tiMicro
Altera Agilex 27 is the First Agilex 2X Platform Adding 800GbE and 1.6TbE and More
1 d ago
ASRock Rack AI Box-A395-AMD Ryzen AI Max 300 Series CPU 1 When AMD first announced its high-end Ryzen AI Max (Ryzen Halo) chips at the start of last year, the company aimed to use them across a wide range of computing markets, including laptops and desktop PCs. The product ecosystem built around AMD’s powerful integrated chips has turned out far wilder than the company expected, largely because demand for local AI inference systems has exploded. We have not seen as many laptops, but the desktop mini-PC space with these has grown like crazy since these make great local AI boxes to run agents and models in an always-on fashion. VIDEO As a result, when one of AMD’s partners does something with the Ryzen AI Max that isn’t just a small black box, it can turn some heads. That is exactly what we are seeing today with the latest Ryzen AI Max+ 395 mini-PC to cross our labs, this time coming from ASRock Industrial. The Taiwanese vendor’s industrial arm, ASRock Industrial, has a reputation for building no-nonsense machines with a few extra features for the industrial market. This gives us a chance to see what they think a no-nonsense Ryzen AI Max box should look like. ASRock Industrial AI Box-A395 Key Specs (As Configured) Processors AMD Ryzen AI Max+ 395, 16C/32T (5.1GHz) Operating System None: Barebones (As Tested: Ubuntu Linux 26.04) Memory 128GB LPDDR5X-8000, Soldered Drive Bays 1x M.2 2280 PCIe Gen4 x4 (Empty) 1x M.2 2280 PCIe Gen4 x4 (Empty) GPU Radeon 8060S Graphics (40 CUs) PSU 400W Internal PSU Form Factor Small Tower PC Dimensions 200 x 100 x 232 mm (7.87 x 3.94 x 9.13 in) Weight 2.8kg (6.2 lbs) Wireless Wi-Fi 7 (2×2) + Bluetooth 5.4 (MT7925) Color Aluminum Ports Top: 2x USB-A 480Mbps Rear: 2x USB-C 40Gbps, 1x USB-C 10Gbps, 2x USB-A 10Gbps, 2x HDMI 2.1, 1x 10GbE (RJ45, AQC113), 1x 2.5GbE LAN (RJ45, RTL8125BG), 1x 3.5mm Line-Out, 1x 3.5mm Mic-In Officially, ASRock Industrial promotes the barebones AI BOX-A395 as an AI workstation PC. While it can certainly be used as one, as we will see when we talk about its build and configuration, that marketing spin is probably a bit too optimistic. Instead, the AI BOX-A395 is best framed as a portable Ryzen AI Max system for industrial environments. It is a Ryzen mini-PC meant for on-the-go use or environments where it can run on its own without babysitting. If you wanted to find the ASRock Industrial AI Box-A395 online, here is a Newegg affiliate link. ASRock Industrial AI Box-A395 External Hardware Overview Starting with a look at ASRock Industrial’s Ryzen AI Max+ 395 mini-PC from the outside, you will immediately notice this is not the usual low-profile, wide-and-low mini-PC design. Instead, ASRock has opted for a more micro-tower style configuration, with the system’s motherboard sitting vertically. This makes it very space-efficient in terms of 2D floor space. ASRock Rack AI Box-A395-Front Angled 1 The other thing you will notice is the system’s aluminum body. There is also a large collection of vent holes on the left side of the system. Now install the optional carrying handle along the top side of the system, and it bears more than a passing resemblance to Apple’s 2003 Power Mac G5, otherwise (lovingly) known as the cheese grater. Between the unpainted metallic case and the small vertical vent holes, you cannot escape the comparison. If nothing else, it is a very distinctive design in a market flooded with a mix of colorful gaming-adjacent designs or minimalistic corporate designs. The one design that we looked at which is roughly a similar size and orientation is the Framework desktop, which we went over a bit in the video. Framework Desktop AMD Ryzen AI Max + 395 128GB Front With Tile Pack 2 Even with this tower design, ASRock Industrial has taken the interesting route of not bothering to use the system’s front at all. Not for I/O ports or cooling. The only thing on the front is the company’s logo. Once we take a look at the internals, the whys of this will become a bit clearer, but for now it serves to underscore how this is not a traditional mini-PC in terms of design, cooling, or how it is meant to be used. ASRock Rack AI Box-A395-Top 1 So if there are no I/O ports or buttons on the front of the system, where are they? As it turns out, it is all on top. In front of another large collection of vent holes (and mounting points for the handle), ASRock Industrial placed the system’s power button and a pair of USB-A ports at the top, as far forward as possible. ASRock Rack AI Box-A395-Top IO 1 The USB ports here are one of a few different surprises with this mini-PC. The black ports are not (just) a design choice. These are not SuperSpeed USB 3.x ports. Instead, these are classic 480Mbps USB 2.x ports. As the only two ports not on the rear of the system, this means there are no high-speed I/O ports on the front. This system assumes you will plug in basic peripherals at the top, and that is it. This is one of a few