WEBVTT 00:00:00.240 --> 00:00:07.040 AMD, I need you to try harder. I mean, it's not like you're actively 00:00:04.880 --> 00:00:12.320 antagonizing your partners, customers, and the media. That's more NVIDIA's jam. 00:00:09.840 --> 00:00:19.760 It's just that if you want to have your Ryzen moment in GPUs, now is the time to 00:00:16.720 --> 00:00:22.560 find the hunger that turned you from a 00:00:19.760 --> 00:00:28.320 player in the CPU space to the A player that you've become. I mean, don't get me 00:00:25.519 --> 00:00:33.520 wrong, the new Radeon 960 XT looks pretty good compared to NVIDIA's 00:00:30.240 --> 00:00:35.920 lukewarm 5060 series, and that might win 00:00:33.520 --> 00:00:40.960 you some sales, but I want you to look at the big picture. This is an 00:00:38.800 --> 00:00:46.079 opportunity while your main competitor is distracted by their rapidly growing 00:00:43.200 --> 00:00:52.000 AI business to change the perception of Radeon from that brand for scrub gamers 00:00:49.440 --> 00:00:57.120 to that brand that I never knew how much I liked until I tried it. It's an 00:00:54.399 --> 00:01:01.760 opportunity to build an army of loyal Radeon customers who through their sheer 00:00:59.600 --> 00:01:06.400 numbers will incentivize your board partners to build exciting solutions and 00:01:04.080 --> 00:01:10.640 incentivize developers to focus on Radeon when they're optimizing their 00:01:08.000 --> 00:01:14.400 games and software. But I'm getting ahead of myself a little bit. Adam just 00:01:12.720 --> 00:01:19.360 walked out of the keynote and he's got the details on everything from the new 00:01:16.479 --> 00:01:23.560 Radeon cards to AMD's fastest Thread Rippers ever. He's also got a segue to 00:01:22.479 --> 00:01:27.799 our sponsor. Touch me again. You're going to 00:01:25.520 --> 00:01:32.320 find out exactly who I am. I'm your sponsor, ODU. They have 00:01:30.479 --> 00:01:35.920 over 50 apps to help you cook up your business. Whether you want to put your 00:01:34.000 --> 00:01:42.240 hands on marketing, accounting, or even HR, they have you covered. The RX960 XT 00:01:39.920 --> 00:01:47.920 is AMD's attempt to appeal to gamers who reject NVIDIA's lackluster 8 GB cards 00:01:45.520 --> 00:01:52.880 while also offering a profit marginfriendly 8 GB option for 00:01:50.159 --> 00:01:56.720 shareholders. To their credit, unlike Team Green, who has a history of quietly 00:01:54.880 --> 00:02:02.079 downgrading their lower VRAMm variants in other ways, AMD has built both of 00:01:59.280 --> 00:02:08.879 these new cards using the same Na'vi 44 GPU with 32 compute units, up to 3.1 GHz 00:02:05.360 --> 00:02:12.400 boost clocks, and up to 821 AI tops 00:02:08.879 --> 00:02:14.640 running through a 16 lane PCIe Gen 5 00:02:12.400 --> 00:02:19.200 interface. Thank you AMD for understanding that budget cards often 00:02:16.879 --> 00:02:24.959 end up in budget systems with older gen PCIe slots and having those extra lanes 00:02:22.640 --> 00:02:30.400 makes a big difference in compensating for those lower speeds. Power targets 00:02:27.040 --> 00:02:33.120 are 150 and 180 watts for the 8 GB and 00:02:30.400 --> 00:02:40.280 16 GB cards respectively. And according to AMD, the 960 XT6 GB will be the 00:02:37.760 --> 00:02:44.480 world's best GPU under $350. and I'd like to believe them 00:02:42.560 --> 00:02:47.519 because is pretty dire for budget gamers right now, but there's a couple of red 00:02:45.920 --> 00:02:51.280 flags that I want to point out. AMD hasn't presented a clear path to ensure 