WEBVTT 00:00:00.000 --> 00:00:05.400 Do you remember that one time when Intel launched the world's first quad-core consumer CPU? 00:00:05.400 --> 00:00:09.680 And then 10 years later, was still doing the same thing? 00:00:09.680 --> 00:00:13.920 Okay, yes, yes. I know there were plenty of other innovations along the way. 00:00:13.920 --> 00:00:19.200 But that doesn't change the fact that Intel's long-term stagnation did create an opening 00:00:19.200 --> 00:00:25.720 for AMD to not only continue existing, but against all odds to close the technology gap, 00:00:25.720 --> 00:00:28.000 taking serious market share from a company 00:00:28.760 --> 00:00:34.320 up until then was functionally a monopoly in both the server and the mobile space. 00:00:34.320 --> 00:00:38.440 Now, that level of stagnation, that's not where AMD is today. 00:00:38.440 --> 00:00:41.760 But if there are new product announcements today 00:00:41.760 --> 00:00:45.720 or anything to go by, the warning signs are starting to show. 00:00:45.720 --> 00:00:49.480 In no particular order we've got, a darn near one-to-one rebrand 00:00:49.480 --> 00:00:53.800 of their last-generation laptop chips, a couple new entries in the incredible, 00:00:53.800 --> 00:00:57.200 but obscenely expensive Strix Halo chip family, 00:00:57.200 --> 00:01:00.200 a better-binned refresh of the 9800X3D 00:01:00.200 --> 00:01:06.400 called the 9850X3D, and some AI performance and usability improvements. 00:01:06.400 --> 00:01:12.040 I mean, none of these are bad things, but the complete lack of any pricing information 00:01:12.040 --> 00:01:15.800 is a little ominous and raises another uncomfortable bit of news 00:01:15.800 --> 00:01:18.800 that AMD didn't mention, 00:01:18.800 --> 00:01:22.360 the rumored price hikes that are coming on consumer GPUs. 00:01:22.360 --> 00:01:29.360 Let's talk through all of that. After an uncomfortable segue to our sponsor. 00:01:30.280 --> 00:01:34.240 As a wise man once said, when you can't innovate in engineering, 00:01:34.240 --> 00:01:37.960 innovate in marketing, the unwashed masses who buy our shit 00:01:37.960 --> 00:01:42.200 won't know the difference anyway. Okay, that's not entirely fair. 00:01:42.200 --> 00:01:47.960 To AMD's credit, there are some improvements in the new Ryzen AI 400 series. 00:01:47.960 --> 00:01:54.960 Everything highlighted in gold here is new, but the part that they leave off is how much new they are. 00:01:54.960 --> 00:01:59.160 Maximum boost clocks of 5.2 gigahertz on a mobile CPU. 00:01:59.160 --> 00:02:04.320 I mean, that's incredible. It's just that that's only 0.1 gigahertz more 00:02:04.320 --> 00:02:08.560 than the Ryzen AI 300 series, which wouldn't be an issue. 00:02:08.560 --> 00:02:13.480 I mean, clock speeds, not everything, except that they're also reusing the same mix 00:02:13.480 --> 00:02:16.600 of Zen 5 and Zen 5c cores. 00:02:16.600 --> 00:02:21.440 So when you're talking about mobile processors that are often limited by their cooling solutions 00:02:21.440 --> 00:02:26.840 and power delivery anyway, no amount of gold text is gonna make a 2% difference 00:02:26.840 --> 00:02:32.440 in boost clock noticeable day to day. The bump in LPDDR5 speed is welcome though, 00:02:32.440 --> 00:02:39.440 especially for applications that rely on the GPU core that is also faster, albeit just 7%, 00:02:39.440 --> 00:02:44.080 and it's not gonna hurt to be able to run local AI slightly faster. 00:02:44.080 --> 00:02:50.040 With all that said, the devil's in the details and while AMD's slide is technically true, 00:02:50.040 --> 00:02:54.560 it's only true for the top skew. And as we make our way down the lineup, 00:02:54.560 --> 00:02:57.720 the generational upgrades slowly fall away 00:02:57.720 --> 00:03:03.120 to make room for what appear in some cases to be generational downgrades. 00:03:03.120 --> 00:03:06.240 The Ryzen AI 7 445, for instance, 00:03:06.240 --> 00:03:13.400 gets lower clocks, less cash, and the same RAM speed compared to the Ryzen AI 5 340. 