Welcome back to Paul’s Tech News! With just one week remaining in March 2024, it’s a good time to take stock, to look back and reflect upon the Many Wonders that Q1 2024 brought to us – like back in January when CES happened and new GPUs launched or in February when it rained a lot. Which brings us to March when we were like “how is it March already what the hell 2024.” So many memories. But enough about the past, let’s now look to the Future which is what Nvidia, AMD, and even Intel are focused on with multiple scintillating Revelations made at multiple events that took place this week here and across the globe. And if all the AI fluffery doesn’t keep your dander up, there’s cheaper PC Hardware to discuss too including a $60 AIO CPU Cooler that’s making me more of a hypocrite than I already was. Let’s see if we can bring water cooling back from the dead on this week’s Tech News.
Excellent! Today’s video is brought to you by the Tower 300 series cases from Thermaltake, expanding their vertical Tower design with a stunning new lineup available in a variety of colors. The larger frame now supports micro-ATX motherboards as well as huge GPUs including the RTX 4090 without compromising airflow thanks to an abundance of filtered ventilation panels. Two CT140 fans come pre-installed, and the right side can house up to 420mm AIO radiators. Finish your build off with a sold separately 3.9in LCD panel kit or even go horizontal with a matching chassis stand for a truly unique look. For more on Thermaltake’s the Tower 300 series cases, click the sponsor link in the video description.
There were multiple conferences this week, most notably for us Nvidia’s GPU Technology Conference or GTC in San Jose, California, the Game Developers Conference or GDC very close by and at the same time in San Francisco, and the AMD-hosted Beijing AI PC Innovation Summit which was held in Texas – Beijing, sorry. We’ll begin with Nvidia’s GTC announcements, which were highlighted during CEO Jensen Huang’s Monday keynote and further elucidated in breakout sessions as the week progressed. The big news is Blackwell, the architecture that powers Nvidia’s next-gen GPUs that have long been rumored, which seem like they’ll continue Nvidia’s dominance in the AI compute sector. Based on the specs alone, the Blackwell B200 is a two-chip monstrosity with each full-sized Blackwell die supporting 104 billion transistors for 208 billion in total, a big increase from the 80 billion transistors in a Hopper H100 or H200 or, by comparison, an RTX 4090 GPU that has 76.3 billion. The B200 even has 192 GBs of HBM3E memory tucked in there along each edge. And yes, PC gaming fans, the Blackwell architecture will likely power the next generation of GeForce video cards, which will probably be the RTX 50 Series. So the specs and performance increase is listed for Blackwell, which TSMC will be manufacturing on their 4ND node, should trickle down or carry over somewhat to GPUs made for gaming if and when they debut as well. But that’s a big question mark since Nvidia has made no promises about the RTX 50 Series as of yet, and it’s clear that the data center AI side of their business is far more lucrative. They already have major industry players such as Amazon Web Services, Dell Technologies, Google Meta, Microsoft Open AI, Oracle, Tesla, and XAI lined up to adopt the Blackwell platform. And while pricing can vary based on the implementation, they’re expecting these GPUs to go for $30,000 to $40,000 each. But Nvidia would rather not sell just the GPUs, as they seem much more interested in providing full data center solutions, full systems with GPUs and CPUs and memory and all, such as the GB200, which combines two Blackwell B200 GPUs and one of Nvidia’s 72-core ARM-based Grace CPUs. Stack 36 GB200s together and you get the GB200 NVL72, a multi-node liquid-cooled data center system that houses 72 B200 GPUs and 36 Grace CPUs. And I don’t even want to ask what the price for one of these systems would be – “how much you got” would probably be Nvidia’s answer. Blackwell will come in three variants initially: the full-fat B200, which can pull up to 1200 watts by itself, a more power-optimized B200 that’ll pull 1,000 watts while still providing 90% of the performance, and a 700-watt B100, presumably for poor people. Maybe they’ll have a coupon for that one or something. Blackwell is expected to ship “later this year,” not to put too fine a point on it. So there’s your high-level recap, but if you want to nerd out about the specs and Nvidia provided early performance numbers, definitely check out the articles linked in the description. Oh, and Jensen got a new leather jacket too. If you want one, they’re only about $9,000. Meanwhile, about an hour’s drive north at GDC, AMD unveiled FSR 3.1, which most notably includes a long overdue update to their temporal upscaling solution, which had remained largely the same since FSR 2.2. The new upscaler will debut in Ratchet and Clank Rift Apart, although there is no set launch date as of yet for the game. It should improve image quality problems that have plagued the tech for some time, such as flickering and ghosting. Although the demo on their blog post about it was limited to a couple of short animations, it does look like there is improvement. The other part of the announcement that’s perhaps more exciting is that FSR 3.1 separates FSR upscaling and FMF frame generation, with the most practical application being that Nvidia GPU users can now run DLSS for upscaling and AMD’s Fluid Motion Frames for frame generation, an update that many users have requested. All of this seems like good news, but the main question is when. We do not know when FSR 3.1 will go public or when the new Ratchet and Clank game will launch. All we know is that game developers will have access to FSR 3.1 starting in Q2, which is in like a week, so hopefully it will go live not long after that. AMD’s other announcement this week came from the Beijing AI PC Innovation Summit, a bit further of a drive, but at least the hardware is consumer-focused. China is getting access to the Ryzen 8000G CPUs like the 8700G, and they’ll also have a couple more budget options to choose from with the 8700F and 8400F CPUs from the desktop APU 8000 series. But the “F” means they don’t have iGPUs. Now, call me crazy, but I thought the whole point of the horribly named 8000 series was to differentiate the newer APUs from the existing 7000 series AM5 CPUs. But what’s an APU without an iGPU? It’s just a regular-ass CPU, in my humble