WEBVTT 00:00:00.160 --> 00:00:08.400 The GPU in my hand, the Intel Arc Pro B60, is quite possibly the most 00:00:05.839 --> 00:00:13.120 important GPU, not just of this year, but maybe the next few years, because of 00:00:11.200 --> 00:00:18.960 the way that it could massively disrupt NVIDIA's near monopoly on professional 00:00:15.440 --> 00:00:21.359 GPUs, drive costs way down for proumers 00:00:18.960 --> 00:00:25.519 and professionals alike, and if the demos they have running in the suite are 00:00:22.960 --> 00:00:29.760 anything to go by, change the game for AI developers and for businesses who are 00:00:28.000 --> 00:00:34.320 looking to leverage AI private. privately without giving their data away 00:00:31.679 --> 00:00:38.399 to companies like OpenAI and Deepseek. And I managed to get some good news for 00:00:36.320 --> 00:00:42.079 gamers, too. It wasn't an easy conversation, but I've got a solid 00:00:40.000 --> 00:00:47.440 commitment from Intel's VP and general manager for discrete graphics, Vivian. 00:00:44.559 --> 00:00:51.840 Yes, the OG land party queen of DFI, to get back to me on what their plan is to 00:00:49.200 --> 00:00:56.800 ensure that ARC consumer cards are going to be available at MSRP in the US, where 00:00:54.399 --> 00:01:00.879 they're often still well over. Stay tuned for that. Coming back to Ark Pro 00:00:58.879 --> 00:01:03.920 though, man. I get it. These are supposed to be for boring people in 00:01:02.160 --> 00:01:07.920 suits or whatever, but I'm almost feeling some of that land party energy 00:01:05.760 --> 00:01:14.240 here. Like, check out this wild board from Maxon. It's dual GPU with each of 00:01:11.439 --> 00:01:19.680 them getting eight lanes of PCIe Gen 5 and a total of 48 gigs of GDDR6 memory 00:01:17.280 --> 00:01:24.400 that yes, my friends, through some software magic that's part of what Intel 00:01:21.520 --> 00:01:31.680 is calling battle matrix can be shared across multiple GPUs for AI workloads, 00:01:27.759 --> 00:01:35.040 enabling models of up to 192 GB to sit 00:01:31.680 --> 00:01:37.119 in GPU memory. And that's just the tip 00:01:35.040 --> 00:01:42.159 of the iceberg. Oh, what's that down there? Oh, that's the tip of this segue 00:01:39.600 --> 00:01:46.880 to our sponsor UG Green. Their find track smartfinder helps you find lost 00:01:44.320 --> 00:01:53.680 items and pinpoint their exact location with an audible sound. Play the 00:01:51.240 --> 00:01:57.200 sound. It's super slim. It's equipped with a rechargeable and removable 00:01:55.520 --> 00:02:03.240 battery. And it's even compatible with the Apple Find My app. So check it out 00:01:59.119 --> 00:02:03.240 using our link today. 00:02:10.319 --> 00:02:16.400 Let's get the basics out of the way first. If you know the B580 Battle Mage 00:02:14.319 --> 00:02:21.440 gaming GPU, then you're already pretty familiar with the new Arc Pro B60. But 00:02:19.440 --> 00:02:25.840 there are some key differences. First up, the clock speeds are actually lower, 00:02:23.680 --> 00:02:29.520 which Intel says is to ensure the utmost instability and reliability for 00:02:27.840 --> 00:02:33.599 professional users. But don't let that fool you into thinking it's some kind of 00:02:31.200 --> 00:02:39.280 slouch for performance. Memory has been doubled from 12 gigs to 24 gigs. And as 00:02:37.280 --> 00:02:43.760 we've covered in the past, certified drivers and OpenGL tuning can result in 00:02:41.760 --> 00:02:48.560 much greater performance in professional applications with Intel showing off a 00:02:46.480 --> 00:02:53.040 heavyweight list of software partners, including the likes of Solid Works, 00:02:50.239 --> 00:02:57.360 Autodesk, and Blender. Here's a little fun fact, by the way. They won't prevent 00:02:55.040 --> 00:03:01.120 you from installing gaming drivers on your Arc Pro card if you're the kind of 00:02:59.120 --> 00:03:05.280 