WEBVTT 00:00:00.080 --> 00:00:09.360 As much as it pains me to say it, this review of NVIDIA's RTX 5050 is probably 00:00:06.000 --> 00:00:11.200 the most important GPU review I will do 00:00:09.360 --> 00:00:15.920 this year. Because the bottom line is that for so many of you out there, 00:00:13.599 --> 00:00:19.600 whether you're building your own PC or getting it in the pre-built your 00:00:17.440 --> 00:00:24.880 grandmother got you for your birthday, this is the GPU you're going to use. 00:00:23.279 --> 00:00:29.840 There's nothing inherently wrong with that. Everybody has their own budget. 00:00:27.359 --> 00:00:34.800 It's just that when a better graphics card could be had for the same price or 00:00:32.079 --> 00:00:40.160 if there's a significant upgrade for a much less significant extra cost, it 00:00:37.040 --> 00:00:43.520 hurts me to see these boxes flying off 00:00:40.160 --> 00:00:46.079 the shelves instead despite my years of 00:00:43.520 --> 00:00:51.520 tactipping to the contrary. Which is why this time I am entrusting everyone else 00:00:49.440 --> 00:00:55.440 to persuade you. Maybe when you hear it from the people who actually have time 00:00:53.280 --> 00:01:01.600 to play video games after work, you'll believe us when we say stay away from 00:00:58.960 --> 00:01:07.680 this launch. Maybe they can also help me warn you about our sponsor sponsors. 00:01:05.119 --> 00:01:11.280 >> That was awful. Dbrand. They paid us to mention their protective ghost cases and 00:01:09.760 --> 00:01:14.560 prison screen protectors, which you can pick up using our link in the video 00:01:13.119 --> 00:01:21.680 description. What they didn't pay us to do is say that they're good. 00:01:18.640 --> 00:01:22.640 Okay, go ahead and continue the video by 00:01:21.680 --> 00:01:28.720 Dbrand. I guess >> let me start by pointing out that we're using a high-end test bench to evaluate 00:01:26.400 --> 00:01:32.960 what will likely be the most entry-level offering from NVIDIA this generation. 00:01:31.040 --> 00:01:36.479 This is done to alleviate any potential bottlenecks that may distort our 00:01:34.400 --> 00:01:42.320 results, but it's not exactly a realistic configuration. I mean, nobody 00:01:38.880 --> 00:01:46.240 is going to pair a $250 RTX5050 with a 00:01:42.320 --> 00:01:49.680 $450 Ryzen 7800 X3D or with fast, 00:01:46.240 --> 00:01:51.920 high-capacity DDR5 memory, or at least 00:01:49.680 --> 00:01:55.200 they shouldn't. Realistically, if you buy this card, your performance will be 00:01:53.360 --> 00:01:58.799 at least a little bit worse than what we're going to show you today. So, with 00:01:56.960 --> 00:02:03.759 that in mind, let's kick things off with gaming. This thing blows. And by that, I 00:02:01.520 --> 00:02:09.520 mean it trades blows with our RTX 4060 at 1080p. In Counterstrike 2, the RTX 00:02:07.280 --> 00:02:15.280 5050 is in the bottom four, beating out the two generation old 3060, but losing 00:02:12.480 --> 00:02:19.440 to last gen 4060. And we see that same behavior in Red Dead Redemption 2. On 00:02:17.599 --> 00:02:25.440 the other hand, The Last of Us Part One at 1080p is just kind of sad. The 12 gig 00:02:22.160 --> 00:02:27.520 RTX 3060 pushes the 5050 from the 00:02:25.440 --> 00:02:31.360 bottom. Probably because the 5050 only has 8 gigs of VRAM, whereas this game 00:02:29.440 --> 00:02:36.319 utilized over 9 gigs on cards that had the capacity. Sure, we're crushing the 00:02:33.760 --> 00:02:40.239 RTX 3050. There's no 4050. So, that's the most recent 50 class card. That's 00:02:38.560 --> 00:02:47.319 like congratulating yourself for beating your 8-year-old baby brother in a boxing 00:02:42.400 --> 00:02:47.319 match. Sorry, little buddy. No mercy. 