reasons why, despite ASRock Industrial’s “workstation” marketing, this is much closer to the kind of industrial PC design the group is better known for. ASRock Rack AI Box-A395-Front Angled 2 Moving on, if we stand the system up again and rotate it slightly to the left, we can see the right side. Like the front face, this side is blank. There are no vent holes or other notable features. Some screws hold the panel on and let you remove it to access the back side of the motherboard. ASRock Rack AI Box-A395-Rear 1 For the rest of the story on the A395’s chassis, we need to turn to the rear side of the system, which immediately illustrates just what kind of a system we are looking at. ASRock Rack AI Box-A395-Rear IO 1 Right off the bat, we see the familiar rim of an ATX I/O shield. The A395 is sold as a barebones PC, and it certainly lives up to those expectations. ASRock has not designed a highly custom case just to fit the Ryzen AI Max+ processor, but instead we are looking at AMD’s high-end AI chip placed inside of a standing mini-ITX form factor case. An integrated power supply, no less. ASRock Rack AI Box-A395-HDMI 2.1 Ports 2 Looking at the port configuration of the system, to the right of the obligatory mic-in/line-out 3.5mm audio ports, we have a pair of HDMI 2.1 ports driven by the Ryzen AI Max+ 395’s integrated Radeon 8060S iGPU. That iGPU can drive up to two HDMI-style ports, and many PC vendors have opted to implement both, including ASRock Industrial. ASRock Rack AI Box-A395-USB 3.2 Gen2 Type-A Port 2 If you need a DisplayPort-type output, ASRock Industrial also covers this via DP Alt Mode. There are three USB-C ports on the system, and all of them can be used for display outputs, taking advantage of the chip’s ability to drive up to four monitors at once. Otherwise, we are looking at a mix of USB 3 and USB4 ports. The left-most USB-C port is a 10Gbps USB 3.x port, while the right two USB-C ports are the two blazing 40Gbps USB-C ports that the Strix Halo silicon natively supports. These are likely to be the last ports you use for display outputs unless necessary, as doing so will eat up one (or both) of the system’s fastest USB I/O ports. For older peripherals and drives, ASRock Industrial also added a pair of USB-A ports on the rear of this system. Both of these are 10Gbps ports, using the industry-standard cyan coloring. And while ASRock Industrial did not use the USB-IF’s official logos here, the fact that they have otherwise clearly labeled the performance and capabilities of each USB port is greatly appreciated as it is something that other vendors have struggled with. ASRock Rack AI Box-A395-RJ-45 10GbE Port 2 As for non-USB I/O, ASRock Industrial has equipped the A395 with a pair of Ethernet ports. First up, we have a 10GbE port with an RJ45 jack, driven by Marvell’s nearly ubiquitous AQC113 controller. For the price and performance capabilities of Ryzen AI Max+ 395 systems, 10Gb Ethernet ports have been rarer than we would like to see. So it is good to see that one is included here. ASRock Rack AI Box-A395-RJ-45 2.5GbE Port 2 To the right of that is the second Ethernet port, a 2.5GbE port driven by Realtek’s RTL8125BG controller. Some Strix Halo systems ship with two 10GbE ports. We have also seen many, including the previously mentioned Framework Desktop, ship with only a single 5GbE or 2.5GbE port. Having 10Gb + 2.5GbE is a solid compromise here. Besides labeling each port with its max speed, ASRock has also used blue plastic with the 2.5GbE port to further distinguish the system’s two Ethernet jacks. ASRock Rack AI Box-A395-Antena 6 Meanwhile, if you instead need wireless networking, ASRock Industrial has that covered as well. The metal case makes internal antennas impractical, so the company added a pair of antenna connectors on the rear of the system that lead back to the Wi-Fi 7 radio. The antennas are included, of course. ASRock Rack AI Box-A395-AC Power Outlet 1 Meanwhile, here is a quick look at the system’s internal power supply from the outside. With the extra volume afforded by the standing system, and to keep it portable, ASRock Industrial opted for an internal power supply. So there is no need to haul around an external power brick here. The A395 can be powered with any standard C13 power cable. Assuming the system will run in headless mode (common for industrial PCs), this makes the A395 entirely self-contained. ASRock Rack AI Box-A395-Bottom 1 Finally, we will take a quick look at the bottom of the system. Although it lacks ventilation on its right side, the A395 is designed to be operated in a standing position, and the case design reinforces this. The included rubber feet keep the system elevated to keep the power supply intake clear and avoid scratching any surfaces. This is also where ASRock Industrial has placed the required stickers, including the model info, serial number, HDMI logo, and regulatory information. Now that we have taken a look at the outside of the AI Box-A395, let us dig into the guts of the system.
