1 00:00:00,000 --> 00:00:08,900 I can't believe it. After NVIDIA and AMD just completely abandoned the sub-$300 price point, or gave us intentionally 2 00:00:08,900 --> 00:00:12,640 nerfed options, Intel has finally done it. 3 00:00:12,640 --> 00:00:17,340 They released the first good budget GPU in over five years. 4 00:00:17,340 --> 00:00:24,260 It doesn't cut any corners. It's got 12 gigs of VRAM, which means that new games like Indiana Jones will run properly, 5 00:00:24,260 --> 00:00:29,540 it's relatively efficient, and best of all, it's got raw performance that beats the RTX 6 00:00:29,540 --> 00:00:33,980 4060 handily at both 1080p and 1440p. 7 00:00:33,980 --> 00:00:37,420 If I had scripted Intel's press conference for this thing, it probably would have gone 8 00:00:37,420 --> 00:00:42,860 something like this. It's been a difficult journey to get here for Team Blue. 9 00:00:42,860 --> 00:00:47,420 Their first generation had famously buggy drivers, with some games running poorly, or 10 00:00:47,420 --> 00:00:52,140 even not at all, which led to such low sales figures that the department has been the subject 11 00:00:52,140 --> 00:00:58,740 of near constant speculation that it would be cut. But believe it or not, they have squashed most of the bugs. 12 00:00:58,740 --> 00:01:02,940 Most of them. We'll talk about that later. First, the TLDR. 13 00:01:02,940 --> 00:01:07,660 Last time, we asked viewers to consider buying Intel Arc because we were desperate for them 14 00:01:07,660 --> 00:01:10,900 to emerge as a disruptor in the GPU duopoly. 15 00:01:10,900 --> 00:01:16,140 This time, we're asking you to consider Intel Arc because it's freaking awesome to the point 16 00:01:16,140 --> 00:01:22,700 where I would say that it is the only choice for gamers who want to spend $450 to $700 on 17 00:01:22,700 --> 00:01:28,240 a tower, or for those folks who have waited so patiently for an upgrade to their faithful 18 00:01:28,240 --> 00:01:31,580 GTX 1660 or 10 series. 19 00:01:31,580 --> 00:01:36,780 Let's start at 1080p, where AMD and NVIDIA are more than happy to gouge you for the luxury 20 00:01:36,780 --> 00:01:40,140 of being able to play. First stop, Night City. 21 00:01:40,140 --> 00:01:45,900 In Cyberpunk, the B580 makes an impressive debut, running neck and neck with the 4060 22 00:01:45,900 --> 00:01:51,020 Ti 16 gig, a card that retails for $450. 23 00:01:51,020 --> 00:01:55,580 You heard me, this one is 80% more expensive. 24 00:01:55,580 --> 00:02:03,140 And check out the B580's direct competition, the $300 4060 and the $250 RX7600. 25 00:02:03,140 --> 00:02:10,620 They're behind by nearly 15%. And Intel's lead increases in Red Dead Redemption 2, where their improvements to Vulcan support 26 00:02:10,620 --> 00:02:15,460 have lifted the fortunes of even their last generation cards, and brought the B580 in 27 00:02:15,460 --> 00:02:21,300 line with the 7700XT, a $400 GPU. 28 00:02:21,300 --> 00:02:29,740 This kind of utter dominance, though, isn't across the board. In The Last of Us Part 1, the B580 falls back in line, but then even this poor result has 29 00:02:29,740 --> 00:02:34,140 them beating NVIDIA and AMD's current generation price competitors. 30 00:02:34,220 --> 00:02:39,620 In F123, AMD takes the lead, but they've historically performed really well on the track. 31 00:02:39,620 --> 00:02:45,060 And in the old gold shadow of the Tomb Raider, the B580 barely inches past the RTX 4060 in 32 00:02:45,060 --> 00:02:50,460 average FPS and loses in 1% lows, which we classify as a loss. 33 00:02:50,460 --> 00:02:57,300 In Returnal, the B580 is neck and neck with the 4060 Ti and the 6700XT, only taking a 34 00:02:57,300 --> 00:03:03,780 clear L from the 7700XT, which, again, I remind you, is a $400 card. 35 00:03:03,780 --> 00:03:08,460 And then finally in Atomic Heart, the B580's sales lose some of their glorious wind, but 36 00:03:08,460 --> 00:03:13,380 hey, the most expensive card on the charts was bound to win at least one game, wasn't 37 00:03:13,380 --> 00:03:17,660 it? Looking at the overall picture, Intel, you've outdone yourself. 