WEBVTT 00:00:00.000 --> 00:00:03.000 Intel has gone through a bit of a rough patch recently. 00:00:03.000 --> 00:00:09.200 One's so bad, maybe we should be calling it a rough mile. Degradation found in their 13th and 14th gen processors 00:00:09.200 --> 00:00:12.800 was bad enough. Then there were the mass layoffs and belt tightening. 00:00:12.800 --> 00:00:15.920 Then they hoped that the Core Ultra 200 series, 00:00:15.920 --> 00:00:21.120 codenamed Aerolake, would make a splash. But unfortunately, despite some interesting changes 00:00:21.120 --> 00:00:24.600 and impressive efficiency gains, it was more of a belly flop. 00:00:24.600 --> 00:00:28.280 So today they made two major announcements. 00:00:28.280 --> 00:00:31.360 Their CEO is out, like what? 00:00:31.360 --> 00:00:36.640 And also their Arc desktop GPUs are finally getting a follow up. 00:00:36.640 --> 00:00:40.040 To say then that they need Arc Battlemage to be a win 00:00:40.040 --> 00:00:44.720 would be a gross understatement. But with the announcements made today, 00:00:44.720 --> 00:00:48.280 I am cautiously optimistic. If the numbers are to be believed, 00:00:48.280 --> 00:00:51.320 it is looking great for budget gamers. 00:00:51.320 --> 00:00:56.320 But if this is yet another failure, I think 2024 could go down 00:00:56.320 --> 00:00:59.760 as the single worst year in Intel's existence. 00:00:59.760 --> 00:01:05.080 So I thought it was only fitting then that this segue, this one that I'm bragging out right now, 00:01:05.080 --> 00:01:08.560 should be remembered as the worst regged segue, 00:01:08.560 --> 00:01:13.240 reg, Black Friday might've just passed, ever on this channel. 00:01:13.240 --> 00:01:25.640 Editor cut. Thank you. There's a lot of ground for us to cover, 00:01:25.640 --> 00:01:30.240 but in a nutshell, you'll find nuts. And in this video, you'll find mostly a recap 00:01:30.240 --> 00:01:33.480 of the Battlemage announcement. Intel has confirmed that we are getting 00:01:33.480 --> 00:01:37.040 two new graphics cards, the B580 in mid-December 00:01:37.040 --> 00:01:40.360 and the B570 about a month later in mid-January. 00:01:40.360 --> 00:01:45.160 More on the architectural and software improvements in a bit, but these cards are 12 gig 00:01:45.160 --> 00:01:48.800 and 10 gig cards respectively, meaning that you should be very comfortable 00:01:48.800 --> 00:01:51.960 gaming at 1080p. And Intel is even going as far as to say 00:01:51.960 --> 00:01:56.640 that the B580 could be the heart of a 1440p gaming machine 00:01:56.640 --> 00:02:01.320 and all of this for less money than NVIDIA's RTX 4060, 00:02:01.320 --> 00:02:06.040 which only has eight gigs of VRAM and is marketed as a 1080p card. 00:02:06.040 --> 00:02:09.840 The pricing, I gotta say, it looks great, 00:02:09.840 --> 00:02:14.480 but you don't have to look far for the motivation for this hyper-aggressive pricing. 00:02:14.480 --> 00:02:17.600 ARC Alchemist has barely even registered 00:02:17.600 --> 00:02:21.720 on the Steam hardware survey, it's been out for two long years. 00:02:21.720 --> 00:02:26.240 So what has changed then to give this new card a chance to compete? 00:02:26.240 --> 00:02:30.280 Well, kind of everything. We first met Battle Mage alongside Intel's 00:02:30.280 --> 00:02:35.200 Lunar Lake mobile platform earlier this year. The big focus, efficiency. 00:02:35.200 --> 00:02:39.280 Their previous XE-based graphics cards required massive amounts of power 00:02:39.280 --> 00:02:42.720 and large silicon dies for meager performance. 00:02:42.720 --> 00:02:47.400 And their changes focus on improving core utilization, enabling smarter work distribution 00:02:47.400 --> 00:02:50.440 between those cores and reducing software overhead. 00:02:50.440 --> 00:02:53.800 They've made vast improvements to pretty much every part of the pipeline. 00:02:53.800 --> 00:02:57.800 Improved vertex fetching, mesh shading, out of order sampling, blending throughput, 00:02:57.800 --> 00:03:02.760 and many more terms that, quite honestly, I don't expect the average gamer to care about, 00:03:02.760 --> 00:03:07.560 but what you might care about is ray tracing. Each ray tracing unit is larger 00:03:07.560 --> 00:03:12.760 and can now handle 1.5x or two times more calculations further improving 00:03:12.760 --> 00:03:16.600 the already decent RT performance of XE graphics. 