Spectra Logic Introduces Storage Reclaim to Help Organizations Recover flash and Disk Capacity They Already Own
1 d ago
Dell Technologies Turns Enterprise Data Into Trusted Context for AI Agents Dell AI Data Platform, the data foundation of the Dell AI Factory, expands with new data orchestration and storage capabilities This is a Press Release edited by StorageNewsletter.com on October 29, 2026 at 2:01 pm Summary: Unified Semantic Layer and Enterprise Knowledge Graph give every application and agent the same trusted view of enterprise data Dell AI Data Platform shortens the wait between question and insight with accelerated data processing that processes data nearly 4 times faster on average than CPUs alone Dell AI Data Platform storage enhancements improve security and benchmark performance New AI-ready data services move organizations from deployment to production quickly and reliably Average than CPUs alone across a range of workloads, and up to 20 times faster on batch processing workloads. Apache Arrow moves data efficiently between Dell storage and processing, so jobs can query data in place. This can reduce data-preparation time and shorten the path from raw enterprise information to trusted context. Dell Technologies expands the Dell AI Data Platform so AI systems deliver better answers, cost less to run and reach production faster, giving applications and agents trusted, current context from a company’s own data. Expanded multitenancy and security: Dell PowerScale now supports up to 500 tenants in a single cluster, with mTLS over NFS to encrypt and authenticate file traffic and more granular role-based access control for managing each tenant. AI service providers and enterprises running shared AI platforms can serve more teams and customers from one cluster while keeping each tenant’s data isolated and secure Benchmark workload performance: The new Dell Storage Performance Tool helps customers confidently size AI infrastructure, compare vendor solutions and keep GPUs fed by testing the performance of their S3-compatible object storage across training, inference and checkpointing Expertise to get to production Dell is expanding Dell AI Data Platform implementation services across its data and storage engines to move organizations from deployment to production results quickly and reliably. Dell Services establish a production-ready foundation, activating the platform’s analytics, processing, search and orchestration capabilities while keeping the platform tuned to the demands of AI workloads. Perspectives “Data without context is just noise. Most enterprises have spent years making their data accessible. That’s not the same as making it usable. An agent that can find a customer record but doesn’t know what it means, how it connects to everything else, or whether it can be trusted isn’t intelligent. It’s just fast. The companies that solve it won’t just deploy AI faster or run more agents. They’ll get more out of the data they already have,” said Arthur Lewis, president, infrastructure solutions group, Dell Technologies. “AI agents are only as effective as the data they can access, understand, and trust, and Nvidia accelerated computing and AI software help turn governed data into AI-ready context. By bringing Nvidia cuDF acceleration directly into the Dell Data Processing Engine, Dell helps shorten the path from stored data to GPU-accelerated AI-ready data their agents can use to drive real world innovation,” said Jason Hardy, VP, storage technology, Nvidia. Availability The Unified Semantic Layer, Enterprise Knowledge Graph and Knowledge Agents will be released in 1H 2027 Dell Data Processing Engine enhancements with the Nvidia acceleration stack will be available in December 2026, with further acceleration using Apache Arrow expected 1H 2027 Dell PowerScale security and multitenancy enhancements will be available in November 2026 The Dell Storage Performance Tool is available AI-ready data services are available
Aivres NVIDIA Vera Rubin and HGX B300 at OCP Summit
4 d ago
ASRock AMD Radeon AI PRO R9700 Creator-Top 1 Thanks to the explosion in demand for AI inference hardware, high-end video cards have become a hot commodity in the PC parts market. This spike has affected video cards at almost every level, but it has hit workstation-grade cards hardest. The crème de la crème of cards, and, importantly, the ones that ship with the most VRAM, have been hit hardest. NVIDIA, AMD, and even Intel have felt this spike in demand, as discrete video cards have become the most effective way to add high-performance inference hardware to a standard-form-factor PC. As part of our overall look at AI inference hardware and performance here at ServeTheHome, we have covered everything from mini-PCs to server-grade hardware, but not much in between. So today we are taking a look at that middle with a Radeon Pro AI workstation card from ASRock, their Creator Radeon AI Pro R9700. This is the fastest and best-equipped tier of workstation cards from AMD’s current-generation RDNA 4 architecture family. The Radeon AI Pro R9700 is the latest in AMD’s long-running line of workstation cards. While it is a video card through and through, with cutting-edge graphics features and numerous display outputs, AMD knew where the market was headed when it released its latest generation of workstation cards and named them accordingly, putting AI capabilities front and