00:03:13.400 --> 00:03:16.760 And this comes at the same time that Intel's Panther Lake 00:03:16.760 --> 00:03:20.760 is rumored to deliver GPU performance gains of like 50%. 00:03:21.840 --> 00:03:25.600 That's the yikes. Next up is the 9850X3D, 00:03:25.600 --> 00:03:30.360 a truly impressive looking 5.6 gigahertz desktop chip 00:03:30.360 --> 00:03:37.080 that promises to once again steal AMD's gaming performance crown and give it to AMD. 00:03:37.080 --> 00:03:41.120 But again, probably not by as much as we'd hope 00:03:41.160 --> 00:03:45.800 or as much as was rumored. See, the internet's been ablaze with rumors 00:03:45.800 --> 00:03:50.800 of a 9950X3D2, a dual 3DV cache 16 core 200 watt monster 00:03:53.760 --> 00:03:58.160 with an alleged 192 megabytes of level three cache. 00:03:58.160 --> 00:04:04.520 But instead, all we got was a speedbind eight core. Once again, this looks like a really nice piece of kit. 00:04:04.520 --> 00:04:10.040 We get the same rated power as last gen along with up to 400 megahertz faster clock speeds 00:04:10.040 --> 00:04:16.840 than the old King, which is nice. But how does that translate to real world gaming performance? 00:04:16.840 --> 00:04:21.880 Well, it's kind of hard to say. See, rather than comparing their own chips directly, 00:04:21.880 --> 00:04:26.280 AMD instead compared their new hotness to Intel's Core Ultra 9 285K, 00:04:26.280 --> 00:04:31.280 which is kind of like a real boxer fighting Jake Paul. 00:04:31.360 --> 00:04:35.520 Fair enough AMD, it isn't really my place to tell you how to bully your competitors. 00:04:35.520 --> 00:04:39.480 And come think of it, it's probably just as well that gamers will need to come to me 00:04:39.480 --> 00:04:44.320 for their benchmarks anyway. Thanks. The Ryzen AI Max Plus announcement. 00:04:44.320 --> 00:04:48.760 Ooh, this one was shaping up to be a real rebranding snooze fest. 00:04:48.760 --> 00:04:53.760 I was sitting there thinking like, really, you guys are gonna remind me that Strix Halo can go on a desktop? 00:04:53.760 --> 00:04:58.240 I literally invested in a company that is doing that, framework investment disclosure. 00:04:58.240 --> 00:05:03.680 But then AMD was like, boom. We heard that you love astonishingly fast 00:05:03.680 --> 00:05:06.680 PlayStation five level graphics in a mobile chip, 00:05:06.680 --> 00:05:11.520 but don't wanna pay an extra $500 for the 16 core model. 00:05:11.520 --> 00:05:14.880 So we put the best graphics on an eight core 00:05:14.880 --> 00:05:18.720 and a 12 core. Man, I've been using the GPD Win 5 00:05:18.720 --> 00:05:23.480 for the last little while for gaming. This thing absolutely shreds Clare Obscure. 00:05:23.480 --> 00:05:26.840 And like epic details, no upscaling Tom Foolery. 00:05:26.840 --> 00:05:29.960 Like if I could save hundreds of dollars on CPU cores 00:05:29.960 --> 00:05:33.680 that I don't need, absolutely killer. 00:05:33.680 --> 00:05:36.760 Unfortunately, Strix Halo is killer for your wallet too, 00:05:36.760 --> 00:05:42.440 since even the entry level is crazy expensive. But hey, cool new product is a cool new product. 00:05:42.440 --> 00:05:46.960 AMD then talked for a while about running near cloud quality local AI models 00:05:46.960 --> 00:05:52.480 that save you from either the limitations of free AI chatbots or the cost of a subscription. 00:05:52.480 --> 00:05:57.840 The partner they showcased was Liquid AI who has scary call to talk about pricing buttons 00:05:57.840 --> 00:06:01.800 on their corporate licenses. But at this time does provide end user access 00:06:01.800 --> 00:06:05.640 to their models for free. So that's actually pretty neat. 00:06:05.640 --> 00:06:10.720 But then things got totally off the rails. I'm writing this based on my pre-briefing. 00:06:10.720 --> 00:06:14.720 So it is possible that the public presentation is a little bit different in the details. 