Meta has announced the next generation of its Meta Training and Inference Accelerator (MTIA) chip, which is designed to train and infer AI models at scale. The newest MTIA chip is a second-generation design of Meta's custom silicon for AI, and it is being built on TSMC's 5 nm technology. Running at the frequency of 1.35 GHz, the new chip is getting a boost to 90 Watts of TDP per package compared to just 25 Watts for the first-generation design. Basic Linear Algebra Subprograms (BLAS) processing is where the chip shines, and it includes matrix multiplication and vector/SIMD processing. At GEMM matrix processing, each chip can process 708 TeraFLOPS at INT8 (presumably meant FP8 in the spec) with sparsity, 354 TeraFLOPS without, 354 TeraFLOPS at FP16/BF16 with sparsity, and 177 TeraFLOPS without.

Classical vector and processing is a bit slower at 11.06 TeraFLOPS at INT8 (FP8), 5.53 TeraFLOPS at FP16/BF16, and 2.76 TFLOPS single-precision FP32. The MTIA chip is specifically designed to run AI training and inference on Meta's PyTorch AI framework, with an open-source Triton backend that produces compiler code for optimal performance. Meta uses this for all its Llama models, and with Llama3 just around the corner, it could be trained on these chips. To package it into a system, Meta puts two of these chips onto a board and pairs them with 128 GB of LPDDR5 memory. The board is connected via PCIe Gen 5 to a system where 12 boards are stacked densely. This process is repeated six times in a single rack for 72 boards and 144 chips in a single rack for a total of 101.95 PetaFLOPS, assuming linear scaling at INT8 (FP8) precision. Of course, linear scaling is not quite possible in scale-out systems, which could bring it down to under 100 PetaFLOPS per rack.Below, you can see images of the chip floorplan, specifications compared to the prior version, as well as the system.

Cadence's digital and custom/analog flows are certified on the Intel 18A process technology. Cadence design IP supports this node from Intel Foundry, and the corresponding process design kits (PDKs) are delivered to accelerate the development of a wide variety of low-power consumer, high-performance computing (HPC), AI and mobile computing designs. Customers can now begin using the production-ready Cadence design flows and design IP to achieve design goals and speed up time to market.

"Intel Foundry is very excited to expand our partnership with Cadence to enable key markets for the leading-edge Intel 18A process technology," said Rahul Goyal, Vice President and General Manager, Product and Design Ecosystem, Intel Foundry. "We will leverage Cadence's world-class portfolio of IP, AI design technologies, and advanced packaging solutions to enable high-volume, high-performance, and power-efficient SoCs in Intel Foundry's most advanced process technology. Cadence is an indispensable partner supporting our IDM2.0 strategy and the Intel Foundry ecosystem."

BIOSTAR, a leading manufacturer of motherboards, graphics cards, and storage devices today, introduces the BIH61-AHA industrial motherboard, designed for smooth, seamless industrial application from AIOT machines, Edge computing, HMI machine, Digital Signage and more. Built based on the Intel H610 chipset, the BIOSTAR BIH61-AHA motherboard is engineered to support the latest Intel Core i7/i5/i3 processors (LGA1700). Featuring the latest DDR5 technology with support for 2x DDR5-4800 MHz LONG-DIMM up to 96 GB max, it offers extensive customization to meet the intricate needs of various industries.

Targeting a diverse audience, from system integrators looking to build sophisticated AIoT machines to enthusiasts aiming to construct powerful automation or edge computing systems, the BIH61-AHA provides unparalleled connectivity and expansion options. With 10 COM ports, 5 PCI slots, dual Intel GbE LAN ports for reliable and high-speed network communication, extensive USB connectivity with 6x USB2.0 & 4x USB 3.2 Gen 1 ports, and multiple storage interfaces, including 4x SATA 6 Gb/s and 1x M.2 M-KEY socket. Its support for a wide operating temperature range (0 ~ 60 degrees C) and ATX power input also ensures the motherboard's adaptability in challenging environments.

