Just like the Rival 600, the Rival 650 uses SteelSeries' unique TrueMove3+ sensor system. It basically consists of a TrueMove3 top-of-the-line optical sensor (based on the PixArt PMW3360) and a secondary depth-sensing linear sensor. The latter makes sure the lift-off distance remains nice and low and that the cursor doesn't jump around when you lift the mouse up. The TrueMove3 has shown its greatness inside a few mice before, which hasn't changed over time—it's still one of the best ones available in even 2019. It's raw, snappy, and provides 1:1 tracking with no added acceleration whatsoever.
The nominal acceleration and maximum tracking speed values are 50 G and 350 IPS (8.9 m/s). You can set the resolution from 100 to 12,000 in steps of 100 CPI. The available polling rates are 125, 250, 500, and 1000, which translates into nominal response times of 8 ms, 4 ms, 2 ms, and 1 ms. By default, the mouse is set to the highest lift-off distance setting of 2 mm, but depending on the mousing surface, you can reduce it to below half a millimeter. For maximum performance, I'd highly suggest using the lowest value that still provides perfect tracking.
Battery
There is a 1000 mAH battery inside the mouse, which can be charged with the micro USB cable in the package. Depending on the lighting and performance settings, the battery can last for over 24 hours. It can quickly be charged to about 30% in under 15 minutes, but a full charge doesn't take too long either. 24 hours might sound low, but keep in mind that there are not one, but two sensors operating in this mouse, and the TrueMove3 wasn't made for power efficiency, but raw performance.
Paint Test
There is no jitter on the reasonable CPI steps, unwanted angle snapping (you can turn this on in the software, but I would highly advise against it), or sensor lens rattle.
CPI Divergence
CPI divergence on the Rival 650 is a tad higher than average, so if you're coming from a mouse that translates these steps accurately, you should adjust your in-game sensitivity accordingly. Please note that this test is not 100% accurate, but resembles reality quite well.
Perfect Control Speed
Perfect control speed (or PCS for short) is insanely high on the TrueMove3+ sensor—there is absolutely no way of hitting this value while gaming, even with incredibly low in-game sensitivity. The nominal perfect control speed is more than 350 IPS, which translate to about 8.9 meters per second. The PCS values don't correlate with the set resolution, which means 100 and 12,000 CPI both hit their limits at about the same value.
Speed Related Accuracy Variance
This test shows the sensor's accuracy at different speeds. You can see me doing a fast swipe to the right before I slowly slide the mouse back to its original position. There is no acceleration or deceleration; any displacement is almost entirely caused by human error in this test.
Polling Rate
The 1000 and 500 polling rate values are not entirely consistent, in both wireless and wired mode. I don't know why exactly, but a later firmware upgrade could fix this "issue". I had no problems with this in game though (there was no strange change in motion, skipping, or any kind of sudden positive or negative acceleration), and the other graphs seem to be unaffected by it.
Input Lag & Smoothing
The first two xCount graphs present a simple swipe from left to right. Strange kinks would indicate sensor smoothing, but there are none on any of the CPI steps I've tested. As for the wireless input lag, there is quite a bit. Unfortunately, SteelSeries' wireless tech is not on Logitech's level as there is about 2–3 ms of delay. This isn't too much (and it's essentially undetectable in game unless you're extremely sensitive to this), but when it comes to competitive gaming, it may not be completely negligible. I've played a couple death matches of CS:GO and some other fast-paced games with it and couldn't really tell—the actual weight of the mouse was more off-putting.
The last two graphs show wired performance, and it is indeed topnotch; there is no smoothing or input lag of any sort I could detect on any CPI level. The control subjects were a Ninox Venator and a Logitech G PRO Wireless.