00:02:47.599 --> 00:02:53.920 The 50/50 sometimes beats the 4060, but 00:02:50.800 --> 00:02:56.000 usually not by a lot. F124 is an extra 00:02:53.920 --> 00:03:02.319 frame the 1% lows and 10 more in the averages. And we see a similar uplift in 00:02:58.319 --> 00:03:04.879 Cyberpunk 2077. But what's this? The 00:03:02.319 --> 00:03:11.360 4060 wins once you enable ray tracing in both these games. Yep. The 5050 may have 00:03:08.000 --> 00:03:13.920 24th gen RT cores, but the whole package 00:03:11.360 --> 00:03:19.760 isn't enough to keep up with the 24 third gen RT cores in the 4060. 00:03:17.040 --> 00:03:24.800 Pathetic. If you thought 1080p was bad, well, check out 1440p. The 50/50 goes 00:03:22.640 --> 00:03:29.920 from trading blows to just blowing it against the 4060. It only wins in a 00:03:27.760 --> 00:03:34.879 single game, and that's Cyberpunk. Adding insult to injury, check out how 00:03:32.000 --> 00:03:40.799 good the 3060 12 gig looks by comparison across every single game we've tested. 00:03:38.000 --> 00:03:46.239 Our poor RTX 5050 is barely beating out a 60class card from two generations ago. 00:03:44.000 --> 00:03:52.640 And in VRAMm hungry games like The Last of Us Part One, the 5050 loses again. 00:03:50.239 --> 00:03:58.879 We'll talk more about pure value later, but even the 9060 XT and RTX 5060, 00:03:56.159 --> 00:04:04.080 despite only having 8 gigs of VRAM, destroy the 50/50 in our gaming tests, 00:04:02.400 --> 00:04:09.120 which is pretty bad, especially when you consider that a more futurep proof card 00:04:06.159 --> 00:04:14.560 like the Intel Arc B580 can now be found easily for around $260. 00:04:12.319 --> 00:04:18.479 And that comes with 12 gigs of VRAM. That will make it a safer bet as AAA 00:04:16.560 --> 00:04:22.720 game requirements creep up over the next few years, assuming Intel's GPU division 00:04:21.199 --> 00:04:27.440 still exists over the next few years. But hey, NVIDIA's frame gen can make up 00:04:24.960 --> 00:04:30.880 for the 5050's poor performance, right? Oh, hey, didn't see you there. You know, 00:04:29.440 --> 00:04:35.840 this isn't the most scientifically accurate test we've ever done, but here's some slow motion footage of 00:04:33.680 --> 00:04:40.000 Cyberpunk running on the RTX 5050 at 1080p. The mouse moves, then the game 00:04:37.919 --> 00:04:44.800 moves 27 frames of footage later, which indicates overall system latency of 27 00:04:42.800 --> 00:04:50.280 millisecond. Now, let's enable frame generation and crank it up to 4x 00:04:46.720 --> 00:04:50.280 alongside DAA. 00:04:53.600 --> 00:05:00.479 Our game is now running at around 200 FPS. Great, but notice the increased 00:04:58.320 --> 00:05:04.400 latency. It's pretty minor, and even in twitchy firstperson shooters, it could 00:05:02.320 --> 00:05:09.199 be imperceptible to many. Sounds like a decent trade-off, right? But our base 00:05:06.400 --> 00:05:13.360 frame rate was already a solid 98 frames per second. Frame generation is what we 00:05:11.280 --> 00:05:17.520 in the MTG community call a win more card. It only really works well if 00:05:15.520 --> 00:05:22.479 you're already in a good situation. Here, watch what happens if we have a 00:05:19.600 --> 00:05:27.919 low starting frame rate. At 1440p, our base frame rate averages around 63 FPS. 00:05:25.360 --> 00:05:32.240 Good, but nowhere near the