38 00:03:17,660 --> 00:03:23,660 You've pulled off a commending generational uplift of 55% and beaten your old flagship 39 00:03:23,660 --> 00:03:31,820 even by more than 20%. You've fallen shy of the 4060 Ti, but the bottom line is, if gamers out there are looking 40 00:03:31,820 --> 00:03:37,580 to upgrade their aging GPUs for high refresh rate 1080p gaming, you've given us something 41 00:03:37,580 --> 00:03:45,140 to finally recommend to them. Not just point at the least worst thing, but to recommend to them. 42 00:03:45,140 --> 00:03:49,980 This is a truly incredible achievement for Intel, and they did a lot to make it happen. 43 00:03:50,980 --> 00:03:56,860 Interestingly, they don't call it a GPU, leaves no stone unturned in the search for 44 00:03:56,860 --> 00:04:03,620 performance improvements. They've jumped to a new process node and made several architectural and driver overhauls 45 00:04:03,620 --> 00:04:10,460 that Intel claims gave them a 70% improvement in performance per core and a 50% energy improvement 46 00:04:10,460 --> 00:04:14,740 over last gen. Which makes me wonder, why did they hold back? 47 00:04:14,740 --> 00:04:22,340 If 20XE2 cores is good, then 32 cores on a B770 would be better. 48 00:04:22,340 --> 00:04:27,620 Speaking of bigger, nearly 37% of Steam users are gaming at a resolution that is greater 49 00:04:27,620 --> 00:04:33,580 than 1080p these days, and that number keeps growing as 1440p and 4K monitors continue 50 00:04:33,580 --> 00:04:38,060 to come down in price. So can Intel compete at 1440p as well? 51 00:04:38,060 --> 00:04:46,900 The short answer is yes. Across our suite of benchmarks, the B580 overtakes the RX 6700 XT and narrows the gap 52 00:04:46,900 --> 00:04:50,540 with the 4060 Ti to just a few percent. 53 00:04:50,540 --> 00:04:55,820 As for Intel's promised 10% lead over the 4060, well, it appears they were actually 54 00:04:55,820 --> 00:05:02,980 being modest. Across our game selection, we found a lead of 20%, though it is worth noting that this 55 00:05:02,980 --> 00:05:06,140 will vary depending on game selection. 56 00:05:06,140 --> 00:05:10,780 And with the famously VRAM hungry The Last of Us Part 1, the B580 holds its position 57 00:05:10,780 --> 00:05:17,580 on our chart, embarrassing NVIDIA again, with the B580's 1% lows besting the 4060's average 58 00:05:17,580 --> 00:05:22,380 FPS. Shadow of the Tomb Raider demonstrates just what an incredible generational leap Intel 59 00:05:22,380 --> 00:05:28,620 has pulled off here. And it seems like something about Returnal really likes Intel, because all of our team 60 00:05:28,620 --> 00:05:36,100 blue cards get a nice little bump in the rankings. I'm kind of running out of ways to say Intel's doing well, so in Red Dead Redemption 2, Intel 61 00:05:36,820 --> 00:05:44,780 did bloop-a-looby! And in Cyberpunk, they blomp-anated the competition, while making it glaringly obvious just how 62 00:05:44,780 --> 00:05:51,940 overdue the 1060 and the 1650 were for a valid, modern 1440p upgrade. 63 00:05:51,940 --> 00:05:56,740 With strong, if not mind-blowing performance in both Atomic Heart and F123, it's clear 64 00:05:56,740 --> 00:06:01,540 that Intel has achieved what they set out to do, made a killer value 1080p gaming card 65 00:06:01,540 --> 00:06:08,780 that's also capable of 1440p. There are some caveats, even Intel admits that they don't win in every game. 66 00:06:08,780 --> 00:06:14,340 And a motherboard with support for resizable bar is mandatory for ArcGPUs. 67 00:06:14,340 --> 00:06:19,580 And the B580 isn't a top-of-the-line benchmark buster, so if you have a higher-tier older 68 00:06:19,580 --> 00:06:24,420 card like a 3060 Ti, you shouldn't feel compelled to upgrade here. 69 00:06:24,420 --> 00:06:30,140 But as far as downsides go, those are pretty minor, and there's more to like than just 70 00:06:30,140 --> 00:06:36,620 the raw performance. Let's talk about ray tracing, which is a little more relevant these days all of a sudden 71 00:06:36,620 --> 00:06:42,900 than it used to be. Especially when you consider that games are starting to list ray tracing as a minimum requirement. 