00:06:14.720 --> 00:06:19.040 But after the local models, AMD shifted to talking about the idea 00:06:19.040 --> 00:06:22.560 of using iterate.ai to feed an AI agent, 00:06:22.560 --> 00:06:25.880 your pay stubs, bank information, tax returns, 00:06:25.880 --> 00:06:31.040 and all that kind of stuff. And then have it generate powerful financial advice, 00:06:31.080 --> 00:06:34.160 saving you the cost of hiring a financial planner, 00:06:34.160 --> 00:06:38.080 direct quote incoming, you can do it yourself. 00:06:38.080 --> 00:06:42.080 Now look, I'm not saying you should spend a bunch of money on a financial planner, 00:06:42.080 --> 00:06:47.440 and there are clear privacy benefits to running workloads like this locally. 00:06:47.440 --> 00:06:52.040 But, and to be clear, this is financial advice. 00:06:52.040 --> 00:06:57.840 Do not trust an AI to give you financial advice without also having it thoroughly vetted, 00:06:57.840 --> 00:07:01.840 at least not for now. AMD then talked for a while 00:07:01.840 --> 00:07:06.360 about the strides that they've made recently on Rock'em with respect to performance improvements 00:07:06.360 --> 00:07:11.040 and Windows compatibility, and they announced a handful of new features, including, 00:07:11.040 --> 00:07:14.320 oh, okay, so it's just integration with Comfy UI 00:07:14.320 --> 00:07:17.480 and a simplified install method? To be clear, I'm not mad. 00:07:17.480 --> 00:07:21.480 Those are both good things. It's just, don't make me read through four enhancements 00:07:21.480 --> 00:07:25.840 when two of them are just, we changed the software version number by 0.1, 00:07:25.840 --> 00:07:29.520 and we have support for the freshly rebranded version 00:07:29.520 --> 00:07:32.640 of last year's AI processor. 00:07:32.640 --> 00:07:36.880 Why is everything called AI? I know I'm just a lowly tech YouTuber, 00:07:36.880 --> 00:07:40.160 but like, is this stuff really working? 00:07:40.160 --> 00:07:45.200 I mean, AMD says it must be working. They've seen two and a half X growth year over year 00:07:45.200 --> 00:07:50.440 for their Ryzen AI family, except that, guys, I'm just not convinced 00:07:50.440 --> 00:07:53.680 that that's why anyone is flocking to AMD mobile chips right now. 00:07:53.680 --> 00:07:57.480 To be clear, I love my Strix Halo Elite Book 8. 00:07:57.480 --> 00:08:04.680 This thing is freaking awesome, but the co-pilot button has nothing to do with why I like it. 00:08:04.680 --> 00:08:07.680 I like the performance and I like the battery life. 00:08:07.680 --> 00:08:13.360 The AI branding thing has gotten so over the top that at this point, it's hard for me to even tell 00:08:13.360 --> 00:08:17.600 what of it is a begrudging, well, we've got to shoehorn this into everything 00:08:17.600 --> 00:08:22.040 or our shareholders will abandon us. And what is a genuine misunderstanding 00:08:22.040 --> 00:08:25.840 of what is driving customers to purchase your products? 00:08:25.840 --> 00:08:30.920 To be clear, I don't think anyone's gonna be upset in the coming years about having a decent NPU 00:08:30.920 --> 00:08:34.080 so they can generate weird hybrid animals 00:08:34.080 --> 00:08:38.000 or 13-fingered French mimes or whatever it is they're into. 00:08:38.000 --> 00:08:41.080 But I promise you that the other parents in my kids' classes, 00:08:41.080 --> 00:08:45.600 they don't care if their PC is AI or not, and they don't even know what it means. 00:08:45.600 --> 00:08:50.440 Then again, they don't use LTT to keep up with the latest tech, so they wouldn't even know 00:08:50.480 --> 00:08:53.480 about our sponsor. If you guys enjoyed this video, 00:08:53.480 --> 00:08:57.240 why not check out the ShortCircuit I did on the GPD Win 5. 00:08:57.240 --> 00:09:02.760 This may not be Strix Halo at its best. It is a little bit thermally constrained in this form factor, 00:09:02.760 --> 00:09:05.840 but it might be Strix Halo at its coolest. 00:09:05.840 --> 00:09:09.840 This is a real house holding it upside down. This is a really awesome gaming handheld.