We’re working with the GTX 1650 SC Ultra Gaming, a 4GB GDDR5 solution that costs $170, about $20 over the price floor for the lowest-end cards, and includes a 6-pin connector for more clock-boosting and overclocking headroom. The EVGA GTX 1650 SC Ultra runs a 90W board power stock, whereas the reference design is suited for 75W. The GTX 1650 SC Ultra uses a dual-fan cooler design that we’ll discuss more in our tear-down video and in the thermals section of this video. This model also includes a metal backplate, one of the primary marketing features of the two-slot, two-fan design.

The card runs 896 CUDA cores, or 7 streaming multiprocessors at 128 CUDA cores per SM, which leaves us wondering when the 1650 Ti will release as this is not a fully populated GPU die. EVGA’s boost clock on this card is 1860MHz with memory at 8Gbps stock on the 128-bit bus, enabling operation at 128GB/s.

Let’s get into benchmarking. We’ll have 1080p and 1440p tests primarily, with overclocking, power, thermals, and frametime charts thrown in – and we managed to get all of this done in the last few hours. Note that we’re still testing with higher settings. You could always lower settings to improve absolute framerate, but relative framerate scaling will remain roughly the same.

GPU Test Bench (Sponsored by Corsair)

F1 2018

F1 2018 and the EGO engine are next. At 1080p, the GTX 1650 SC Ultra runs at 61FPS AVG, with room to improve if settings were reduced toward High or Medium. The RX 570, for competitive reference, runs at 66FPS AVG. This is a lead of about 9%. The GTX 1060 6GB still holds a strong lead over both, running at 75FPS AVG, demonstrating that the 1650 isn’t the successor to the 1060, but rather the 1660 non-Ti was. An overclock gets us to 66FPS AVG, which ties the RX 570 stock card. Not a great place to be.

At 1440p, the 1650 runs at 46FPS AVG with ultra-high settings, which really isn’t all that bad when considering where we came from a few years ago. A settings reduction would even allow an average of 60FPS, if so desired. Unfortunately for the 1650, the RX 570 still performs better here – not just in average framerate, but also measurably and perceptibly in frametime performance. This is something we’ve demonstrated relating to F1 2018 in the past, too. Overclocking pushes to 50FPS AVG.

Apex Legends

Apex Legends is up next. This is a modern DirectX 11 game that we set to all “High” settings, then use a highly controlled multiplayer test course. We test away from other players, but in the actual game. Our previous content discovered that this was highly representative of real gameplay and is the most accurate way to benchmark.

At 1080p and with all High settings, the GTX 1650 SC Ultra places at 64FPS AVG, with lows reasonably well-timed at 49 and 41FPS 1% and 0.1% lows. The the RX 570 isn’t as advantaged here as it was in Sniper, as the workload has moved back to DirectX 11 and drops asynchronous command support. The 570’s 70FPS AVG allows it a lead of 9% over the stock 1650 SC Ultra. Overclocking the 1650 got it close to the stock RX 570, but the card still fell short with a 69FPS AVG, or an improvement of about 7% over baseline. The GTX 1060 6GB card hasn’t been retested since our initial round, but was at 72FPS AVG at the time of last test. It may be a few percentage points faster now, but the results overall haven’t changed much for other retested cards.

Moving on to frametimes, again being mindful that lower is better, but more consistent is better than lower, we see that the Gigabyte RX 570 4GB card managed performance in the range of roughly 11ms to 25ms, with the average falling closer to 15-16ms. Although the data is more sporadic than Sniper’s smooth frametime plot, there still aren’t any extreme swings in frame-to-frame interval. This is important.

Finally, we’ll plot the GTX 1050 Ti just to get it on the chart, although it’d need a serious settings reduction to really play fluidly in this game. The 1050 Ti doesn’t have extreme excursions from the mean, either, but is certainly slower on average.

Sniper Elite 4

Sniper Elite 4 will start us off. This game uses the Asura engine and is one of the best-built DirectX 12 titles out. We use Dx12, Asynchronous Compute, High settings, and test primarily at 1080p for this class of video card; that said, we also have some 4K tests strictly for synthetic comparison to a wider range of products. For now, let’s start with the more realistic 1080p setting.

At 1080p, the GTX 1650 SC Ultra runs Sniper at 71FPS AVG, with lows at 61FPS and 53FPS 0.1%. This particular game is compute-intensive, and so AMD’s RX 570 leverages architectural advantages to lead the 1650 by about 36%, with a 97FPS AVG. The GTX 1060 6GB Gaming X leads both the RX 570 and the GTX 1650, although the latter is led more significantly, with its 104FPS AVG. Overclocking the GTX 1650 to about 2085MHz gets it to 77FPS AVG, an improvement over baseline of 8%. Improvement in the 1650 over the 1050 Ti is meaningful, posting 24% uplift, but it doesn’t compete well with the RX 570 in this title.

Here’s a frametime plot for Sniper 4. Remember that frametimes are the base metric for framerate, and are important as they don’t average-out potential issues that AVG FPS and even 1% lows can obscure. The RX 570 4GB card runs exceptionally well here, with an average frametime of about 10ms frame-to-frame. This line is nearly perfectly flat, which is what we want to see. The flatter the line, the less the variance. For reference, 16.667ms is 60FPS. We have observed that excursions beyond 8-12ms frame-to-frame will cause users to notice stutter. As for the new GTX 1650, the line is relatively smooth, it’s just slower than the RX 570.

Finally, the GTX 1050 Ti 4GB card ends up at around 15-22ms on average, with more sporadic frametime consistency along its line. It’s not bad, but this card has clearly never been particularly high-end.