center. The underlying architecture reflects that, with RDNA 4 adding a slew of AI-centric upgrades, including faster matrix math performance and support for lower-precision formats down to FP8. As a result, the R9700 is AMD’s most AI-capable workstation video card yet, and has arrived at a time when those capabilities are in very high demand. If you want to check current prices or find one online, here is an Amazon Affiliate link. ASRock Creator Radeon Pro AI R9700 Hardware Overview As AMD is not selling their own reference design workstation cards for this generation, it falls to AMD’s partners to sell the cards instead. This production strategy is familiar to AMD on the consumer side, but it is a bit more unusual for a workstation card. Most notably, AMD has kept its Radeon Pro partners on a rather short leash with card designs. AMD is curating a very specific image for a very specific market, chief among them the ability to install multiple cards in a single workstation. As a result, though AMD isn’t selling its reference card, its partner cards are all quite similar in high-level design: dual-slot blowers, a form factor you will not readily find outside workstation cards these days. Xidax System-AMD Radeon AI Pro 3 Large That brings us to ASRock’s Creator Radeon Pro AI R9700. ASRock has been a long-time partner of AMD’s at every level, from client and workstation motherboards to video cards. So it is not surprising to see them get into workstation video cards as well, now that AMD is offering that option to its closest partners. ASRock AMD Radeon AI PRO R9700 Creator-Top Angled 1 ASRock’s R9700 card is, in all practicality, exactly what you would expect from a workstation card given the form factor and AMD’s constraints. This dual-slot video card follows the PCIe specifications to a T in width and height to ensure it (and perhaps a few of its friends) fits in a workstation chassis. ASRock Creator Radeon AI Pro R9700 Specifications CUs 64 AI Accelerators 128 GPU Boost Clock Speed 2.92GHz Peak Matrix Throughput (FP8) 382.7 TFLOPS Peak Vector Throughput (FP32) 47.8 TFLOPS Memory 32GB GDDR6 Memory Bandwidth 640GB/sec (20Gbps x 256-bits) ECC Yes (Linux Only) Infinity Cache 64MB I/O PCIe Gen5 x16 TBP 300W Form Factor PCIe CEM (Air Cooled) Dimensions 271 x 111 x 39 mm (10.7 x 4.4 x 1.5 in) Architecture RDNA 4 GPU Navi 48 Under the hood, the card uses AMD’s Navi 48 GPU, the most powerful RDNA 4 GPU in this generation. The GPU offers 64 of AMD’s CUs, which in turn are comprised of a total of 4096 shading units for vector math and 128 AI accelerator cores for matrix operations. Notably, AMD does not include discrete matrix math cores in the RDNA architecture as it does in its CDNA architecture. Instead, AMD relies on Wave Matrix Multiply-Accumulate (WMMA) for matrix operations, which run on the same CUs that make up the vector unit. Consequently, while these CU-integrated AI accelerators are not quite like the matrix hardware on AMD’s server-grade hardware, they are no slouch, either. Especially after AMD doubled its floating-point performance (and quadrupled its INT8 performance) coming from RDNA 3 to RDNA 4. RDNA 4 Overview With a peak clock speed of 2.92GHz, the R9700 delivers 382.7 TFLOPS of peak matrix math throughput with dense FP8 math, or double that with sparsity. One key improvement in this generation of RDNA architecture is added FP8 support, which not only boosts the card’s compute throughput but also better aligns AMD’s hardware with the industry trend toward lower-precision weights. While the card does not support faster execution of even lower precisions such as FP6 and FP4, it still supports them, as they are processed at FP8. The bigger benefit is that it still allows smaller weights and thus larger model parameter counts, making the most of the card’s 32GB of VRAM. The most typical example is to use FP6 weights to get a ~30B parameter model working on a 32GB card with some VRAM space to spare. RDNA 4 Compute Engine As is traditional for a workstation card, AMD and ASRock packed the R9700 with as much memory as the card can handle. With Navi 48 sporting a 256-bit bus, the card has 32GB of GDDR6 memory organized as 16 2Gbit chips in clamshell mode, meaning chips on the front and back of the card. Above all else, memory support is a key differentiator between AMD’s consumer and professional cards, so you will find 32 GB only on the Radeon AI Pro cards. Patrick’s note: At an Advancing AI 2025 event, I asked AMD why the R9700 only had 32 GB. They looked at me like, why would you want more, since that is the max the memory configuration supports. My feedback was it might cost a lot more, but the market would probably absorb a higher cost and higher memory part for AI. I still wish AMD had something between this and the MI350P. The GDDR6 memory clocks at 20Gbps, giving the card a total of 640GB/second of memory bandwidth. Meanwhile, as a workstation card, the R9700 has another ace up its sleeve: ECC support. This is implemented as soft ECC and is only available under Linux (and even then, it must be manually enabled), but it adds an extra layer of reliability for systems and racks built around the card. Again, with the ability to install up to 4 in standard ATX systems, the need for reliability increases because the odds of failure rise, so it is not just a checkbox feature. AMD-Navi-48-Die-Shot Navi 48 sits somewhere between a mid- to