From Display-Corner
Jump to: navigation, search



Model:FORIS FG2421
Panel type:VA
Max. refresh rate:120 Hz
Panel:SHARP LQ235D1LW03
Processor:MST 8757T (MSTAR)
Backlight type:LED (white, edge)
LED driver:BD9276EFV
TachistoMode:no, not really
Price (approx.):USD 600
Available since:11/2013



EIZO FORIS FG2421 review on FlatPanelsHD.

At a glance

The EIZO FORIS FG2421 is the first monitor with a VA panel (Vertical Alignment) operating at 120Hz and a flashed LED backlight. The VA panel promises much better color quality and higher contrasts over a wider range of viewing angles as compared to 120Hz monitors with TN panels that also offer flashed backlight mode in form of 2D LightBoost (e.g. BenQ XL2420T or Asus VG248QE). Another plus on the FG2421’s side is the high brightness (400cd/m2), which is still high (250cd/m2) when the monitor is operated in flashed backlight mode.
First reviews on the EIZO FG2421 were extremely positive, but meanwhile also more critical opinions have been voiced. The bottom line of this mini review is that the FG2421 is not so much more than a proof of concept, the concept being to combine a VA panel with 120Hz flashed backlight technology. There are numerous problems which are rather attributable to sloppy design and manufacturing than to technology itself. This is disappointing and somewhat unexpected given that EIZO usually knows how to build good monitors and given that the price tag of this monitor is well above other 120Hz monitors. But EIZO is to blame only for part of the problems as the VA panel, which accounts for the other part, is manufactured by SHARP.


In the following, several issues are discussed in more detail. First off, when reading the respective forums one can get the impression that there is a huge variability in the manufacturing quality. But most of this is probably just because of different observation standards and, more importantly, different settings and test conditions. Also, some problems seem to surface only occasionally or under very specific conditions. Nevertheless, such problems should not be ignored unless it is clear under which circumstances they do occur and how to work around them. Otherwise, according to Murphy’s law, such problems will not only appear in tests that have been tailored to make them appear but will also pop up in real applications.

Figure 1: Histograms of systematic luminance errors for 256 pixel values, where errors are given in percent of local step size. The distribution for the EIZO (shown in red) is wider, reflecting larger errors (see also Measuring color resolution).

Color resolution

The VA panel is a 10bit device but operated at 8bit+FRC (i.e., virtually 10bit, according to the TFT CENTRAL review). However, the effective color resolution also depends on the pixel value processing, which seems to be worse (by 40% or so) for the EIZO in terms of systematic errors as compared to the BenQ XL2420T which only has a 6bit+FRC panel (see Figure 1 and Measuring color resolution). This becomes most obvious at dark levels where the perceived step sizes between adjacent input levels are not only quite big but also quite non-uniform (Lagom black level test). That this non-uniformity is actually caused by systematic round-off errors and not just by some other non-linearities can be also seen when sweeping through the monitor’s contrast setting, which causes according changes in the pixel value calculations. What can be observed then is how the luminance of the different gray patches are "jumping" to the next level at different contrast settings (best seen in a dark room when looking at the monitor from above).
Note that the effective color resolution is difficult to quantify, as there is a lot of variation, and the perceived step sizes along with the step size errors are getting exaggerated by EIZO's deep black levels. Although it seems to be safe claiming that the EIZO is doing worse, it is not to an alarming extent.
A general problem regarding color resolution comes with the high contrast ratio the EIZO provides, because the dynamic range that has to be covered by the available pixel values is much wider than for monitors with more traditional contrast ratios. With only 8bit per channel, cut-backs in perceived color resolution are inevitable. These cut-backs occur mainly at low luminance, which is a consequence of the (historical) choice of gamma. So there is obviously a trade-off between color saturation and perceived color resolution, especially when it comes to low luminance (see also the Black level section).
By the way, a high black:white contrast ratio does not necessarily mean high color saturation. A red pixel, for example, gets washed out (i.e., desaturated) because the blue and green sub-pixels of that pixel are not perfectly dark. This is due to an imperfectly dark black level (as measured for the maximal contrast ratio) and due to stray light coming from the red sub-pixel. Note that colors on an LCD are produced by color filters sitting at the very end of the light path, so stray light is still white before passing the filters.