almost 100 we just came from. Rendering natively, 00:05:29.759 --> 00:05:36.000 we're starting with the same 36 frames of latency that we saw at 1080p with 00:05:34.240 --> 00:05:41.199 frame gen set to four in the previous clip. So then what happens when we add 00:05:38.880 --> 00:05:45.840 frame gen at 1440p? You're looking at MFG3X where we're getting an additional 00:05:43.840 --> 00:05:49.600 12 frames of latency even though our total frames per second or higher. Can 00:05:47.680 --> 00:05:54.320 an average person feel it? Not necessarily. But some of you out there 00:05:52.000 --> 00:05:58.960 are going to be sensitive to it. Let's be clear, we don't hate frame 00:05:56.479 --> 00:06:02.880 generation. And to NVIDIA's credit, they continue to improve the tech with every 00:06:01.120 --> 00:06:08.639 single generation just like they did with DLSS. What we do hate is how NVIDIA 00:06:06.400 --> 00:06:12.560 uses these features to mislead consumers by posting unfair representations of 00:06:10.880 --> 00:06:18.120 video card performance, which is especially bad because not every game 00:06:14.479 --> 00:06:18.120 supports those features. 00:06:21.360 --> 00:06:27.759 Sure, you could use an inexpensive thirdparty tool like Lothless Scaling, 00:06:25.440 --> 00:06:31.520 but even with the bells and whistles, if you buy this card, you're getting 00:06:29.440 --> 00:06:34.880 fleeced. Unlike if you buy the new transparent screwdriver from 00:06:32.639 --> 00:06:39.280 ltstore.com, it's a clear winner when it comes to nifty tools. Let's talk price 00:06:37.600 --> 00:06:45.199 and value. Starting with some of our competitors, AMD's 9060 XT8 GB, which is 00:06:42.720 --> 00:06:53.039 at the top of our 1080p Geomine results, costs us $2.32 per frame. The B580, $242 00:06:50.080 --> 00:06:57.840 is very nice. Then we get to the 5050 at $265. 00:06:54.639 --> 00:07:00.720 But it gets worse. The RTX5060, 00:06:57.840 --> 00:07:06.160 which costs around $299, just 50 bucks more than the 50/50, beats the 50/50 in 00:07:03.599 --> 00:07:09.599 value at $2.52. These aren't massive swings in value 00:07:07.680 --> 00:07:13.919 when you're looking at averages, but it still shows what a terrible deal the 00:07:11.440 --> 00:07:18.400 50/50 is. A huge part of why you get such better value with the more 00:07:15.599 --> 00:07:23.599 expensive card is that for $50 more, you're getting literally 50% more 00:07:20.639 --> 00:07:27.520 hardware. The 5060 has 50% more CUDA cores, 50% more texture processing 00:07:25.759 --> 00:07:31.919 clusters, 50% more streaming multipprocessors, 50% more tensor cores, 00:07:29.919 --> 00:07:37.360 50% more RT cores, 50% more texture units, and 50% more ROP units than the 00:07:34.319 --> 00:07:41.039 5050. And while this card is on the same 00:07:37.360 --> 00:07:44.160 128 bit bus, it's using GDDR6 instead of 00:07:41.039 --> 00:07:46.240 seven. Why? Because NVIDIA says G6 is 00:07:44.160 --> 00:07:50.800 best for desktops and the more power efficient G7. Well, that's that's best 00:07:48.240 --> 00:07:54.560 for laptops. Cool. So, then why not put the superefficient stuff in the one card 00:07:52.880 --> 00:07:59.039 in the desktop lineup that might actually want an extra low TDP? I just I 00:07:57.280 --> 00:08:04.000 don't get it. I quit. I can't I just can't do it anymore. >> Productivity. If there's anything 00:08:01.759 --> 00:08:09.520 positive to say about the 50/50, it's that it blends pretty okay. Sure, it 00:08:06.639 --> 00:08:14.319 renders our scene slower than the 4060 and hardly