00:02:48.959 --> 00:02:55.120 that the 960 XTs will be actually available at their MSRPs. And for 00:02:53.519 --> 00:02:59.440 another, I couldn't help but noting the way that AMD focused on comparisons 00:02:57.280 --> 00:03:04.560 between their 16 GB card and NVIDIA's 5060 Ti 8 GB. It's a valid comparison, 00:03:02.800 --> 00:03:09.360 and the numbers do look impressive with AMD claiming a roughly 15% better 00:03:06.879 --> 00:03:13.680 performance per dollar in raster and in rateracing at 1440p. 00:03:11.519 --> 00:03:21.480 How much of that margin comes from VRAM issues that will pop up on the 960 XT 8 00:03:17.040 --> 00:03:24.159 GB as well? Also, what is that? And 00:03:21.480 --> 00:03:28.480 hence, is that just three display outputs? AMD at least did confirm during 00:03:26.959 --> 00:03:35.760 the press briefing that these are DisplayPort 2.1A, UHPR 13.5, and HDMI 00:03:33.080 --> 00:03:41.120 2.1B. So, at least you'll be able to natively drive all but the highest 00:03:37.840 --> 00:03:42.640 bandwidth displays. And I guess if you 00:03:41.120 --> 00:03:46.959 really need more monitors, there's always DisplayPort multiream transport. 00:03:44.799 --> 00:03:52.000 But it just feels like a bit of a weird spot for them to cut costs. Although if 00:03:50.319 --> 00:03:59.280 it helps contribute to their competitive pricing, then maybe that's a good thing. 00:03:54.799 --> 00:04:01.120 Well, sort of. See, the 960 XT 16 gig is 00:03:59.280 --> 00:04:09.200 looking pretty good, undercutting the MSRP of the 5060Ti 16 gig and even the 8 00:04:05.200 --> 00:04:12.000 gig by 80 and $30, respectively. But the 00:04:09.200 --> 00:04:15.599 8 gig version of the 960 XT is stunningly 00:04:13.400 --> 00:04:20.239 non-competitive. Something that AMD tacitly acknowledges when they gave it 00:04:17.799 --> 00:04:25.259 approximately 2 seconds of screen time during their 00:04:26.919 --> 00:04:33.360 announcement. Yikes. I mean, sure, they 00:04:30.400 --> 00:04:37.680 price match NVIDIA's 5060, but this is exactly what I was getting at during the 00:04:35.199 --> 00:04:42.880 intro. price matching the guy that everyone hates for overcharging is not 00:04:40.400 --> 00:04:47.919 enough. And it doesn't seem like they're even trying to compete with Intel's B 00:04:45.120 --> 00:04:53.040 series GPUs. The whole thing even just feels like backwards progress compared 00:04:50.320 --> 00:04:58.520 to AMD's own lineup with gamers expected to pay $30 more than last gen's 00:04:56.000 --> 00:05:03.199 disappointing 8 gig card, the RX7600. Anyway, these cards will be 00:05:01.120 --> 00:05:07.440 launching on June 5th, so keep your eyes peeled for our upcoming coverage. While 00:05:05.520 --> 00:05:11.919 we don't know for sure if their prices will be fake, what we do know for sure 00:05:09.520 --> 00:05:15.759 is there will be fake frames. The latest version of FSR, AMD's AI upscaling tech, 00:05:14.160 --> 00:05:20.960 brought massive improvements earlier this year, and AMD plans to keep mining 00:05:18.080 --> 00:05:24.479 for extra performance with FSR Redstone, which focuses on improving path tracing 00:05:22.880 --> 00:05:28.160 workloads. Specifically, to demonstrate this, they showed off this unflattering 00:05:26.560 --> 00:05:33.199 demo that looks like it was ripped out of a Project Gotham Racing from Xbox 360 00:05:30.720 --> 00:05:37.120 era. Like I like, what is this? Why did Why does it look like this? I thought it 00:05:34.720 --> 00:05:42.560 was a tech demo. Anyways, the updates are one, neural radiance