opinion. Hey AMD, don’t you already have a few 8-core Ryzen 7000 CPUs? Is this new 8700F actually just a Ryzen 7700 or 7700X? Well, it will probably also have AI hardware built-in, which AMD calls their XDNA engine. I just wish they would actually show that actually being used for something. Back to the presentation though, apart from the Chinese market announcements, they also shared a roadmap that lists Strix Point, which has been rumored for some time and will apparently be an APU lineup powered by Zen 5 with RDNA 3 Plus graphics and an XDNA 2 AI engine. They moved to version 2 with XDNA before they even showed us how version one helps with anything. Again, we have no clear launch date for these new APUs, just later this year. But if the Zen 5-based Ryzen 9000 Series CPUs launch at or sometime after Computex in early June, I would expect these Strix Point APUs to land not long after. I have a basic philosophy about PC hardware, and I guess consumer products in general: the more companies you see coming to market with a specific type of product, the higher the profit margins on that product will be. Headphones and headsets are a good example – just how many brands that make headphones are there? A lot. And all-in-one liquid CPU coolers, I will note, seem to have a new brand jumping into the market each week. To me, that means there’s a well-established set of OEM companies who can provide parts or ad hoc product solutions for AIO liquid coolers, and that it’s relatively lucrative to sell based on market perception. With custom water cooling being so pricey, the AIO market hasn’t had a problem selling coolers for upwards of $100 for quite some time. I’ve talked about AIO coolers a lot recently, and I still think that they are impractical and uncompetitive, especially when it comes to price in light of some of the excellent tower air coolers out there, most specifically Thermalright’s $30 to $40 Phantom Spirit 120 and 120 Evo. But my opinion in this matter has been challenged once again, and quite strongly, by none other than Thermalright themselves with the debut of their new 360mm all-in-one liquid cooler, the Frozen Edge 360, which costs $60. Less than $60, actually. And yes, it includes all three 120mm fans and even a bit of ARGB lighting. Performance is what really matters though, and yes, according to WCCF Tech’s review, this cooler is right up there with other 360mm AIOs that cost 50 to 100 bucks more. The only real disadvantage was with noise, where it’s in the middle of the pack. But when normalized and when tested at real-world wattage loads, the acoustics are just fine. So if I may correct my headline from a few weeks back, liquid cooling is not dead, but I still think it’s unnecessary unless you’re cooling a higher-end CPU or going small form factor where tower coolers just don’t fit. It does also introduce more points of failure to your system that you should be aware of. For $59, though, or even $54 for the white version, I would have much less of a problem pairing this with, say, an Intel 13700K or Ryzen 7700X. Speaking of more news and less time, Tech Briefs intro, and speaking of cheaper PC hardware, the Radeon RX 7900 XTX, the best alternative to an RTX 480 Super, is now $900, or MSI’s Gaming Trio version is at least as well as XFX’s Speedster Mark 310 if you count a $30 instant promo code, which means more GPU horsepower can now be obtained for less money, which is supposedly a continuous and ongoing trend in the PC hardware space, at least if you’re not the person at Nvidia who decides what the prices should be. If you are not too impressed by ray tracing and you’d like 24GB of VRAM instead of 16, the 7900 XTX for $900 is clearly a better choice than the $1,480 Super. Intel is midway through building out new fabs in Licking County, Ohio, but the State Department of Development seems a bit tongue-tied about the progress. Intel was supposed to have coughed up $20 billion in investment money towards the buildout in order to get things up and running by 2025. But as of the end of 2023, they only accounted for $1.5 billion spent, and it’s now estimated that the fabs won’t spin up until 2027 or 2028. $3 billion more has been laid out in contracts, but even $4.5 billion is less than a quarter of the amount that they promised. So, Intel, the people of Licking County deserve more. Break out the checkbook and fab faster. Apple’s message to DIY PC builders has been pretty consistent over the years: don’t do that. Still, there has been a thriving, if relatively small, community of Hackintosh builders over the years who recognize that you can get a lot more hardware for your money if you’re okay with jumping through some hoops to get macOS running on non-Mac hardware. Unfortunately, though, the latest macOS 14 Sonoma kills off legacy Wi-Fi drivers that were crucial for Hackintosh builds, and a change to the driver format makes workarounds difficult if not impossible. Good old Apple, making arbitrary software changes that break basic functionality for DIY users. So, rip Hackintoshes, and I guess we should also welcome former macOS DIY builders to our community. Good to have you here, guys. Windows isn’t the greatest thing ever, but at least Microsoft doesn’t make you feel like a criminal for using it with a homebuilt PC. Speaking of Apple and criminals, there’s a new side-channel attack specifically aimed at Apple M1, M2, and M3 processors. That kind of sucks a lot. It’s called Go Fetch, and it can steal cryptographic keys from the cache of these modern Apple processors. Check the Bleeping Computer article if you want the full technical description. But the bad news is that the exploit does not require physical access to the PC, and it is a hardware vulnerability that can’t be patched with a software or firmware update. If you do use an M1, M2, or M3-based Mac, there’s not much you can do besides practicing safe browsing habits, keeping your system updated, and regularly running antimalware scans to prevent the type of infection that can leave your system vulnerable to Go Fetch. But there you have it, guys. Tech news for the week. And if you liked it, click that like button or leave me a comment down below. While you’re down there, all the articles I talked about today are linked in the video’s description if you’re interested. And you can check out my store at paul.net for high-quality merchandise, t-shirts, hoodies, beer sets, and more. Subscribing to my channel is always a good call too. Thanks again, everyone, and we’ll see you next week. [Music]