person who likes to mix business and pleasure on your PC. Let's take a closer 00:03:03.440 --> 00:03:09.760 look at some of these cards. I absolutely love that Intel is not taking 00:03:08.080 --> 00:03:14.959 the same approach as others in the space, locking their board partners into 00:03:12.480 --> 00:03:19.360 sometimes imperfect reference card designs. We've got a variety of 00:03:16.879 --> 00:03:22.959 everything from power connectors. This one has dual eight pins. This one has 00:03:21.040 --> 00:03:27.680 single eight pins. This guy over here with the dual GPUs gets the much 00:03:24.959 --> 00:03:31.599 maligned 12vt 2x6 connector. The point is not whether one of these approaches 00:03:29.760 --> 00:03:36.799 is superior to the other. The point is that you as the customer have an option. 00:03:34.480 --> 00:03:42.239 You want three fans. There's this two fans, one fan. How about no fans, a 00:03:40.400 --> 00:03:46.560 passive design, and use the chassis's air flow in order to blow air through 00:03:44.159 --> 00:03:51.680 the cooler. NVIDIA in particular has this weird allergy to cards being used 00:03:48.920 --> 00:03:56.640 for no not quite what we prescribe it for applications and it's refreshing to 00:03:54.400 --> 00:04:00.000 see somebody take a different approach. Regardless of the cooler design or 00:03:58.159 --> 00:04:07.360 power, you're going to get the same HDMI 2.1 as well as DisplayPort 2.0 UHBR 00:04:03.120 --> 00:04:09.519 13.5 and PCIe Gen 5. That is an upgrade 00:04:07.360 --> 00:04:13.840 over the consumer ARC cards which are PCIe Gen 4. I know it's practically 00:04:12.080 --> 00:04:17.919 cliche at this point, but it's been so long since I've seen a dual GPU card. I 00:04:15.680 --> 00:04:22.960 kind of can't help but tear it down with my handy convenient LT store precision 00:04:20.880 --> 00:04:27.919 screwdriver. Thanks, James. That right there is the 00:04:25.440 --> 00:04:32.080 biggest change from the consumer cards. Instead of just having memory on the 00:04:29.440 --> 00:04:35.600 front side of the PCB, each of our GPUs has memory on the back side of the PCB 00:04:34.160 --> 00:04:44.560 as well. That's what gives you double the capacity. Now before you ask, 00:04:39.000 --> 00:04:47.120 no SLI, Crossfire, XCSS, Fire, whatever 00:04:44.560 --> 00:04:52.080 is not coming. This is purely to enable customers to install more GPUs in fewer 00:04:49.759 --> 00:04:59.560 PCIe slots, but I still think it's super cool. Should we look at the other 00:04:55.080 --> 00:05:04.000 side? Let's go, boys. Oh, that's so 00:04:59.560 --> 00:05:06.160 cool, dude. Shout out, Maxon. 00:05:04.000 --> 00:05:09.600 Okay, there's lots of cool stuff to kind of talk about here. Starting with the 00:05:08.000 --> 00:05:15.360 fact that this just uses a standard four pin power connector. Thank you. Why 00:05:12.960 --> 00:05:20.160 can't GPU coolers just use these? Then it would be so much flipping easier for 00:05:17.120 --> 00:05:21.759 me to just like, I don't know, run a 00:05:20.160 --> 00:05:26.240 regular fan. Okay, it doesn't really change that much. But this is also cool. 00:05:24.320 --> 00:05:30.080 Check this out. This is very early hardware. You can actually see they did 00:05:27.680 --> 00:05:36.400 a hand rework on this with this little hair finness copper wire here. 00:05:34.160 --> 00:05:40.560 Man, that's a lot of GPUs on one board. Okay, it's two, but that's a lot more 00:05:38.400 --> 00:05:45.600 than usual. Last thing, this is pretty cool. So, you can actually see the PCIe 00:05:42.880 --> 00:05:50.320 lanes routed off to the separate GPUs. So, each of them gets a B8 interface. 