high-end GPU in die size (357mm2) and transistor count (53.9B), which makes it a bit of an oddity in AMD’s lineup as far as card progression goes. AMD did not produce a true high-end GPU this generation, and as a result, the R9700 actually has less VRAM than the previous-generation 48GB Radeon Pro W7900. This means it is a step back in overall memory capacity, but AMD is looking to make up the difference with compute throughput, thanks in large part to the newer chip architecture. One advantage of being a smaller chip, however, is that it keeps TDPs in check. AMD (and ASRock’s) Total Board Power figure for the card is just 300 Watts, which makes it relatively easy to cool a blower-style card without a noise blowout. This also means that even 4 cards in a system is not going to result in excessive power requirements, though it will certainly be pushing the limits of a single power supply. ASRock Creator Radeon Pro AI R9700 Build and Design As for video card design, ASRock’s Creator Radeon AI Pro R9700 is a typical dual-slot blower design. To ensure the card can be installed in multiples, ASRock has kept things tightly tuned to PCIe specifications, with a dual-slot card measuring 10.7 inches long and 4.4 inches high. Interestingly, the blower design here is open-backed, meaning the back side of the card has open vents so it can pull in air from both sides. ASRock AMD Radeon AI PRO R9700 Creator-Top 2 The fan itself is a pretty typical blower design. A single radial blower sits at the rear of the card, pushing cool air over the heatsink and toward the card’s PCIe bracket, where a large vent exhausts the resulting hot air. According to ASRock, it uses a vapor chamber to efficiently pull heat from the Navi 48 GPU, with a layer of Honeywell’s PTM7950 as the thermal interface material between the GPU and cooler. Focusing on the rear of the card, we can better see the venting cut into the card’s PCB and its shroud. We can also see that the card has a shroud on its back side. As noted earlier, the 32GB of GDDR6 is organized in a clamshell configuration on the R9700, meaning 8 chips are on the rear of the card and need protection and cooling, just like the chips on the front side. ASRock AMD Radeon AI PRO R9700 Creator-Rear 1 Since this is a workstation card, ASRock has also put thought into its power connector placement. The Creator R9700 uses an ATX 12V-2×6 connector, placed on the rear of the card instead of the top, another workstation-centric design element that makes it easier to install multiple cards in a system. The 12V-2×6 connector is rated for up to 600W, so the 300W TBP R9700 is well within the cable’s limits. ASRock AMD Radeon AI PRO R9700 Creator-Front 1 Meanwhile, at the front of the card, you will find a not-too-subtle reminder that this is a video card, with 4 display outputs, the maximum the Navi 48 GPU can drive. Unique among R9700 vendors, ASRock went with an all-DisplayPort configuration, forgoing the usual HDMI port for a fourth DisplayPort. All DisplayPorts are DisplayPort 2.1a, and with DSC, the card can drive displays up to 12 K. ASRock AMD Radeon AI PRO R9700 Creator-Front Angled 2 Otherwise, aside
Apple unveils new iPad mini with OLED display and faster A20 Pro chip
4 d ago
Apple on Monday announced a new iPad mini with a larger 8.5-inch OLED display. The tablet features a thinner and lighter design than its predecessor and is powered by an A20 Pro chip. The new iPad mini starts at $699 for the Wi-Fi model and $849 for the Wi-Fi and cellular model. The tech giant revealed the new device just days before its “Pro and Tell” event that’s scheduled to take place on Wednesday. The new iPad mini is 5.4 mm in thickness, and available in two colors. The device delivers peak brightness of up to 1,200 nits, more than twice that of the previous-generation device, along with full screen brightness of 600 nits for SDR and HDR content and a 2 million:1 contrast ratio. Apple says the iPad mini is ideal for dim-light environments. Image Credits:Apple The iPad mini also comes with a new nano-texture display option that reduces glare in bright environments, a feature already available on the iPad Pro. The updated tablet now also features a 12-megapixel center stage front camera, faster wireless and cellular connectivity, and fast-charge support. The Super Retina XDR display uses OLED technology for pixel-level lighting control, delivering deeper blacks, richer contrast, and support for the P3 wide color gamut for more vibrant colors. It also features an “Acoustic Display” that Apple says enhances the audio experience when gaming, watching videos, or editing content. Specially designed woofers integrated into the display deliver deeper bass and higher volume for richer sound, per the company. Image Credits:Apple The new iPad mini launches with the next generation of Apple Intelligence and Siri AI, which can draw on a user’s personal context to help them find what they need, answer questions, and understand what’s on their screen. Powered by a 7-core GPU with Neural Accelerators in each core, the iPad mini has up to 4.5-times faster AI performance than the previous generation, according to Apple. The A20 Pro also features a 6-core CPU, delivering up to 60% faster multithreaded performance for everyday tasks such as photo editing and multitasking across apps. The device can be preordered on October 28, with availability beginning November 6.