faster than a 3060, but sub 2 00:08:12.479 --> 00:08:18.400 minutes ain't half bad. We see similar results in Puget Bench's Premiere Pro 00:08:16.000 --> 00:08:23.680 benchmark, where our 50/50 is clearly pulling ahead and only overshadowed by 00:08:20.479 --> 00:08:25.680 the 5060. The Arc B580 puts up a good 00:08:23.680 --> 00:08:30.639 fight both here and in Puget Bench's Photoshop benchmarks. But then that card 00:08:28.000 --> 00:08:35.440 gets a big fat DNF with Puget's Da Vinci Resolve test, allowing NVIDIA to get a 00:08:32.719 --> 00:08:39.919 couple of blurry W's. Despite NVIDIA's obsession with AI, they failed to make 00:08:37.440 --> 00:08:45.120 the 5050s AI performance anything but underwhelming. In Proon AI image 00:08:42.479 --> 00:08:49.680 generation, the 50/50 is barely beating the 3060 again. And in text generation, 00:08:47.760 --> 00:08:55.279 the 50/50 is nothing to write home about. See what we did there? My biggest 00:08:52.399 --> 00:09:00.640 takeaway here is how great the ARC B580 is. Look at how much it crushes even our 00:08:57.279 --> 00:09:04.240 5060. Just look at what extra VRAMm 00:09:00.640 --> 00:09:06.080 NVIDIA improper support AMD can do. If 00:09:04.240 --> 00:09:11.279 Intel can figure it out as the new kid showing up with open vinyl, nobody has a 00:09:08.480 --> 00:09:15.839 good excuse. In summary, the RTX 5050 only makes sense as a product designed 00:09:13.279 --> 00:09:19.760 to make the RTX 5060 look good. And if you saw that review, you know that it 00:09:17.680 --> 00:09:23.920 does not look good. We can't in good conscience recommend the 50/50. It flat 00:09:22.000 --> 00:09:27.360 out sucks. And there are so many options that we've mentioned in this video that 00:09:25.519 --> 00:09:30.320 exist and we'll have all of those linked in the video description for you to 00:09:28.720 --> 00:09:35.279 check out. There's a saying that there are no bad products, just bad prices. 00:09:32.720 --> 00:09:40.959 And in RTX50 series fashion, the price is bad. But, you know, if nobody buys it 00:09:39.279 --> 00:09:46.000 and it gets discounted to maybe like $200, 00:09:43.519 --> 00:09:48.399 maybe then I could finally segue to our sponsor, 00:09:46.800 --> 00:09:51.440 >> Dbrand. We all know that to prevent accidents, you should use proper 00:09:50.000 --> 00:09:55.600 protection. So, if you keep dropping stuff like me, consider protecting what 00:09:53.920 --> 00:09:59.920 is most likely your most precious possession, your phone with Dbrand. 00:09:57.760 --> 00:10:03.920 Their ghost cases are ultra durable, super scratch resistant, designed to 00:10:01.920 --> 00:10:07.200 never yellow, and they're grippy for a nice hand feel. Plus, if you add on one 00:10:05.600 --> 00:10:11.519 of their Prism screen protectors, you'll have 360° protection. They are stupid 00:10:09.680 --> 00:10:15.440 simple to apply, and the whole process, like some things in life, only last 00:10:13.519 --> 00:10:19.440 seconds. And also consider picking up a Glow Circuit skin to add a bit of style 00:10:17.600 --> 00:10:23.920 and illumination to your nighttime activities. Smash our link in the 00:10:21.120 --> 00:10:28.000 description. Oh, I get it. To learn more today, thanks for watching. If you like 00:10:25.760 --> 00:10:32.320 this video, go check out our 960 XT review. The 16 GB model is really the 00:10:30.399 --> 00:10:36.880 way to go, but if it's too far out of your price range, then a B580 or even an 00:10:34.160 --> 00:10:41.360 8 gig 960 XT would be better than garbage. Just dog.