00:05:47.919 --> 00:05:54.160 And then check this out. Each of them gets a single HDMI and a single display 00:05:52.240 --> 00:05:58.479 port. So, this card is effectively just like installing two by8 GPUs in your 00:05:57.039 --> 00:06:05.600 system. You just have to have a slot that supports bifurcation. So cool. Moving on to the 00:06:02.639 --> 00:06:10.800 Ark Pro B50, this one's a bit more of an unknown. Like its bigger brother, it 00:06:07.759 --> 00:06:12.800 gets a PCIe Gen 5 by8 interface and a 00:06:10.800 --> 00:06:17.759 generous allocation of VRAM that makes it very competitive against memory 00:06:15.120 --> 00:06:22.880 starved cards like the NVIDIA RTX 10000 that Intel is positioning it against. 00:06:19.919 --> 00:06:30.240 But unlike the B60, it's not just a repurposed consumer GPU. It gets just 16 00:06:26.560 --> 00:06:35.039 XE cores compared to 18 in their B570 00:06:30.240 --> 00:06:37.840 card, which is a bit of a downer, but it 00:06:35.039 --> 00:06:42.400 enables it to hit just 70 watts of total board power, which lets it operate 00:06:40.400 --> 00:06:46.319 without a separate power connector for some pretty cool use cases. Imagine, if 00:06:44.560 --> 00:06:50.400 you will, that you have a giant chassis on wheels like this. Now, imagine you 00:06:48.720 --> 00:06:54.720 have the opposite of that. you have a little slimline office workstation like 00:06:52.960 --> 00:06:59.440 you'd see deployed in an architectural office. This will fit perfectly in that 00:06:56.880 --> 00:07:03.759 without upgrading the power supply. Or hear me out, if you had a shop 00:07:01.360 --> 00:07:08.800 environment where you don't want active cooling drawing sawdust and crud into 00:07:06.479 --> 00:07:13.599 your machine and then kicking it around, I'd love to see someone do a passively 00:07:11.280 --> 00:07:17.919 cooled one of these. The coolest stuff is what's coming down the road, though. 00:07:15.599 --> 00:07:23.759 Let's turn our attention to Battle Matrix, which Intel is still working on 00:07:20.639 --> 00:07:25.440 the software magic for, but has it 00:07:23.759 --> 00:07:32.160 working enough that they're able to show off some pretty compelling demos. Behind 00:07:28.400 --> 00:07:35.759 me is full fat Deepseek. Okay, quantized 00:07:32.160 --> 00:07:38.960 4bit, but still DeepSeek with 671 00:07:35.759 --> 00:07:41.680 billion parameters running locally on 00:07:38.960 --> 00:07:46.479 one system. Now, not everything fits into VRAMm. They've got about 256 agents 00:07:44.800 --> 00:07:51.759 that don't get used as much on the system RAM, but this is it. Full fat 00:07:48.960 --> 00:07:56.160 Deepseek running on local hardware. This is perfect for a firm that has valuable 00:07:54.080 --> 00:08:03.120 intellectual property and doesn't want to entrust any of it to an offsite 00:07:58.960 --> 00:08:05.280 cloud-based AI. Boom, one box. And the 00:08:03.120 --> 00:08:10.800 secret sauce that I can't emphasize enough is that the VRAMm sharing that's 00:08:08.000 --> 00:08:16.639 going on with those GPUs is done in software rather than needing a fancy 00:08:13.360 --> 00:08:19.039 NVLink interface. Say for example, 00:08:16.639 --> 00:08:23.919 they've got another demo here. This is a code debugging demo that's showing 00:08:21.360 --> 00:08:30.479 corrupted video output on a Lunar Lake engineering sample right here. The model 00:08:26.720 --> 00:08:32.880 and the context take about 96 gigs. And 00:08:30.479 --> 00:08:38.240 uh while this demo is actually playback rather than real time because it took so 00:08:35.360 --> 00:08:43.360 long, what it's showing here is multiple AI agents working together to ultimately 00:08:41.039 --> 00:08:49.120 take the corrupted video encoding output and turn it into something that works. 00:08:46.160 --> 00:08:52.880 The mic drop moment though is the price. Unlike the consumer cards, Intel is 00:08:51.200 --> 00:08:58.160 leaving pricing to their board partners on these. So, I can't give you an exact 00:08:55.200 --> 00:09:03.120 price for the B60, but they are giving us enough information about these 00:08:59.760 --> 00:09:05.600 shockingly competent AI cards to infer a 00:09:03.120 --> 00:09:09.279 price. Battle Matrix is expected to land in the 5,000 00:09:07.080 --> 00:09:16.000 to$10,000 range, depending on whether you have a single B60 or up to eight. 