Figure 2: Gamma curves for viewing angles of 0° (straight view, in blue) and 15° (view from above, in red). The luminance has been normalized for each curve to the respective maximal luminance (the luminance for 15° was about 80% of the luminance for 0°).

Viewing angle

TN panels are known to have small viewing angles, meaning that the color saturation and color hue do change substantially when looking at the monitor from different angles. VA panels are better in this respect but suffer from a different effect called gamma shift. Although the color saturation and color hue stay pretty constant over a wide range of viewing angles, the gamma transfer function does not, as shown in Figure 2. This might cause, for example, dark levels to vanish in the black when looking at the monitor from a straight angle.

Flashed backlight

The monitor can operate in TURBO-240 mode (TURBO mode, for short), in which case the LED backlight is not constantly on but is flashed in synchrony with a 120Hz frame rate. This mode is only free of timing artifacts with an input frame rate of 120Hz, which is internally doubled to 240Hz (about applicable frame rates, see below). The important thing regarding backlight is that the backlight is actually flashed twice per 120Hz frame, as can be seen in the right panel of Figure 3. The first pulse (pre-strobe) is very short though (270µs), whereas the second pulse (main strobe) is between 2.1 and 2.5ms long, depending on the Brightness setting (below Brightness=75%, the LED current is changed, above 75%, the strobe pulse width is changed). The pre-strobe is placed in the middle of two main strobes and might have just been added to somewhat alleviate flickering. This goes, however, somewhat against the idea of motion blur reduction.
On the positive side is the output luminance range, which goes from 45 cd/m2 to 285 cd/m2 (245 cd/m2 at Brightness=75%, i.e., max. luminance at minimal pulse width).

Figure 3: Settling curves for three different step sizes while TURBO mode was off (left) or on (right). For TURBO=off (left) the curves were shifted and scaled individually according to the respective step size, whereas for TURBO=on (right) the curves were just scaled to the maximal luminance.
The vertical gray lines mark the ends of the VSync intervals (i.e., when the 1st line of a frame was about to be sent out by the computer). Measurements have been taken at the screen center with the photo diode PDA36A (Thorlabs).

Anyway, from a technological point of view, flashing the backlight is not a big deal. What is a big deal though is the adjustment of the overdrive amplitude so that it results in a fast and accurate output luminance settling in combination with the pulsed backlight. Here is where 240Hz can make the difference, not because it is double the refresh rate but because it allows the panel to be updated in half the time, leaving enough settling time even for the bottom pixels before the backlight pulse renders them visible. Whether the internal 240Hz refresh frequency is used to its full potential, like calculating and applying overdrive also at a 240Hz frequency, is a different question; but even if it is just the faster update speed that is put to a use, it is an improvement over, for example, the BenQ XL2420T which is only able to update the panel within about 6ms.
Besides the small pre-strobe that might somewhat diminish the motion blur reduction, there is another issue with the TURBO mode. This is a 60Hz luminance modulation: every other long pulse is a bit darker than the other. This modulation is permanent and the modulation amplitude seems to depend on the vertical pixel position, which indicates that this effect is not caused by backlight control. The modulation phase, i.e., which pulse is higher, does not depend on the input signal. The modulation can bee seen in the red and the green curves at the right panel of Figure 3. Note the hi-lo pattern for the green curve is reversed as compared to the red curve. This is so by accident and just indicates that the modulation phase does not depend on when, for example, the first bright frame was presented.

Figure 4: Pixel-walk artifacts (pixel inversion) while in TURBO mode (monitor mode setting FPS1). Normally, the image should be just gray. See the Lagom Pixel-walk tests for details.
The artifacts are the stronger the more of the screen is covered by the test pattern. As the browser window is made smaller, the patterns look more like they are supposed to look, i.e., the horizontal lines become weaker and the color does change towards gray.
With some other test patterns the monitor flickers very strongly.
The artifacts are much weaker in non-TURBO mode (e.g. monitor mode setting WEB).