4K is obviously not a real-world scenario for these cards, but we’re going to briefly look at a 4K Sniper chart for purposes of synthetic comparison and seeing the card hierarchy. The RX 570 again leads, this time by 41% over the GTX 1650 SC Ultra. Overclocking the SC Ultra marginally helps, but not enough to change the stack. The 1650 and 570 are both still led by the GTX 1060 6GB.

Far Cry 5

Far Cry 5 at 1080p has the GTX 1650 SC Ultra at 56FPS AVG, which is again behind the RX 570’s 62FPS AVG. The 570 maintains a steady lead of 10.5% while managing to be cheaper. The GTX 1060, for what it’s worth, still leads, but we’re only pointing it out because some rumors had suggested the 1650 might surpass the 1060, and that’s clearly wrong. Overclocking puts the 1650 to 61FPS AVG, an improvement of about 8%.

1440p doesn’t change the story much: The RX 570 again leads the GTX 1650. Neither card is particularly meant to play this game at 1440p, so we’ll move on quickly to the next chart.

GTA V

GTA V has typically posted much better performance on NVIDIA than AMD when compared relatively to other games, and that remains true here. At 1080p, the GTX 1650 SC Ultra ties the RX 570 for the first time all review. The two results are roughly within margin of error of one another for averages, although we’d like to see better lows than the 1650 posted. The GTX 1060 still leads both handily, with an 83FPS AVG – it’s not even close to what some of the rumors projected, with the 1060 6GB still holding a lead of 31% over the 1650. Overclocking allows an improvement to 68FPS AVG, which is enough to tie the stock 780 Ti reference model, although obviously the overclocked 780 Ti holds a significant lead. The RX 570, for what it’s worth, could also be overclocked here, but we’re arguing over single-digit movements at this point. The two cards are effectively the same in AVG FPS for GTA V.

At 1440p, the GTX 1650 SC Ultra continues its tie with the RX 570, with an overclock stretching up toward 46FPS AVG, nearing the RX 580 Gaming X stock card. That’s not bad, but GTA V has been the only instance of this thus far. An overclocked 570 would encroach similarly on the 580, as we demonstrated years ago with the RX 400 and 500 launches.

SOTTR

Shadow of the Tomb Raider is another modern, DirectX 12 title. For this one, the GTX 1650 SC Ultra operates an average of about 50FPS, predictably allowing it to lead the 960 and 1050 Ti, but also allowing the RX 570 a lead of about 9.3%. Please note again that dropping settings further would obviously allow relative playability in Shadow of the Tomb Raider. The 50FPS AVG is fine for comparative purposes with our higher settings, but if you wanted to play the game with a higher FPS, it could be done at the drop of quality to about medium. Overclocking gets us to 52FPS AVG, although instability in the OC has dropped our low-end performance for frametime consistency. This could be improved with more time – for example, if NVIDIA released drivers to press ahead of launch.

At 1440p, the 1650’s 33FPS AVG doesn’t inspire much confidence with these settings. The game would need a serious settings reduction to bring this closer to fluidity. Strictly for comparative purposes, though, we learn that the RX 570 holds a lead once again, this time at 38FPS AVG, a lead of about 15%.

OC Stepping

This table shows our overclock stepping, providing a look at the very quick process we followed to get to our overclock. We were able to sort of hold 2100MHz, but it wasn’t stable in all games tested, and so we had to pull back to 2085MHz for the final average frequency. Peaks depend on thermals, as always, and this is illustrated immediately in the top 3 rows of our table. We changed nothing between those 3 rows and just let the temperature slowly rise, eventually dropping us from 1980MHz to about 1935MHz once nearing 60 degrees Celsius. After this, we started actually overclocking the card, settling for just a 160MHz offset and 400MHz for the memory offset. We sort of had 175MHz offset stable, but 0.1% lows were dropping too hard, and so we reverted to 160MHz core offset.

Power Consumption

Conclusion

The GTX 1650 has basically one use case when compared to its on-market competition: Small, low-power video outs that can still handle some kind of minimal GPU workload. If you don’t care about the difference in power consumption – and at this price point, not many do – and you also don’t care about needing a power cable, then the RX 570 is a much better purchase for the gaming audience. Most of the ones we saw on launch day were around $130, versus an initial 2017 price of ~$180-$220, and that $130-$150 pricing is just objectively better than the GTX 1650 options on market. Further driving the knife into the heart of the 1650, the RX 570 is a card from 2016, but disguised: In 2016, AMD launched its RX 470 video card, which the RX 570 usurped in 2017. They’re identical, aside from a higher stock clock on the RX 570. We didn’t even overclock the 570 in this review, mostly for time reasons, and it still remains in better positioning than the GTX 1650’s overclocks. The GTX 1650 is simply not competitive in its price class. It is handily beaten by a 2-year-old card that's built atop a 3-year-old GPU, dishing NVIDIA the same embarrassment we dished to AMD for competing with the 1080 Ti two years too late.

EVGA’s GTX 1650 SC Ultra performed reasonably well as an overclocker, hitting 2085-2115MHz (depending on thermals, averaging 2085) when pushed, but still couldn’t match the RX 570. On average, the RX 570 is 7-10% ahead of the GTX 1650, with the GTX 1650 tying in GTA V.

We’ll follow-up more with this card in the coming days, primarily on the YT channel, but it’s looking like a pass from us for now. There are some places it’d make sense, like a 75W setup where no power cables are permissible (maybe replacing an old OEM PC GPU?), but that’s about it.

Editorial, Testing: Steve Burke
Video: Andrew Coleman