDs, where the luminance onset – on the single-pixel level – can indeed be assumed to be instantaneous, this smear effect is the overall-limiting factor.
For more detailed information on the measurement method and the presentation of the results, see Flicker-free settling.
Settling curves
Figure 6 shows the luminance signal over time for the horizontal stripe being switched from black to white and back. Since the pixels in OLED monitors respond virtually instantaneously, the exact shape of the luminance transition is of minor interest here. Nevertheless, Figure 6 still provides some noteworthy insights.
- The latency (or input lag) is as short as can be expected, namely about half the duration of a refresh cycle. Note that these measurements were taken at the screen center; therefore, and assuming that the pixels are updated at the very time they are received, half of the frame had to be received when the update process arrives at the vertical screen center where the photodiode was placed, which is why latency was measured to be half a refresh cycle.
- The amplitude for the first refresh cycle after the switch is higher than for later refresh cycles, as if the monitor was using overdrive. This is only the case for specific luminance values though, which will be shown in the Settling matrix measurements section.
- Each refresh cycle ends (or – depending on how you look at it – starts) with a very brief low-pulse. This is typical for OLED monitors and reflects a pixel reset phase that is part of the pixel refresh procedure. We cannot infer any precise per-pixel pulse timing from our simple measurements here, because the photodiode lumps together several pixel lines into one signal, thereby smearing out such short time events (as explained above). But the smearing is why the relative pulse amplitude is seemingly higher for 240 Hz than for 120 Hz; at double the refresh rate, the recorded pixel lines are simply updated twice as fast, which results in less smearing and, thus, in a less washed-out pulse.
The vertical gray lines mark the times when the OpenGL command sequence SwapBuffers();glFinish(); returns control to the PC program, which is about when the 1st line of a frame is sent out to the monitor. This is also when, for the black to white frame switch, a hardware trigger is updated (pink trace), which is recorded along with the photodiode signal. For this recording, the photodiode was placed at the vertical center of the screen.
Settling matrix measurements
The metrics typically presented in this section, such as settling times, fall/rise times, and similar measures, are intended to characterize the shape of the pixel response curves. However, for OLED monitors with their ultra-fast pixel response time, most of these measures have become largely irrelevant. Those that remain useful may have to be tweaked in order to account for new signal features, such as the pixel reset pulse, which otherwise would contaminate the measurements.
Figure 6, in particular, suggests that overshoot – or, alternatively, the average luminance error in the first refresh cycle after a pixel level change – might still be worth looking at. Since luminance remains mostly constant throughout a refresh cycle, the average luminance is the measurement of choice, provided that time intervals potentially affected by the limitations of the measurement method are excluded. To this end, the pixel reset pulse was excluded from averaging.
Note the different z-axis scales for the two monitors. The z-axis for the MSI matrix plots has been clipped at 6% to improve the scale for the values of interest.
A "positive" error corresponds to the luminance being higher in the first refresh cycle than in later refresh cycles, no matter whether the FROM pixel value was larger or smaller than the TO pixel value.
To assess the aforementioned luminance error, the methods described in Flicker-free settling were used, with some modifications: the photodiode gain was set to 60 dB (instead of 70 dB), a low-pass filter frequency of 4 kHz (instead of 70 Hz) was applied to preserve the features of the fast OLED pixel response, and the signal averaging was limited to a time interval that is free of potentially contaminated samples, as outlined above. Note that, with these procedural adjustments, the signal bandwidth is ultimately constrained by the width of the measured stripe on the screen.
Figure 7 shows the luminance errors for the first refresh cycle following a pixel level change, for 120 Hz and 240 Hz, compared against the MSI MPG 271QRX. The large negative errors for the MSI are mainly caused by slow rise times rather than by too low target luminances in the first refresh cycle. In direct comparison to the ASUS, the MSI exhibits smaller errors for low-to-high pixel changes (FROM < TO) but larger errors for high-to-low pixel changes (FROM > TO). Even though ASUS exhibits errors up to 5%, and only when switching from black to anything brighter, these values are still way smaller than for LCD monitors, where anything below 20% would be considered above-average good already. Whether or not a 5% error holds any practical relevance, the observed error pattern raises questions about the underlying causes.
Could it be that the measured luminance deviations are actually not errors but compensations for, e.g., detrimental effects of the pixel reset pulse, similar to how overdrive in LCD monitors accelerates (and compensates for) slow rise and fall times? On the one hand, the reset pulse – on a single pixel level, i.e., not as measured here – is considered to be too short to have any relevant effect on the luminous output energy per refresh cycle; on the other hand, these pulses obviously carry enough energy to leave traces even in our bandwidth-limited signal (see Figure 6). Note that the reset pulse is always negative, thereby giving high-to-low pixel level switches some advantage, regarding switching latency, over low-to-high switches, at least for the more extreme target levels (i.e., when switching from dark to bright or vice versa).
Here is another observation regarding the pixel reset pulse. Although we cannot measure the pulse width and amplitude on a single-pixel level directly, we still can infer the relative pulse energy from our bandwidth-limited measurements – at least for the settled portion of the pixel response curves –, namely by comparing two simple measures: the average luminance (LP) over the entire refresh cycle (i.e., including the pulse) and the average luminance (L0) over the refresh cycle excluding the pulse. We then define the relative reset pulse energy e simply as e = 1 - LP / L0. If, on a single-pixel level, the pulse level is always zero and the pulse duration is constant (i.e., independent of the pixel level), then we would expect the relative pulse energy to be the same for all pixel levels. This, however, is not at all what was measured – see Figure 8. Although the relative pulse energy scales nicely with the refresh rate, which is in line with the reasonable assumption that the pulse duration does not depend on the refresh rate, the relative pulse energy strongly depends on the luminance level. Mind you, this is just for settled refresh cycles, i.e., for refresh cycles where effects of any potential pixel value switches have completely faded. This is why we cannot draw any conclusions from this finding other than that the luminance settling behavior might be more affected by the reset pulse than expected – in whichever way. There is one caveat: inferring the pulse energy from a bandwidth-limited signal assumes that the measurement system is sufficiently linear. However, given that the ASUS and the MSI, which are two monitors with rather different OLED panels, appear to have such similar pulse energy signatures raises doubts regarding the measuring method used here. Maybe these results just reflect some shortcomings of the measurement devices (photodiode, amplifier, oscilloscope) and/or the analysis method.
Back to the luminance error pattern. The luminance error, as a however defined measure, might largely depend on where exactly the refresh cycle is considered to start and how the effects of the signal bandwidth limitations are taken into account. There are a number of reasonable definitions for this error measure, and which one is the most appropriate might also depend on the use case one has in mind. One measure might be better for capturing luminance artifacts (e.g., of moved edges), another measure might be better for capturing color distortions. For example, the start of the refresh cycle could be determined for each FROM–TO level transitions individually by aligning it with the 50%-point of the respective luminance transition edge, in contrast to using one absolute time point for all FROM–TO level transitions, which meets a certain criterion, like the 50%-point, only on average. Or, as another example, the pixel reset pulse could be either excluded from the luminance averages, or not. Here, we chose to exclude the pulse, assuming that its impact is – if not negligible – at least the same for all level transitions. However, the findings summarized in Figure 8 suggest that both of these assumptions are likely wrong. Neither is the impact necessarily small (see dark levels), nor is the impact the same across different levels.
In conclusion, the results presented in this section should be interpreted with caution.
Color spectra
.png){kind=link}
.png){kind=link}
.png){kind=link}
.png){kind=link}