Alphawave Semi (LSE: AWE), a global leader in high-speed connectivity for the world's technology infrastructure, today announced successful collaboration with Keysight Technologies, a market-leading design, emulation, and test solutions provider, demonstrating interoperability between Alphawave Semi's PCIe 6.0 64 GT/s Subsystem (PHY and Controller) Device and Keysight PCIe 6.0 64 GT/s Protocol Exerciser, negotiating a link to the maximum PCIe 6.0 data rate. Alphawave Semi, already on the PCI-SIG 5.0 Integrators list, is accelerating next-generation PCIe 6.0 Compliance Testing through this collaboration.

Alphawave Semi's leading-edge silicon implementation of the new PCIe 6.0 64 GT/s Flow Control Unit (FLIT)-based protocol enables higher data rates for hyperscale and data infrastructure applications. Keysight and Alphawave Semi achieved another milestone by successfully establishing a CXL 2.0 link setting the stage for future cache coherency in the datacenter.

Thanks to the leaked slides obtained by @InstLatX64, we have more details and some performance estimates about Intel's upcoming 5th Generation Xeon "Emerald Rapids" CPUs, boasting a significant performance leap over its predecessors. Leading the Emerald Rapids family is the top-end SKU, the Xeon 8592+, which features 64 cores and 128 threads, backed by a massive 480 MB L3 cache pool. The upcoming lineup shifts from a 4-tile to a 2-tile design to minimize latency and improve performance. The design utilizes the P-Core architecture under the Raptor Cove ISA and promises up to 40% faster performance than the current 4th Generation "Sapphire Rapids" CPUs in AI applications utilizing Intel AMX engine. Each chiplet has 35 cores, three of which are disabled, and each tile has two DDR5-5600 MT/s memory controllers, which operate two memory channels each and translating that into eight-channel design. There are three PCIe controllers per die, making it six in total.

Newer protocols and AI accelerators also back the upcoming lineup. Now, the Emerald Rapids family supports the Compute Express Link (CXL) Types 1/2/3 in addition to up to 80 PCIe Gen 5 lanes and enhanced Intel Ultra Path Interconnect (UPI). There are four UPI controllers spread over two dies. Moreover, features like the four on-die Intel Accelerator Engines, optimized power mode, and up to 17% improvement in general-purpose workloads make it seem like a big step up from the current generation. Much of this technology is found on the existing Sapphire Rapids SKUs, with the new generation enhancing the AI processing capability further. You can see the die configuration below. The 5th Generation Emerald Rapids designs are supposed to be official on December 14th, just a few days away.

Moore Threads Technology (MTT), a Chinese GPU manufacturer, is reportedly testing its next-generation graphics processors for client PCs and data centers. The products under scrutiny are the MTT S90 for client/gaming computers and the MTT S4000 for data centers. Characterized by their Device IDs, 0301 and 0323, this could imply that these GPUs belong to MTT's 3rd generation GPU lineup. While few details about these GPUs are available, the new Device IDs suggest a possible introduction of a novel microarchitecture following the MTT Chunxiao GPU series. The current generation Chunxiao series, featuring the MTT S70, MTT S80, and MTT S3000, failed to compete effectively with AMD, Intel, and NVIDIA GPUs.

Thanks to @Löschzwerg who found the Device Hunt submission, we see hardware identifiers in PCI ID and USB ID repositories earlier than launch, as this often signals the testing of new chips or drivers by various companies. In the case of MTT, the latest developments are complicated by its recent inclusion on the U.S. Entity List, limiting its access to US-made technologies. This introduces a problem for the company, as they can't access TSMC's facilities for chip production, and will have to turn to domestic production in the likely case, with SMIC being the only leading option to consider.

EK, the world-renowned manufacturer of premium water-cooling equipment, is introducing the specialized PCI expansion slot cover, the EK-Loop PCI Bracket Pass-Through. This cleverly designed PCI bracket provides an efficient solution for PC enthusiasts seeking to elevate their custom water-cooling experience by providing seamless connectivity for external cooling solutions, such as external radiators. With the EK-Loop PCI Bracket Pass-Through, users can maximize the potential of their water-cooling setup without being limited by the case size.