caching, which 00:05:40.320 --> 00:05:46.160 learns how light bounces within a scene to try and predict ray intersections 00:05:44.479 --> 00:05:49.840 before they happen. Two, ray regeneration, which is NVIDIA's ray 00:05:47.759 --> 00:05:53.680 reconstruction, but more AMD earer. Three, super resolution, which is super 00:05:51.919 --> 00:05:58.479 resolution. And four, frame gen, which is frame gen. Sounds 00:05:56.120 --> 00:06:04.560 cool. What would be cooler is having more games supporting the tech. Oh. Uh, 00:06:01.440 --> 00:06:06.720 okay. What next? We got AI. AI. Oh, 00:06:04.560 --> 00:06:12.639 right. Uh, new Threadripper. The king is back, baby. And the new Threadripper 00:06:08.880 --> 00:06:14.880 9000 lineup looks wild. While AMD isn't 00:06:12.639 --> 00:06:19.520 packing any more cores than last year, Shimata Peak seems to squeeze a lot of 00:06:17.440 --> 00:06:26.000 juice out of the move to AMD's more efficient Zen 5 architecture. Starting 00:06:21.840 --> 00:06:28.240 with a 300 MHz higher boost clock, which 00:06:26.000 --> 00:06:33.680 doesn't sound like much, but when it's across this many cores, it adds up to a 00:06:31.280 --> 00:06:37.759 lot. Now, if you were hoping for 3D Vcash on these models though, and let's 00:06:35.759 --> 00:06:42.080 be real, we were all hoping for 3D Vcash, you're going to have to keep 00:06:39.440 --> 00:06:47.360 waiting. But despite the minor changes to the on paper specs, AMD showed off a 00:06:45.199 --> 00:06:52.080 pretty decent performance uplift over their last gen Inbench before they 00:06:49.600 --> 00:06:56.960 pivoted to shaming Intel for a while. You know what? Fair enough AMD. Enjoy 00:06:54.720 --> 00:07:01.759 your victory lap. Now, naturally, given that these are workstation chips, AMD is 00:06:59.199 --> 00:07:05.599 also emphasizing how improved CPU performance will allow for improved AI 00:07:04.240 --> 00:07:10.479 performance when you're working with large models that require multiple GPUs, 00:07:08.000 --> 00:07:14.960 which is becoming more and more relevant as larger local models are gaining 00:07:12.680 --> 00:07:19.680 popularity. The best part, though, is these new chips are dropping compatible 00:07:17.120 --> 00:07:25.280 with existing 7000 series motherboards with just a BIOS update. Good guy AMD 00:07:23.440 --> 00:07:29.759 finally delivering on a Threadripper upgrade path just like we delivered 00:07:27.759 --> 00:07:33.479 these great cargo pants available in two inseam lengths for shorter and for 00:07:32.000 --> 00:07:38.000 taller kings ltstore.com. There is bad news though. 00:07:36.000 --> 00:07:41.759 If you want to get your fingers on a 96 core Threadripper Pro when they arrive 00:07:39.440 --> 00:07:45.599 in July, they will only be available through partners like Lenovo, meaning 00:07:43.599 --> 00:07:50.960 that enthusiasts are going to be stuck with the nonpro lineup that caps out at 00:07:48.000 --> 00:07:56.080 just 64 cores with fewer PCIe lanes, half the number of memory channels, and 00:07:52.880 --> 00:07:58.479 far less supported memory capacity. AMD 00:07:56.080 --> 00:08:03.919 has well and truly forgotten the Threadripper enthusiasts who supported them in 00:08:00.400 --> 00:08:06.240 the early days of Ryzen. Cool. Also, 00:08:03.919 --> 00:08:11.000 there's no pricing information today. It really is amazing how much AMD acts like 00:08:08.479 --> 00:08:16.080 Intel and NVIDIA when they are the top dog. Also coming to the workstation is 00:08:13.680 --> 00:08:20.240 some new AMD GPU tech. Earlier this