Pixel inversion ( pixel-walk)

The pixel state depends on the DC voltage applied to the liquid crystals. In order to not compromise the molecular structure of the liquid crystals, die polarity of the voltage needs to be switched from time to time. This is what pixel inversion refers to. It is difficult, however, to maintain the exact absolute voltage level for both voltage polarities and, thus, the pixel state varies accordingly. So in order to not induce flicker by this voltage polarity switching, the two different polarities “+” and “-“ are distributed across the sub-pixels and are switched with each panel refresh. Normally, special test patterns are used to make just the difference between the “+” and the “-“ voltage visible. But these patterns do not only reveal static differences between the voltage polarities but also show how stable the voltage sources are, possibly depending on the currently applied pixel pattern and the pixel location. This is where the EIZO FG2421 shows some unexpected weakness. Especially in TURBO mode, where basically only every other internal refresh phase becomes visible and thus half of the averaging potential regarding pixel inversion gets lost (namely averaging inversion effects over time), weird artifacts appear. This would not be so alarming if these artifacts were not so severe for certain monitor settings, suggesting that under certain circumstances pixel inversion effects can be clearly visible also without provoking them by tailored test patterns.
Such strong artifacts can be observed when looking at the Lagom Pixel-walk tests while in TURBO mode and while sweeping through the monitor's black-level setting (Figure 4). Here, the black-level setting is just used to easily change the gray levels of the dark sub-pixels in the test patterns. It is likely that the same effect can also be elicited while keeping the black-level setting fixed but, instead, sweeping through different gray levels as the test pattern are generated (which is not implemented at the Lagom site).
These test patterns are tailored to reveal pixel inversion artifacts. However, this does not necessarily mean that the observed artifacts are all caused by pixel inversion. Other causes are also thinkable which are triggered by these test patterns just by accident.

Input lag

The EIZO FG2421 shows more input lag than many other 120Hz monitors. Input lag is difficult to measure and there are also several definitions out there of what input lag means, especially when it comes to monitors with flashed backlight. If taking the end of the input signal's VSync as a reference, which is when the computer sends out the first pixel line, the luminance curve measured at the screen center crosses the 50% mark with a delay that is about 10ms longer for the EIZO FG2421 than for the BenQ XL2420T. This is the case in TURBO/LightBoost mode (pre-strobes in TURBO mode not taken into account) but also when backlight is permanently on (see Figure 3 and Figure 5 for comparison).

Figure 5: Settling curves for the BenQ XL2420T for three different step sizes either with LightBoost disabled and AMA=on (left) or with LightBoost=80% (right). In the left panel the curves were shifted and scaled individually according to the respective step size, whereas in the right panel the curves were just scaled to the maximal (settled) luminance.
The vertical gray lines mark the ends of the VSync intervals (i.e., when the 1st line of a frame was about to be sent out by the computer). Measurements have been taken at the screen center with the photo diode PDA36A (Thorlabs).

Frame rates

The high input lag already suggests that the EIZO FG2421 first buffers the input frames before forwarding them to the panel. Although the monitor accepts refresh rates other than 120Hz, the internal processing seems to keep running at a 120Hz frame rate no matter what. Note that there is no tearing when the monitor is fed with, for example, a 100Hz signal, but input frames are sometimes presented for two 120Hz time periods in order to bridge accumulating time gaps that are inevitable when trying to fit 100Hz in a 120Hz time raster. This makes 120Hz basically the only applicable refresh frequency.

AG coating

The EIZO FG2421 comes with a semi-glossy anti-glare coating that reduces reflections much less than the matte coating of the BenQ XL2420T or Asus VG248QE. On the other hand, this also means that less stray light coming either from the monitor itself or from the room illumination, which helps to preserve the image contrast and results in slightly less image blur. Which coating strength is best really depends on the application and personal preference. However, the coating of the EIZO FG2421 suffers from something called cross-hatching, an artifact probably caused by the adhesive between the coating foil and the panel glass not being distributed homogeneously. Some describe this artifact as patterns of faint diagonal lines across some screen areas, possibly seen only at certain gray levels and/or at certain colors and/or under certain viewing angles. But even if line patterns are not clearly visible, the panel might still look kind of patchy or splotchy because of cross-hatching. This is one of these effects which are almost impossible to document by camera pictures, and reports regarding these effects might be very subjective. Therefore, it is not so clear whether this is a general problem basically all FG2421 monitors will show or whether only some bad samples are affected.