This specialized PCI expansion slot cover boasts two integrated pass-through fittings and a convenient cable management hole, ensuring a tidy and organized setup. With two G1/4" ports on each side, users can choose their preferred tubing type and size, granting them complete control over their water-cooling configuration. Specifically designed for this PCI expansion slot cover, the integrated pass-through adapter fittings ensure a secure and reliable connection. The cable management hole, equipped with a rubber gasket, protects cables from friction damage as they pass through the metal gap on the bracket.

During the China Chip Storage Future 2023 Storage Industry Trend Summit, Yingren Technology, widely recognized as InnoGrit outside of China, announced the initiation of mass production of its enterprise-level YR S900 PCIe 5.0 SSD controller. Marking a significant breakthrough, the YR S900 stands as China's first domestic PCIe 5.0 SSD controller. Operating on an open-source RISC-V architecture, the YR S900 is engineered to align with U.S. export restrictions, ensuring a seamless design and manufacturing process of the SSD controller. While Yingren Technology remains discreet about the specific process node to produce the YR S900, it's known that the controller embodies a versatile design, with compatibility extending to mainstream NAND from eminent manufacturers, and exhibits an impressive synergy with NAND from Yangtze Memory Technologies Corp (YMTC).

The YR S900 is a quad-channel controller, offering sequential read and write speeds peaking at 14 GB/s and 12 GB/s, respectively, and is equipped with InnoGrit's third-generation ECC engine to optimize 4K LDPC encoding and decoding. This collaboration with Kioxia's XL-Flash results in a low 4K random read latency of 10us, highlighting its potential to deliver higher data throughput, increased stability, and extended service life. The YR S900 encompasses a comprehensive feature set, including FDP, SR-IOV hardware virtualization, CMB, and a range of data encryption algorithms. While the mass production of the YR S900 underscores a monumental stride in SSD solutions within China, it remains to be seen whether adopting this new Chinese technology will enter markets beyond China.

IBASE Technology Inc. (TPEx: 8050), a renowned provider in the design and manufacturing of embedded computing solutions, rolls out the low-power IB838 3.5" single board computer powered by Intel Core i3 N-series (formerly Alder Lake-N) processor. Combining robust computing performance with advanced features, it is positioned to meet diverse needs across multiple industries, including industrial automation and control, retail displays, transportation, and automotive applications.

Fortified with intelligent power handling and a 9 V~36 V DC input range, the IB838 ensures reliability and consistent operations in transportation applications that may encounter variable power conditions. Versatile connectivity makes it ideal for automation processes and data acquisition systems. The SBC offers Type-C, DP++, and eDP or LVDS display options, along with a rich set of I/O interfaces, including two Intel PCI-E 2.5G LAN ports, two USB 2.0 ports, and four USB 3.2 ports (1x Type-C + 3x Type-A).

Twitter/X user @momomo_us has unveiled official documentation detailing a new GC-HPCE power connection standard to supply additional GPU power throughout motherboard. This connector can deliver over 600 Watts of power, surpassing 12VHPWR connector's capacity. The GC-HPCE power connector features four groups of pins. Sixteen pins are dedicated to power delivery, while the remaining twelve facilitate communication between the motherboard and the graphics card. Its size is comparable to the regular PCIe x1 connector. Positioned in alignment with the primary PCIe x16 slot on a motherboard, the GC-HPCE power connector is situated behind the x16 slot, typically where the motherboard chipset heatsink is found. This strategic placement ensures the graphics card can be easily slotted into both connectors, simplifying installation. The connector's design negates the need for an additional locking mechanism, as the x16 slot's existing lock sufficiently secures the graphics card.