72 00:06:42,900 --> 00:06:46,740 While we didn't have time to develop a test for our boy Indy, we did check out the RT 73 00:06:46,740 --> 00:06:52,380 performance in a few other games. Tracing those rays still results in a significant performance here, but NVIDIA's mature and 74 00:06:52,380 --> 00:06:57,780 well-supported RT tech allows them to beat the B580 in a big way in Atomic Heart at 1080p. 75 00:06:57,820 --> 00:07:02,940 In Returnal, the B580 catches right back up to the 4060 Ti, but falls back into the pack 76 00:07:02,940 --> 00:07:09,340 in F123. Overall, you can have a solid plus 60fps ray traced gaming experience, assuming you're 77 00:07:09,340 --> 00:07:13,900 willing to fiddle with the settings just a little bit. Speaking of which, ultra settings at 1440p. 78 00:07:13,900 --> 00:07:18,140 That's beyond the reach of pretty much every card we've tested today, the entire lot failing 79 00:07:18,140 --> 00:07:27,700 to break 60fps average in any title. But the extra VRAM on the B580 earns it a sizable lead over the 4060 in Returnal and in F123. 80 00:07:27,780 --> 00:07:31,580 Even if it can't pull out a win in Atomic Heart. AMD's RX7600? 81 00:07:31,580 --> 00:07:35,900 Oof. It's basically crying in the corner trying to figure out why it's even here. 82 00:07:35,900 --> 00:07:39,980 And if you're wondering where the flagship ray tracing title Cyberpunk is, well, we had 83 00:07:39,980 --> 00:07:46,500 some Intel problems. It seems to not like running at ultra ray tracing settings on CPUs with 3DV cache. 84 00:07:46,500 --> 00:07:50,540 But before you start haranguing Intel about game compatibility, they are aware, and they 85 00:07:50,540 --> 00:07:54,140 are working on it. Look mom, we made the patch notes. 86 00:07:54,140 --> 00:07:57,700 And since the launch of ARC, compatibility has massively improved. 87 00:07:57,700 --> 00:08:03,620 Hardware box recently showed in a test of 250 games that 233 were completely playable. 88 00:08:03,620 --> 00:08:08,460 There were still issues and some of those were in major titles that took a lot longer 89 00:08:08,460 --> 00:08:14,060 to fix than we would have liked. But hopefully Intel will continue this upward trend in compatibility. 90 00:08:14,060 --> 00:08:17,740 Now knowing the price and the 1440p performance numbers, you wouldn't expect this to be 91 00:08:17,740 --> 00:08:22,100 a 4k gaming card and you'd be right. It's not a 4k card. 92 00:08:22,100 --> 00:08:25,940 Even in low settings in Cyberpunk, we don't see great frame rates. 93 00:08:25,940 --> 00:08:30,980 At least not natively. But Intel's got some more tricks up their sleeves. 94 00:08:30,980 --> 00:08:35,380 Desperate to not completely miss the boat on the AI boom, Intel has packed some juicy 95 00:08:35,380 --> 00:08:38,580 AI into the B580 in order to up its performance. 96 00:08:38,580 --> 00:08:42,140 Oh, I guess if I'm going to talk about AI, I might as well dress the part with my tech 97 00:08:42,140 --> 00:08:45,740 bro vest from LTTstore.com. 98 00:08:45,740 --> 00:08:49,820 Intel's AI-powered render enhancement sauce comes in the form of XCSS2. 99 00:08:49,940 --> 00:08:57,580 XCSS2 has three main parts. An AI upscaler that renders the game at a lower resolution than uses AI to upscale 100 00:08:57,580 --> 00:09:03,300 to display resolution, XCSS frame generation, which creates extra frames by interpolating 101 00:09:03,300 --> 00:09:08,260 visual and in-game vector data, and then taking the two frames and making a middle 102 00:09:08,260 --> 00:09:12,780 point to enhance animation smoothness, and a latency reduction component that helps to 103 00:09:12,780 --> 00:09:17,100 cancel out some of the extra latency from the aforementioned frame generation. 104 00:09:17,100 --> 00:09:23,020 In fewer words, they've invented NVIDIA DLSS, NVIDIA FrameGen, and NVIDIA Reflex. 