00:09:12.000 --> 00:09:18.800 So, by my math, that's about 700 bucks a 00:09:16.000 --> 00:09:24.360 pop, which means that Mac Studio that I bought turned out to be, pardon my 00:09:20.480 --> 00:09:27.519 French, a complete waste of money. But 00:09:24.360 --> 00:09:29.839 hey, that's progress, baby. So, that's 00:09:27.519 --> 00:09:35.680 what's happening today. But what about the future? Intel is clearly committed 00:09:32.640 --> 00:09:38.080 to ARC integrated graphics. And looking 00:09:35.680 --> 00:09:42.480 at the road map for professional, not to mention the importance of AI in the 00:09:40.720 --> 00:09:48.160 future, they're clearly committed to that. But what about us gamers? What's 00:09:45.360 --> 00:09:51.440 the commit level there? Well, no matter how hard I pressed, they weren't willing 00:09:49.760 --> 00:09:54.800 to talk themselves into the same traps they ran into in the leadup to Arc 00:09:53.440 --> 00:09:58.959 Alchemist. They don't want to overcommit. But what they would say is 00:09:57.200 --> 00:10:03.680 that they're working really hard on building a hardware development pipeline 00:10:01.200 --> 00:10:08.480 that enables them to much more easily leverage their hardware across 00:10:05.519 --> 00:10:13.519 integrated gaming and professional which I'm taking as a yes. So, I hope it's 00:10:11.200 --> 00:10:16.880 clear to everyone, the Intel folks in this room who worked so hard to bring 00:10:15.200 --> 00:10:22.320 these products to life and put these demos together, and to you at home, how 00:10:20.000 --> 00:10:28.000 hard it is that I am rooting for them to succeed here. Just like I'm 00:10:24.680 --> 00:10:30.160 rooting for our sponsor, Acuity 00:10:28.000 --> 00:10:33.920 Scheduling. Managing your workday isn't just about doing your job. It's about 00:10:32.000 --> 00:10:38.720 staying organized, keeping your schedule in check, and optimizing your time. 00:10:36.320 --> 00:10:42.160 Sometimes the actual work takes less time than all of the prep and 00:10:40.240 --> 00:10:46.000 coordination and communication that comes with it. That's where Acuity 00:10:44.000 --> 00:10:50.560 Scheduling steps in. Making scheduling effortless so you can focus more on what 00:10:48.399 --> 00:10:54.160 you actually need to get done. You gain access to a powerful and intuitive 00:10:52.640 --> 00:10:58.720 calendar management system that synchronizes seamlessly across multiple 00:10:56.399 --> 00:11:03.360 platforms, reducing the chance of double booking internal or client meetings. 00:11:01.360 --> 00:11:07.040 Speaking of clients, Acuity Scheduling allows you to customize your online 00:11:05.040 --> 00:11:11.920 booking page with your brand's fonts, colors, and logo, ensuring a 00:11:09.120 --> 00:11:14.880 professional and personalized touch. You can keep your appointments running 00:11:13.120 --> 00:11:19.040 smoothly with automated confirmation emails, text reminders, intake forms, 00:11:17.200 --> 00:11:23.200 and even integrations with trusted payment providers. So, don't wait. Take 00:11:21.440 --> 00:11:26.640 control of your schedule and make booking stress-free with Acuity 00:11:25.120 --> 00:11:32.720 Scheduling. Click the link in the description and use offer code LTT20 to 00:11:29.360 --> 00:11:34.320 unlock 20% off your first subscription. 00:11:32.720 --> 00:11:38.480 If you guys enjoyed this video, why not check out the conclusion of the Ark 00:11:36.000 --> 00:11:43.240 Battle Mage Challenge? Spoiler, one of the participants went out and bought an 00:11:41.120 --> 00:11:48.000 Ark based on their experience. That's a pretty strong 00:11:45.120 --> 00:11:48.000 endorsement.