First showcased at Computex, this connector was featured in several prototype motherboards and graphics cards, emphasizing its potential to enhance cable management and aesthetics. By eliminating supplementary power cables from the graphics card and channeling power through the motherboard, the graphics card's appearance remains uncluttered, and cable management behind the motherboard tray becomes more streamlined. Intriguingly, this connector is a familiar design. It's a modified version of the High Power Card Edge (HPCE) standard prevalent in the server industry. While ASUS has been the primary proponent of this connector, it remains to be seen if other companies will adopt this standard for their consumer products. Additionally, routing power throughout the motherboard will require manufacturers to include additional power regulating circuitry, potentially driving motherboard costs up. It is also not an agreed upon industry standard, which could block some future GPU upgrades from happening.You can check out ASUS'es implementation of the Back-to-Future (BTF) motherboard that supports this connector below.

Sparkle has announced the Titan OC graphics card, becoming the company's flagship Intel Arc A770 16 GB graphics card. It stands out with its 16 GB onboard memory, factory-overclocked GPU and memory, and an advanced cooling system for additional overclocking potential. After Sparkle's re-entry into the graphics card market, it unveiled a series of Intel Arc-based products. The Arc A770 Titan, in particular, showcases the company's technological advancements. The Sparkle Intel Arc A770 Titan OC Edition features Intel's ACM-G10 GPU with a base clock of 2300 MHz, which is a significant increase from Intel's stock 2100 MHz, and boasts a 16 GB memory with a 17.5 GT/s data transfer rate.

The card has two eight-pin auxiliary PCIe power connectors, yielding a 650-watt PSU requirement, and an expansive triple-fan cooling system for optimal performance even under heavy loads. A unique feature is the ThermalSync thermal sensor atop the cooler, which adjusts the LED light bar's color based on temperature, making it easier for users to monitor. When Intel discontinued its Limited Edition Arc A770 16 GB graphics card, there was a gap in the market for a high-performance graphics card to rival offerings from ASRock and Acer. Sparkle hasn't revealed the pricing for its Intel Arc A770 Titan OC Edition GPU.

IBASE Technology Inc. (TPEx: 8050), a global leader in network appliances and embedded solutions, proudly announces the release of the INA8505 enterprise 1U edge server. Powered by the Intel Xeon D-2700 processor and offering versatile connectivity options, this state-of-the-art appliance is specifically designed to excel in demanding 5G Open vRAN & MEC applications such as real-time data analytics, autonomous vehicles, and smart city deployments. It enables full control over resource allocation in the RAN and MEC, and has the potential to seamlessly integrate AI capabilities to dynamically optimize network performance in real time at the edge of the 5G network infrastructure.

The INA8505 delivers unmatched performance, scalability, and efficiency with flexible storage, offering two SATA/NVMe 2.5" HDD/SSD slots, 2x M.2 (M-key) SATA/PCI-E storage slots, and one 16 GB/32 GB/64 GB eMMC. With an FHFL PCI-E (x16) Gen 4 (supports 75 W) and an FHFL PCI-E (x8) Gen 4 (supports 75 W) configurable as PCI-E (x16) Gen 4 double FHFL (supports 120 W), the INA8505 adapts effortlessly into different network environments and meet future demands for increased scalability. It boasts a rich array of I/O connectivity options, including a VGA port from BMC (Aspeed 2600, IPMI 2.0 support), two USB2.0 Type-A ports, an RJ45 console port, and four 25 GbE SFP28 ports, ensuring enhanced adaptability to various connectivity needs.

IBASE Technology Inc. (TPEx: 8050), a leading provider of embedded computing solutions, introduces the IB837 3.5" single board computer that is designed for a wide range of IoT applications, including industrial automation, smart retail, healthcare, smart city and edge AI solutions, combining powerful processing capabilities with extensive connectivity options.

At the heart of the IB837 is the onboard Intel Celeron N & J series (formerly Elkhart Lake) processor, offering a good balance of performance and power efficiency. With support for up to 16 GB of DDR4-3200 SO-DIMM memory, the SBC ensures smooth multitasking and efficient operation, delivering exceptional performance while consuming minimal power for applications requiring a compact form factor.