00:03:16.600 --> 00:03:20.240 And thanks to the B580's 12 gigs of GDDR6, 00:03:20.280 --> 00:03:25.160 not 6x, sadly, you are less likely to find your ray tracing performance bottlenecked 00:03:25.160 --> 00:03:28.640 with Intel claiming a 64% frame rate improvement 00:03:28.640 --> 00:03:32.440 over the RTX 4060 in one cherry-picked example 00:03:32.440 --> 00:03:35.720 where the B580 still runs below 60 frames per second, 00:03:35.720 --> 00:03:42.160 making that argument pretty questionable lab. But AI, XE Matrix extensions, 00:03:42.160 --> 00:03:47.000 which can be described kind of as the tensor cores of Intel GPUs are specifically designed 00:03:47.000 --> 00:03:51.200 to maximize performance with the matrix math that is typical of AI workloads. 00:03:51.200 --> 00:03:56.440 And they got some nice juice here. XMX was already pretty impressive, 00:03:56.440 --> 00:04:01.880 albeit under-supported on their server hardware. But here we get upgrades to support more data types 00:04:01.880 --> 00:04:05.240 and we get improved clock speeds, which could be a huge boon for people 00:04:05.240 --> 00:04:09.480 who are interested in running local models on the cheap. Intel is so confident in this 00:04:09.480 --> 00:04:13.200 that they're providing a front end for generating your own AI images. 00:04:13.200 --> 00:04:15.800 Maybe they could use it to generate one like this. 00:04:16.920 --> 00:04:20.080 I kid, I kid. All of these improvements and features 00:04:20.080 --> 00:04:24.040 are coming on a card that will draw just 8.5% more power 00:04:24.040 --> 00:04:28.400 than its namesake predecessor. It's still a lot more power than the RTX 4060, 00:04:28.400 --> 00:04:32.320 but is an impressive improvement in performance per watt nonetheless. 00:04:32.320 --> 00:04:36.880 But we always gotta keep in mind, these results are cherry-picked samples from Intel. 00:04:36.880 --> 00:04:40.560 So as always, I would encourage you guys to wait for our full review. 00:04:40.560 --> 00:04:46.920 Speaking of sampling, super sampling. Launching alongside this new discrete card is XCSS2, 00:04:46.920 --> 00:04:50.400 which is Team Blue's answer to DLSS3 plus Framegen. 00:04:50.400 --> 00:04:55.080 It packs a suite of AI-enhanced features. Super sampling to improve performance 00:04:55.080 --> 00:05:00.200 by rendering the game at a lower resolution, then using AI to upscale it to your display, 00:05:00.200 --> 00:05:05.000 combined with AI-powered frame generation that will add fake frames based on both 00:05:05.000 --> 00:05:08.400 in-engine vector information and optical flow information. 00:05:08.400 --> 00:05:12.800 To deal with the added latency of trying to add fake frames, a latency reduction solution 00:05:12.800 --> 00:05:16.400 that is plainly titled XC low latency until claims we can expect 00:05:16.400 --> 00:05:19.520 a 45% latency reduction over native rendering. 00:05:19.520 --> 00:05:24.120 Furthermore, they're bringing XCSS support to Vulkan and DirectX 11, allowing developers 00:05:24.120 --> 00:05:27.440 to more easily implement the tech in existing and future games. 00:05:27.440 --> 00:05:33.600 Sounds promising, but naturally, all of the caveats around the image quality of DLSS and FSR 00:05:33.600 --> 00:05:38.000 still apply here. Tell me, see it, maybe it's perfect, but I doubt it. 00:05:38.000 --> 00:05:42.920 But what about the price? I keep saying it's good, but I haven't said what it is. 00:05:42.920 --> 00:05:47.440 Well, you're gonna have to wait for our review to see where this card actually lands 00:05:47.440 --> 00:05:50.760 when it comes to 1080p and 1440p performance. 00:05:50.760 --> 00:05:55.320 But what we know now is at $249. 00:05:55.320 --> 00:05:58.760 I am, well, I'm extremely excited. 00:05:58.760 --> 00:06:02.320 Both of the incumbent players have treated the budget gaming segment 00:06:02.320 --> 00:06:05.440 like it isn't even worth their time and attention. 