ADT R27A-BK3 U.2 to PCIe Adapter Review
5 d ago
Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 12 At OCP Global Summit 2026 in San Jose, we stopped by the Aivres booth. The lineup featured a lot of NVIDIA, spanning a Vera Rubin NVL72 rack down to a PCIe GPU server, with an HGX Rubin NVL8 tray, Vera CPU hardware, and an HGX B300 server in between. VIDEO We saw the Aivres Vera Rubin rack at its NVIDIA GTC 2026 booth, so this was a chance to see more of the hardware around it. Aivres KRS8000V4 NVIDIA Vera Rubin NVL72 Rack The big rack at the booth was the Aivres KRS8000V4, the company’s NVIDIA Vera Rubin NVL72 liquid-cooled AI rack. At the show, this was a mix of dummy blanks and a live node. These racks are rough to transport to trade shows, but we have seen them many times before. Aivres OCP Summit 2026- NVIDIA Vera Rubin NVL72 Liquid-Cooled AI Rack 2 Aivres lists 18 1U compute trays and nine 1U NVLink switch trays, as shown above, with the switch trays in the middle and power shelves at the top and bottom. That is 72 Rubin GPUs and 36 Vera CPUs on sixth-generation NVLink. That creates a giant GPU domain over NVLink with over 20TB of HBM. Aivres says this is 100% liquid-cooled and can operate at 45C to help cooling efficiency. A hot tub is usually 38-39C, so that gives you an idea of how warm the fluid is in the rack. Aivres OCP Summit-NVIDIA Vera Rubin NVL72 Liquid-Cooled AI Rack 1 With eighteen compute trays, plus switch trays, management switches, an integrated NVLink spine, power, and liquid cooling, these racks have a lot going on. Aivres KL2288 2U Liquid-Cooled HGX Rubin NVL8 Compute Tray The Aivres KL2288 is the 2U liquid-cooled compute tray for NVIDIA HGX Rubin NVL8, pairing eight Rubin GPUs with dual AMD EPYC 9006 “Venice” processors. This exhibit shows how an eight-GPU HGX platform can be hosted by x86 CPUs in a fanless design. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 1 Aivres lists the KL2288 at up to 400 PFLOPS for NVFP4 inference and 280 PFLOPS for NVFP4 training. The booth placard also lists 2.3TB of HBM4 memory with 176TB/s of total HBM bandwidth. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 2 The front has a UART port, two USB ports, two RJ45 ports, and a row of eight E1.S drive bays. It also has space for an NVIDIA BlueField DPU. Here is a quick look at the compute tray, where you can see it heavily featuring liquid cooling. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 6 The rear of the system has the OCP rack quick disconnects and uses a busbar for power. You can also see that the rear does not have exhaust area for fans since it is liquid-cooled. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 14 Aivres calls this a modular, fanless, fully liquid-cooled design. Inside, the two CPUs have copper cold plates flanked by banks of DDR5 DIMMs that are also liquid-cooled. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 12 Distribution blocks with red (warm) and blue (cooler) fittings route coolant through the tray. Aivres OCP Summit-2U Liquid-Cooling NVIDIA HGX Rubin NVL8 AI Compute Tray 17 This was a neat look at an upcoming system.
Block’s new voice model sounds eerily human
5 d ago
Jack Dorsey’s Block released a new voice model out of the company’s research lab that runs locally and sounds strangely human with its expressions like laughs, pauses, and cries. The company said the model, called Dessa, is trained fully on human speech and a single NVIDIA GPU rack. Dessa is a full-duplex model, which means it can talk and listen at the same time, just like the GPT-Live model OpenAI released earlier this year. Typically, speech models are paired with a large language model for knowledge retrieval. Block said that Dessa’s knowledge stems from the conversational data it has. Block said that the current models can sound unnatural and annoying because they have text in the pipeline that drives speech generation. A lot of models like ElevenLabs v4 or Fish Audio Drama 3 use voice tags to drive emotion in speech generation. Dorsey’s company claims that emotion is built into its model. It’s not clear if that emotion is being steered by some kind of tags. “Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can’t carry. The most powerful AIs in the world can’t hear how you feel. We think that’s ridiculous, leads to a worse experience, and is potentially dangerous,” the company said in a blog post. The company’s researchers posted multiple examples of conversations with Dessa where the agent had different voices, cadences, and emotions. For instance, in the conversation, the bot said it was tired from work, and at that point it sounded tired. In the example below, the bot was talking about the person making it comfortable while sounding emotional and almost breaking down. Another one shows a Block team member playing a game with the voice assistant. The company said that the model can run locally on a consumer-grade computer such as a MacBook M4 Max or better. However, there is no public access to Dessa at the moment, and Block has put out a call for safety researchers to test the model. This is likely to test out cases where a voice agent could trick people on the other end. Block is in the business of finance and selling its solutions to small businesses. Square already uses voice AI to handle ordering. But with its own voice model that can run on-device, it can power a lot more business and consumer experiences at a lower cost.
Nvidia CEO plays with adorable Microduck robots, 20,000 sell
7 d ago
Earlier this week, a video of Nvidia CEO Jensen Huang playing with a flock of waddling Microducks during his keynote at the company’s GTC Berlin conference went viral. The robots, stomping their little feet and making tiny R2-D2 noises, sat, fell down dead for a “nap,” and summersaulted on voice command. They were so darn cute that by the next day, preorders of the Microducks surpassed 20,000 units, Hugging Face’s Clem Delangue said in a tweet. The Microducks are expected to be delivered in time for Christmas, meaning these robots might already be on their way to becoming the hottest toy for the holiday season. Delangue thinks they might end up being the best-selling robot of the year, and possibly not even for kids. The promotional videos show adults playing with them, too. Microducks are open source devices produced by Pollen Robotics, a company acquired by Hugging Face in 2025, which means they are now an Nvidia company. The GPU giant bought Hugging Face in September. This is actually Hugging Face’s second robot, after the also-cute Reachy Mini. Microducks were trained using Nvidia’s technology stack, wherein simulations of them were taught basic tricks (walk, sit, stand, etc.) that they ship with. Pollen says that owners can train their robots with new tricks and then share them with the community. Here’s the video below. Enjoy it as a palate cleanser after the scary black eye the AI industry gave itself this week, as AI involvement was blamed for cyberattack after cyberattack. OK, AI might seriously damage the world. But it can be fun, too. It’s hard to watch this video without ordering one.