Black level

VA panels generally exhibit very dark blacks, which results in a high white:black contrast ratio. However, as far as the EIZO FORIS FG2421 is concerned, this comes at a price. The panel needs a rather long time to switch from completely black to any other level. This is especially true when switching from deep black to levels which are just close to black, a fact also mentioned in the TFT CENTRAL review. That there is something to it can be also seen from measurements for black-to-white switches (not shown here) where the rising edge of the luminance curve is delayed by about 2.5ms as compared to a dark-to-white switch. To some extent, this can be also seen in the left panel of Figure 3, where the onset of the blue curve (black-white switch) clearly lags behind the other curves for the rising but not for the falling edge.
In general, a higher maximal contrast means that a wider dynamic range has to be covered by the limited number of available pixel values. Even if the monitor made optimal use of the available value space by minimizing digital noise and by applying an optimal gamma function (which would have to be supported by the imaging software as well), the perceived difference between two adjacent gray levels would still have to become bigger at some point. The generally used gamma value is, however, not optimized for such high contrasts and neither is the nominal color resolution of 8bit per channel really sufficient to cover contrast ratios like 5000:1 as found in the EIZO.

Figure 6: Photograph of the EIZO FG2421 "glowing". The picture resembles how the glow appeared on the monitor. In order to minimize gamma shift effects, the camera was put at a distance of 1m at approximately the height of the upper monitor edge.

Backlight bleed, glow

The EIZO FORIS FG2421 does not show any signs of conventional backlight bleeding, which otherwise would be visible on a pitch black screen already. However, at levels a bit brighter than pitch black the monitor seems to glow here and there, especially along the right edge and with some greenish tint, as shown in Figure 6. This has been reported by several people and it is seen as one of the weakest points of the EIZO FORIS FG2421.

VESA mount

The EIZO FORIS FG2421 does not provide holes for a VESA mount. However, it is apparently pretty simple to remove part of the back cover and still attach a VESA mount.

Matrix measurements

Flicker-free backlight

Evaluating the flicker-free backlight mode (TURBO240=off) is relatively easy because the EIZO FORIS FG2421 does not offer any user settings regarding overdrive which could have an impact on the settling behavior. As already mentioned in the Black level section above and as can be inferred from Figure 7, the EIZO is rather slow when switching from black (FROM=0) to dark gray levels (TO<5 or so). Apparently, the EIZO is driven well into saturation when presenting black and needs quite some time to get out of saturation, especially if the target level is close to black and, thus, does not provide much of a pull. Raising the Blacklevel setting improves the situation somewhat and might also help to alleviate the problems mentioned earlier regarding excessively high B/W contrasts, obviously at the price of a slightly increased black luminance and, thus, reduced B/W contrast (e.g., matrix plots for BL=60%,C=50%). Another concern is the Contrast setting, which might be too high to provide enough overdrive margin at all luminance levels. But reducing the Contrast setting from 50% to 40% does not result in an improvement; it is rather the contrary. A few more conditions are shown in the comparison chart (Figure 8).

Figure 7: Settling measurements with Blacklevel=50% and Contrast=50% at 120Hz.
For further details on the graphs and the measurement method see Flicker-free settling.

Figure 8: Comparison chart for the some flicker-free backlight test cases (120Hz). Note that smaller bars/values are better. The colored bars refer to the maximum of 90% of all step sizes (max90) and the gray bars refer to the maximum of 100% of all step sizes (max100). BLxx stands for Blacklevel setting, and Cxx for Contrast setting.
For further details on the measurement method see Flicker-free settling.