00:06:05.440 --> 00:06:11.480 But Intel looks dead set to bring back the capable $500 gaming box. 00:06:11.600 --> 00:06:14.840 And this is the best outcome that I could have hoped for 00:06:14.840 --> 00:06:20.440 back when they threw their hat in the ring. Their LTT hat pro, lttstore.com. 00:06:20.440 --> 00:06:24.040 Of course, we should be taking all of this with a grain of salt, so big you could choke on it. 00:06:24.040 --> 00:06:27.920 But what Intel is claiming is 26% average uplift 00:06:27.920 --> 00:06:32.840 in raster performance per dollar over the RX7600 at 1440p Ultra, 00:06:32.840 --> 00:06:37.960 which is across 47 games on a card that is $20 cheaper at MSRP. 00:06:37.960 --> 00:06:41.920 And then if we look at Ray Tracing, which is one of Team Red's biggest weak points 00:06:41.920 --> 00:06:47.200 when compared to NVIDIA, Team Blue comes out 37 points ahead of that same card. 00:06:47.200 --> 00:06:53.160 There's less data points here with only nine games tested, but that is still very hopefully impressive. 00:06:53.160 --> 00:06:58.160 And at $299, the RTX 4060 is allegedly getting slaughtered 00:06:59.000 --> 00:07:03.960 by Intel's new B580. That NVIDIA card costs about 20% more 00:07:03.960 --> 00:07:08.880 while delivering allegedly 24% less raster performance 00:07:08.880 --> 00:07:12.200 per dollar and 20% less Ray Tracing performance 00:07:12.200 --> 00:07:16.760 per dollar at 1440p. So it looks like while the CPU team at Intel 00:07:16.760 --> 00:07:23.600 was busy trying to do something great, the GPU division may have been busy succeeding, allegedly. 00:07:23.600 --> 00:07:27.560 I can't wait to start testing these cards and finding out where they actually land 00:07:27.560 --> 00:07:31.040 both against the competition and against themselves. 00:07:31.040 --> 00:07:34.880 This card should crush it in just about every way. 00:07:34.880 --> 00:07:40.560 Yes, the memory bus width is substantially lower, but we've got a lot more VRAM and much higher clock speeds 00:07:40.560 --> 00:07:45.800 to help make 1440p gaming possible. They're also launching alongside years of groundwork 00:07:45.800 --> 00:07:48.960 that was laid by the driver team that took the original arc cards 00:07:48.960 --> 00:07:54.860 from the stuttery mess at launch to basically playable across the majority of PC games. 00:07:54.860 --> 00:07:58.560 Now, when we were discussing the leaks and the battle mage launch internally, 00:07:58.560 --> 00:08:04.120 we all agreed that this thing needed to be no more than $250 and at minimum, 00:08:04.120 --> 00:08:07.440 it had to be on par with an RTX 4060 for performance, 00:08:07.440 --> 00:08:11.240 for it to even matter. Anything more expensive or less performant 00:08:11.240 --> 00:08:15.680 and they would have shot themselves in the foot moments before the race even began. 00:08:15.680 --> 00:08:19.280 Well, Intel, battle mage and XC2 graphics, 00:08:19.280 --> 00:08:22.800 they might be a little late to the party, but if these slides are to be believed, 00:08:22.800 --> 00:08:27.840 they were just busy stopping for more beverages and guests so they could keep the party going. 00:08:27.840 --> 00:08:31.000 Their timing, also kind of good in my opinion. 00:08:31.000 --> 00:08:37.040 It could be months before we see something like an RTX 5060 from NVIDIA and an RX 8600 00:08:37.240 --> 00:08:41.440 or whatever's next from AMD. Now, there's absolutely no telling how much better they might be, 00:08:41.440 --> 00:08:44.680 but even a 20% improvement at the same price 00:08:44.680 --> 00:08:49.400 could still keep Intel on top of that lowest rung on the ladder. 00:08:49.400 --> 00:08:53.400 So the way that I see it, if we have to sacrifice Pat Gelsinger 00:08:53.400 --> 00:08:58.040 to summon our dark magician, well then that's just the heart of the cards. 00:08:58.040 --> 00:09:03.160 Now I summon Pot of Greed, which lets me give you three segues to our sponsor. 00:09:03.160 --> 00:09:06.600 Thanks for watching. If you guys liked this video, go check out our Intel Arc livestream 00:09:06.680 --> 00:09:10.520 back when they first launched. It's, um, has improved a lot.