Nvidia GTC Berlin 2026: Dell Technologies Advances Enterprise Agentic AI from the Deskside to Data Center with Nvidia
7 d ago
Blog written by Uttara Kumar, senior product marketing manager, Nvidia, published October 21, 2026 General availability of the Nvidia instances is coming next month, with planned expansion of availability to Europe — bringing an integrated platform for training, inference and large-scale reinforcement learning to cloud customers.A5X instances bring together Nvidia Vera Rubin NVL72, Nvidia ConnectX-9 SuperNICs, Google Cloud Virgo Network and Google Cloud’s fourth-generation advanced liquid-cooling technology for demanding agentic inference and reinforcement learning workloads at scale. Google Cloud plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites. Ineffable Intelligence, a London-based technology company founded by AlphaGo creator David Silver, was the first customer to access A5X. Ineffable is using the platform to advance large-scale reinforcement learning and shape “superlearner” systems that learn continuously from experience. Learning through interaction requires a continuous cycle of generating experience, evaluating outcomes and updating models, requiring infrastructure that can support the entire process at scale. A5X brings Vera Rubin into Google Cloud’s AI Hypercomputer as an integrated system, combining compute, networking and storage to support experience-driven learning at scale. “We’re building superlearners that continually improve through their own experience,” said David Silver, founder and CEO, Ineffable Intelligence. “That demands both substantial compute and a tight connection between generating experience and training on it, so each improvement in the agent can quickly shape its next experience. Vera Rubin’s combination of compute performance and high-bandwidth connectivity, between the Vera CPU and Rubin GPUs and between the Rubin GPUs within the rack, allow us to push the scale and ambition of our reinforcement learning research further.” WPP, a London-based marketing and communications company, has also demonstrated its creative agent reference architecture on Nvidia Vera Rubin-based A5X instances using Nvidia Nemotron models and Nvidia OpenShell, part of the Nvidia OpenShell Reference Architecture for safe AI agents. The agents turn conversational requests into multistep creative workflows, such as video generation, upscaling and restyling, with human approval before execution. OpenShell provides governed sandboxes for agent execution, helping control access to sensitive brand assets and creative tools. In WPP’s initial tests, workloads that typically required a multi-node cluster ran on a single A5X instance. As organizations deploy agentic systems, they need to protect the proprietary models and sensitive data that power these systems — even when running on infrastructure they don’t own. Google Cloud will be among the first cloud providers to offer Nvidia Confidential Computing at rack scale on bare metal instances such as A4X Max instances powered by Nvidia GB300 NVL72. Additionally, Google Cloud is making Confidential Computing generally available on G4 VMs powered by Nvidia RTX PRO 6000 Blackwell Server Edition GPUs, offering a flexible platform for confidential inference and fine-tuning. Customers including Vertebrae, Polaris Cloud and Confidential Core AI are using Confidential G4 for conversational privacy, verifiable agent execution and policy-governed enterprise intelligence. To support energy-efficient deployment at scale, Google Cloud and Nvidia are exploring Nvidia DSX MaxLPS capabilities for Vera Rubin-based infrastructure to help optimize performance per watt and expand capacity within available data-center power budgets.
Nvidia GTC Berlin 2026: Supermicro Introduces the Nvidia Vera CPU Rack, a 256 CPU, 100% Liquid-Cooled Rack-Scale Solution, Complementing the Nvidia Vera Rubin Platform, for Agentic AI at Scale
7 d ago
Summary: HPE Private Cloud AI, the turnkey AI factory solution, expands agentic AI capabilities to provide developers, applications, and AI agents with secure access, governed AI agent integration, and greater GPU flexibility to help organizations move AI applications into production New infrastructure innovations for AI factories include Nvidia Vera CPU Rack by HPE, support for secure agent architectures and modular infrastructure to help organizations scale increasingly complex AI workloads Expanded security and operational capabilities help enterprises and sovereign AI providers protect sensitive workloads, maintain control over infrastructure, and improve AI factory efficiency HPE will support the Nvidia OpenShell Reference Architecture for Safe Agents, delivering full-stack AI factory governance, isolation, and security with Nvidia Vera Rubin, Nvidia Vera CPUs, and Nvidia BlueField-4 HPE announced new innovations across the HPE AI Factory with Nvidia portfolio that help enterprises securely deploy and operate increasingly autonomous AI across hybrid and sovereign environments.HPE AI Factory with Nvidia brings together models, infrastructure, data, and workflows to help enterprises run AI with greater control over performance and data access. “AI agents are becoming an essential part of how enterprises operate and create value,” said Cheri Williams, SVP & GM, private cloud & flex solutions, HPE. “As organizations deploy AI across industries and increasingly sensitive environments, they need infrastructure that gives them greater control over security, sovereignty, and how enterprise data is accessed and used. We’re helping customers scale AI across hybrid environments while maintaining the governance and operational control required to move from experimentation to production.” HPE Private Cloud AI expands secure access to enterprise AI services and agentic capabilities For enterprises seeking to securely develop and scale AI applications, HPE is expanding HPE Private Cloud AI, the turnkey AI factory solution within the HPE AI Factory with Nvidia portfolio, with new HPE AI Essentials capabilities that make it easier for