105 00:09:23,020 --> 00:09:28,540 And what's nice is that unlike the time that AMD invented NVIDIA Reflex with Antileg Plus, 106 00:09:28,540 --> 00:09:32,660 Intel has implemented this in a way that the game developer bakes it into the game, meaning 107 00:09:32,660 --> 00:09:38,980 you won't get permaband for turning it on. We're not going to be doing a deep dive into image quality at this time, but we will be 108 00:09:38,980 --> 00:09:43,580 looking at performance and we will make some anecdotal remarks about image quality. 109 00:09:43,580 --> 00:09:49,020 XCSS has already proven itself to be a solid upscaling technology, and XCSS 2 successfully 110 00:09:49,020 --> 00:09:56,100 builds upon that foundation. When turning on supersampling to its highest quality level in Cyberpunk 2077 at 1440p, Intel 111 00:09:56,100 --> 00:10:00,780 wins an FPS thanks to the raw performance advantage they already had, but their super 112 00:10:00,780 --> 00:10:03,980 resolution solution doesn't scale as well. 113 00:10:03,980 --> 00:10:08,700 NVIDIA sees a performance uplift of 42 and 32% in the lows and averages respectively, 114 00:10:08,700 --> 00:10:12,400 while Intel only gets around a 20% boost. 115 00:10:12,400 --> 00:10:19,520 We see the same trend at 4K. The B580 doesn't gain as much performance from XCSS as NVIDIA does from DLSS, but wins 116 00:10:19,520 --> 00:10:25,440 by just being a more powerful card. The 4060, though, doesn't even hit 30 FPS here. 117 00:10:25,440 --> 00:10:30,920 But the story changes a bit in F124, the only title currently available for us to test XCSS 118 00:10:30,920 --> 00:10:34,120 2's Framegen, and this is where things get wacky. 119 00:10:34,120 --> 00:10:39,400 In this game, the upscalers are better matched, but in Framegen, it's a blowout. 120 00:10:39,400 --> 00:10:44,120 Intel is introducing way more generated frames, with performance skyrocketing to 70% over 121 00:10:44,120 --> 00:10:49,760 the upscale result, and double over native rendering, and it is almost indistinguishable 122 00:10:49,760 --> 00:10:52,980 from native rendering unless you really know what to look for. 123 00:10:52,980 --> 00:10:56,640 And if you ignore those entire artifacts that arise from the upscaling, not the Framegen, 124 00:10:56,640 --> 00:11:00,600 it's weird. It has double the FPS with barely a drop in quality. 125 00:11:00,600 --> 00:11:05,320 Heck, you can even get over 100 FPS at 4K with XCSS turned on. 126 00:11:05,320 --> 00:11:09,000 Granted, Intel probably worked pretty closely with the devs to make sure that this launch 127 00:11:09,000 --> 00:11:14,400 title would leave a good impression, but a good impression is a good impression. 128 00:11:14,400 --> 00:11:18,160 While Intel pitched this as a gaming card first and foremost, they did include some 129 00:11:18,160 --> 00:11:22,360 other value adds, like their enhanced media engine that now supports more codecs than 130 00:11:22,360 --> 00:11:29,400 NVIDIA or AMD. In modern AV1 encoding, we can see a clear generational uplift, but Intel can't quite 131 00:11:29,400 --> 00:11:36,420 match the speed of NVIDIA's NVENC encoder. And it's worth noting here, by the way, that while AMD does compete on this chart, it does 132 00:11:36,420 --> 00:11:42,940 so by outputting the wrong resolution, 1920 by 1082, and yes, that is a hardware level 133 00:11:42,940 --> 00:11:50,420 problem. Yikes. In the more ubiquitous H.264, the B580 snatches the crown, although margins are pretty small 134 00:11:50,420 --> 00:11:56,820 across the board here. Now thanks to their optics rendering, Blender sees some good old fashioned NVIDIA domination, 135 00:11:56,820 --> 00:12:05,300 at least on the cards that have RT cores. And interestingly, the B580 performed significantly worse than Intel's last-gen cards. 136 00:12:05,980 --> 00:12:09,020 This is expected, but didn't really go into why. 137 00:12:09,020 --> 00:12:13,180 Our best guess then is that it's due to the relatively smaller number of XE cores on this 138 00:12:13,180 --> 00:12:19,180 card compared to the last gen. But that didn't slow it down in gaming, so... 139 00:12:19,180 --> 00:12:25,780 Yay gaming. Moving into our AI testing, NVIDIA's dominant position means that most AI software is developed 140 00:12:25,780 --> 00:12:30,500 with CUDA in mind, making it easy for folks at home to get up and running quickly. 