week, we did a video on Intel's B60 GPU, 00:08:18.639 --> 00:08:24.560 and we talked about why it was their most important GPU yet, offering an 00:08:22.960 --> 00:08:29.280 affordable option for entry-level workstations that required certified 00:08:26.720 --> 00:08:34.000 drivers and creating a lowcost platform for running pretty large AI models 00:08:31.360 --> 00:08:39.200 locally. Meanwhile, AMD is hoping to wrigle their way into the ProGPU market 00:08:36.159 --> 00:08:43.039 by throwing a whopping 32 GB of VRAM on 00:08:39.200 --> 00:08:44.880 their new Radeon AI Pro R9700, a GPU 00:08:43.039 --> 00:08:48.000 name optimized for shareholder value. AMD is also making big strides to 00:08:46.640 --> 00:08:52.399 improve their software, bringing improved RockM support to Windows and 00:08:50.480 --> 00:08:56.880 Linux alike, which enables developers to bring their AI and HPC applications to 00:08:54.560 --> 00:09:00.160 Radeon. Get it away from CUDA is what they say. They say, "Please, please, no 00:08:58.399 --> 00:09:03.760 CUDA, sir." They didn't say how much this GPU would cost, but they did spend 00:09:02.160 --> 00:09:09.440 a fair amount of time comparing it to the 5080. So, a a thousand probably not. 00:09:08.399 --> 00:09:13.519 It's probably not It's probably not going to be a thousand. AMD's 00:09:11.440 --> 00:09:18.959 announcements this year then were a mixed bag of exciting and also 00:09:16.560 --> 00:09:22.480 underwhelming. I mean, compared to the NVIDIA keynote, at least there was 00:09:20.640 --> 00:09:30.240 something worth talking about. And we took this incredible GPU and we shrunk 00:09:26.240 --> 00:09:32.880 it in here. Does that make any sense? 00:09:30.240 --> 00:09:37.040 And AMD is clearly busting their buns to catch up to NVIDIA in the workstation 00:09:34.640 --> 00:09:40.480 and data center spheres. But like the rest of the industry, they seem to have 00:09:38.720 --> 00:09:46.000 forgotten about their customers who don't have six, seven, or eight figure 00:09:43.519 --> 00:09:51.839 annual budgets. I guess we're all just chopped liver. You know what isn't 00:09:47.920 --> 00:09:53.680 chopped liver though? Our sponsor. L is 00:09:51.839 --> 00:09:57.839 getting older and older. He really need to retire. That's why I use our sponsor 00:09:55.760 --> 00:10:02.080 Odu to help him reboot Lannis's cat tips. Odo has over 50 apps for business. 00:10:00.320 --> 00:10:06.160 So while using a project app to track our progress, Lance told me to try this 00:10:04.000 --> 00:10:10.480 method to lure the cats out. An hour later, only Noodle shake my hand to work 00:10:08.480 --> 00:10:14.640 together. In our brainstorm session, Missy and Brownie remain missing, but 00:10:12.399 --> 00:10:18.000 Noodle is one hardworking cat. We listed the ideas out and structure our 00:10:16.320 --> 00:10:22.000 workflow. Noodle agreed to letting me cook the cat. Cook with the cat. We 00:10:20.560 --> 00:10:25.760 scheduled the shoot within a couple of clicks, keeping things organized. 00:10:23.839 --> 00:10:29.519 Filming with him has been a blast, but this sneaky cat better not think I will 00:10:27.680 --> 00:10:33.279 work for free. Luckily, I can easily lock my time sheet along with the 00:10:31.200 --> 00:10:37.360 project or to keep test organized. But to be honest, Noodle didn't contribute 00:10:35.120 --> 00:10:40.720 much. Lance better not retired. If you guys enjoyed this video, maybe check out 00:10:39.040 --> 00:10:46.480 the one we did on Intel's most disruptive GPU yet, the Ark Pro B60.