Strobed backlight

The measurements for the strobed backlight (TURBO240=ON) were taken when the strobe pulse width was set to its minimum (2.1ms, i.e., with a Brightness setting of below 75%, providing a maximal luminance of 245 cd/m2). Note, that the luminance errors refer to the average luminance during one entire refresh cycle, meaning that the errors are also affected by the 270µs long pre-strobe. Because the overdrive margin depends on the contrast setting, measurements were taken at two different Contrast settings (40% and 50%), but overall the C=40% case is actually a bit worse, which suggests that overdrive has been optimized for C=50%. So only the C=50% case is shown here (Figure 9). Note that the occasionally big worst case errors are dominated by switching from dark to dark, which might be only of minor practical relevance. Thanks to the internally doubled refresh frequency, which comes with an accordingly high panel update rate and gives the pixels a quite long settling time before the main backlight strobe, the errors are well balanced across the different screen locations (top vs. bottom). Actually not so good are the high residual errors at later refresh cycles. The measurements might even underestimate the true errors as they are averaged. This is because at least part of these errors are caused by the weird 60Hz luminance signal modulation, which can be well seen in the right panel of Figure 3. Since the measurements were not synchronized to the phase of this luminance modulation, the according negative effects on the error measure did have a good chance to average out, at least to some extent.

Figure 9: Settling error measurements for TURBO240=ON mode at 120Hz, minimal pulse width (Brightness setting of <75%), a Backlight setting of 60%, and a Contrast setting of 50%. Note that the Z axes are scaled the same within columns (i.e., across screen locations) but not within rows (i.e. across refresh cycles).
For further details on the measurement method and presentation see LightBoost settling.

Figure 10: Comparison chart for the some 120Hz strobed backlight test cases (TURBO240=ON). Note that smaller bars/values are better. The colored bars refer to the maximum of 90% of all step sizes (max90) and the gray bars refer to the maximum of 100% of all step sizes (max100). The scaling of the bar lengths is the same within columns (i.e., for a particular refresh cycle), but different within the rows.
Note that the occasionally 100% errors (gray bars) might be of minor practical relevance as they are often dominated by just dark-to-dark switches.
For further details on the measurement method see LightBoost settling.

Connector on the LED driver board
Pin Signal
1 18.2V
2 18.2V
3 18.2V
4 18.2V
8 3.3V
9 LED current control, (3.3V PWM @ 18kHz)

The PWM ratio controls the DC LED current level, which
is then gated by the LED on/off signal (pin 10).
TURBO=OFF: Max. on:off ratio = 21.9:33.5µs
TURBO=ON: Max. on:off ratio = permanently on.

10 LED on/off, Strobe (high-active)

Accessing the backlight signals

Due to its extremely low black luminance, the EIZO FORIS FG2421 would be perfectly suited for tachistoscopic applications. Unfortunately, the LED driver is not so well suited. Preliminary findings reveal several problems. After a few seconds of inactivity the LED driver switches off completely and can only brought back to life by switching the power off and on again. The onset behavior after an inactivity of up to 500ms is pretty good, but the longer the pause the more noise does the onset make. Even if the LED driver would tolerate such operation conditions on the long run, the noisy indication of the onset might be prohibitive for the application at hand.
Disassembling the monitor is not particularly difficult. First, the upper plastic cover of the stand has to be clipped off, which uncovers 3 screws that have to be removed. Somewhat unusual, the back cover is screwed to the interior by another set of 3 screws. Other than that, the back cover is just clipped in, but care should be taken when taking it off, so that the plastic hooks connecting the bezel to the back cover do not get ripped off (see Figure 11).
Inside, there is the main box with three PCBs, the main board, the LED driver, and the power supply (Figure 12 and Figure 14). The main board actually does not connect to the panel controller but connects to a forth PCB first, which is directly mounted on the panel (Figure 13), possibly a signal level converter of some sort.

Figure 11: View at the back of the LC panel which is framed by the bezel. The bezel has two types of elements connecting to the back cover which work along different directions (as being marked by the red arrows). Be careful to not break off the small rails (vertical arrows) when unlocking and removing the back cover.
Figure 12: Main electronic case, photographed from the side which normally faces the panel. Main board (left, LED driver (middle), and power supply (right).

Figure 13: The signal converter, which is directly mounted on the back of the LC panel.
Figure 14: The unmounted PCBs flipped over.

Personal tools