developers, ISVs, and AI agents to securely access, integrate and consume enterprise AI services. Through production-grade APIs and MCP servers, organizations can connect applications and autonomous agents to inference, model, and knowledge services while maintaining centralized governance, identity management, and operational control. New capabilities include: Governed APIs for enterprise AI: Developers and ISVs can programmatically and securely consume inference, models, knowledge services, and AI pipelines to integrate HPE Private Cloud AI into applications and industry solutions First-class MCP services for agentic AI: MCP servers are versioned and governed by role, tenant, and project, with auditing designed to enable AI agents to interact with enterprise systems while maintaining organizational governance OpenAPI-based development: OpenAPI serves as the source of truth for supported interfaces and enables automatically generated documentation, software development kits and code samples that help partners build integrations faster Enterprise-grade identity and governance: Role-based access control, machine identity support, and comprehensive audit logging are designed to help organizations manage human and autonomous access to AI services Support for regulated and sovereign environments: API-first and Kubernetes-assisted onboarding are now supported across both connected and air-gapped HPE Private Cloud AI deployments HPE is also expanding GPU flexibility within HPE Private Cloud AI to help enterprises efficiently support diverse AI workloads. Now generally available, HPE Private Cloud AI can combine Nvidia H200 GPUs and Nvidia RTX PRO 6000 Blackwell Server Edition GPUs within the same cluster, enabling organizations to support a broader mix of AI workloads through a common, governed operating environment. Nvidia’s open-source KAI Scheduler provides quota- and priority-based scheduling to improve GPU utilization and workload placement across projects. Schneider Electric advances agentic industrial AI with HPE Private Cloud AI Schneider Electric plans to combine HPE Private Cloud AI, co-engineered with Nvidia, with its industrial and operational technology expertise to support Industrial AI Agents at the Edge, bringing intelligence closer to where operational decisions are made while maintaining control over sensitive data. HPE and Nvidia deliver purpose-built infrastructure for deploying and securing AI factories and agents As organizations scale AI beyond individual applications, they require infrastructure that can efficiently support autonomous agents, secure execution environments and increasingly complex computing demands. HPE is expanding its AI infrastructure portfolio with Nvidia technologies designed to help organizations build and deploy AI factories at scale. New infrastructure innovations include: Dedicated compute for agentic AI: HPE is introducing the Nvidia Vera CPU Rack by HPE, a high-density, liquid-cooled system for CPU-intensive agentic AI and reinforcement learning (RL) workloads. With 256 Nvidia Vera CPUs, support for 22.5K sandboxes, and 80% faster agentic CPU performance, it helps AI factories scale agent sandboxes, RL evaluations, and data-processing pipelines alongside GPU compute Secure execution for autonomous agents: HPE will support the Nvidia OpenShell Reference Architecture for Safe AI Agents, an open, full-stack architecture for building, containing, observing and securing AI agents across AI factories, enterprise infrastructure, and the edge. The architecture combines the Nvidia OpenShell secure runtime on Nvidia Vera, Nvidia Sentry with Nvidia BlueField-4 for hardware-based security enforcement, and Nvidia Vera Rubin NVL72 for model inference, helping enterprises govern agent activity, isolate workloads and enforce security policies as agentic AI deployments scale. HPE and Nvidia are collaborating to extend security and protection from the agent runtime to the underlying AI infrastructure throughout the HPE AI Factory with Nvidia portfolio Simplified AI factory deployment: HPE will also support Nvidia DSX POD, a reference design for pre-fabricated and validated infrastructure in modular containers to reduce onsite integration work and shorten deployment timelines for large-scale AI factory environments. HPE will work with Nvidia to deliver DSX POD AI factories, building upon the modular data center expertise and services offered today with AI Mod Pod. HPE strengthen security and operations for sovereign AI infrastructure with Nvidia Organizations deploying AI in sovereign and regulated environments need infrastructure that protects sensitive data and models while maintaining control over security, performance and operations. HPE is expanding support for Nvidia technologies that address these requirements across AI factory environments: Confidential computing for sensitive AI workloads: To help address security and sovereignty requirements, HPE is expanding support for Nvidia Confidential Computing on HPE ProLiant Compute DL380a Gen12 servers equipped with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs, helping organizations secure frontier AI models and protect sensitive data on sovereign infrastructure while workloads are in use. On Nvidia-Certified Servers, confidential computing establishes hardware-based trusted execution environments designed to isolate AI workloads, including virtual machines, containers, and bare metal deployments, from underlying infrastructure Availability Updates to HPE Private Cloud AI with the HPE AI Essentials API platform, including administrator-enabled developer and ISV access, are rolling out in late 2026 Included in HPE Private Cloud AI is a self-service HPE AI Essentials developer experience for application and agent registration, API and MCP discovery, and credential management. This will be rolling out in 2027 Heterogeneous GPU support for Nvidia H200 and Nvidia RTX PRO 6000 Blackwell Server Edition GPUs within HPE Private Cloud AI is generally available now KAI Scheduler capabilities for HPE Private Cloud AI are generally available now