141 00:12:30,500 --> 00:12:34,620 The good news is things are getting easier for AMD and Intel users. 142 00:12:34,620 --> 00:12:38,700 And if you want to run some computer vision, large language models, or image generation 143 00:12:38,700 --> 00:12:44,060 at home, you can probably get it working on your non-NVIDIA GPU in a few extra steps. 144 00:12:44,060 --> 00:12:48,260 And generationally, Intel has made a massive leap in performance, taking the lead in our 145 00:12:48,260 --> 00:12:52,220 stable diffusion image generation benchmark by a pretty big margin. 146 00:12:52,220 --> 00:12:56,380 In computer vision, it loses to the 4060, although just barely. 147 00:12:56,380 --> 00:13:02,540 So if you can get it running, it works great. As long as you don't mind using a bunch of power to generate a portrait of your aunt 148 00:13:02,620 --> 00:13:07,580 as an actual aunt. Speaking of power, we haven't even talked about power yet. 149 00:13:07,580 --> 00:13:12,420 The physical PCB of our card is dinky, and Intel takes advantage of this by implementing 150 00:13:12,420 --> 00:13:19,900 a new flow-through cooler design. Thermals are well under control with the hotspot on the B580, never passing 76 degrees in either 151 00:13:19,900 --> 00:13:25,860 synthetic or gaming workloads. The B580 is targeting a TDP of 190 watts. 152 00:13:25,860 --> 00:13:34,500 That's 25 watts more than the RX7600 and 4060 Ti, and 75 watts more than the 4060 non-Ti. 153 00:13:34,500 --> 00:13:39,100 So it's clear that even with all of their improvements, Intel does have a ways to go 154 00:13:39,100 --> 00:13:43,420 with NVIDIA pulling off a clean victory when it comes to performance per watt, even if 155 00:13:43,420 --> 00:13:46,980 NVIDIA doesn't seem to understand what performance per dollar is. 156 00:13:46,980 --> 00:13:52,940 The B580 uses just one PCIe 8-pin for power, so you won't need to upgrade your power supply. 157 00:13:52,940 --> 00:13:57,160 That is, unless you have a power supply that is less than 600 watts. 158 00:13:57,160 --> 00:14:02,340 Even though it should only draw a max of 225 watts with that power connector, we saw some 159 00:14:02,340 --> 00:14:06,620 transient spikes up to 241 watts in F1. 160 00:14:06,620 --> 00:14:11,620 That is significantly higher than Intel's advertised 190 watts. 161 00:14:11,620 --> 00:14:16,180 So in conclusion, maybe it's just because it's been so long since we've seen a great 162 00:14:16,180 --> 00:14:19,580 budget card, but man, I'm excited. 163 00:14:20,100 --> 00:14:25,740 How many times have I had to say, well, you could buy this shiny new thing, but it kind 164 00:14:25,740 --> 00:14:31,340 of sucks. And you'd be way better off going with something last gen or second hand or just not upgrading 165 00:14:31,340 --> 00:14:38,940 at all. Well, not today. I mean, eBay deals obviously still exist, but this is a new GPU that doesn't suck with 166 00:14:38,940 --> 00:14:44,900 modern features and it's worthy of celebration, even if Intel is a little late to the party. 167 00:14:44,900 --> 00:14:50,780 See, the thing is AMD and NVIDIA are poised to release their next generation of GPUs early 168 00:14:50,780 --> 00:14:55,340 next year, so it isn't out of the question for a couple of companies that have had their 169 00:14:55,340 --> 00:15:02,620 stock prices double or ten-uple to make a move to nuke Arc-B series from orbit. 170 00:15:02,620 --> 00:15:06,740 Also, the looming threat of tariffs could impact the affordability of these cards. 171 00:15:06,740 --> 00:15:11,460 But hey, it's going to be a pretty sweet ride up until that time. 172 00:15:11,460 --> 00:15:16,420 It's a great GPU that will give you great performance without destroying your wallet. 173 00:15:16,420 --> 00:15:20,220 If you guys enjoyed this video, go check out our review of the original Arc. 174 00:15:20,220 --> 00:15:25,060 The benchmarks and graphs aren't really relevant, other than to give you the context for how 175 00:15:25,060 --> 00:15:33,740 far we have come.