According to the Linux Kernel Mailing List, AMD started to enable next-generation processors by submitting patches to the Linux kernel. Codenamed Family 1Ah or Family 26 in decimal notation, the set of patches corresponds to the upcoming AMD Zen 5 core, which is the backbone of the upcoming Ryzen 8000 series processors. The patches have a few interesting notes, namely few of them being: added support for the amd64_edac (Error Detection and Correction) module and temperature monitoring; added PCI IDs for these models covering 00h-1Fh and 20h; added required support in k10temp driver.

The AMD EDAC driver also points out that the Zen 5 server CPUs will max out with 12-channel memory. Codenames 0-31 correspond to next-generation EPYC, while 40 to 79 are desktop and laptop SKUS. Interestingly, these patches are just the start, as adding PCI IDs and temperature drivers are basic enablement. With the 2024 launch date nearing, we expect to see more Linux kernel enablement efforts, especially with more complicated parts of the kernel.

A few days ago, we reported that NVIDIA and PCI-SIG have been doing some work around the existing controversial 12VHPWR power connector to make some improvements and prevent issues that happened along the way. Called 12V-2x6, the connector features minor revisions, and you can read more about it here. Today, we learn that NVIDIA has already been shipping implementations of the new 12V-2x6 connector in GeForce RTX 4070 graphics cards, confirmed by Igor's Lab. Majority of RTX 4070 cards from AIBs shipped with 8-pin PCI connector; however, it turns out that cards that didn't use the reference 12-pin connection carried a new 12V-2x6 revision.

Below, you can see the picture comparison with the connectors from the Founders Edition of GeForce RTX 4070 and RTX 4080. The new connector on the RTX 4070 has shorter sensing pins, which can detect whether the connector is fully plugged in. If not plugged all the way, the card will not draw the required maximum power load. This should, in theory, solve connector burning issues happening in the past; however, we are still left to see. In addition, we are not sure if this is the final implementation of the new connector, as the labeling refferest to it as "H+", whereas the new connector should carry the label "H++" on its module. Igor's Lab article notes, "Since the GeForce RTX 4070 FE, NVIDIA has already been using a modified 12VHPWR connector with significantly recessed pins! While the GeForce RTX 4080 Founders Edition still relies on the header from the CEM 5.0 with only 0.45 mm offset, the GeForce RTX 4070 Founders Edition has been using a modified variant with 1.7 mm inward offset since its launch, similar to the 12V-2x6 connector."

The European brand expands its portfolio of PCIe Gen 4 x4 SSDs in M.2 format with a model dedicated to the PS 5 console. It is the IRDM PRO SLIM - ideal not only for gamers, but also professionals and enthusiasts.

The IRDM PRO SLIM is a twin model of the IRDM PRO M.2 SSD. Unlike its predecessor, the new drive is offered without a heatsink, making it 100% compliant with Sony's requirements for PlayStation 5 console drives. The technical parameters speak in favour of choosing an SSD from a European manufacturer. Maximum read and write speeds are 7000/6850 MB/s, respectively. The PCIe 4 x4 NVMe interface delivers twice the speed of older 3 x4 solutions. The memory speed is also influenced by DRAM buffer of up to 2 GB, thanks to which the drive works stably even under heavy use.

NVIDIA's compute and AI-oriented H100 GPU is supposedly getting an upgrade. The H100 GPU is NVIDIA's most powerful offering and comes in a few different flavors: H100 PCIe, H100 SXM, and H100 NVL (a duo of two GPUs). Currently, the H100 GPU comes with 80 GB of HBM2E, both in the PCIe and SXM5 version of the card. A notable exception if the H100 NVL, which comes with 188 GB of HBM3, but that is for two cards, making it 94 GB per each. However, we could see NVIDIA enable 94 and 64 GB options for the H100 accelerator soon, as the latest PCI ID Repository shows.

According to the PCI ID Repository listing, two messages are posted: "Kindly help to add H100 SXM5 64 GB into 2337." and "Kindly help to add H100 SXM5 94 GB into 2339." These two messages indicate that NVIDIA could prepare its H100 in more variations. In September 2022, we saw NVIDIA prepare an H100 variation with 120 GB of memory, but that still isn't official. These PCIe IDs could just come from engineering samples that NVIDIA is testing in the labs, and these cards could never appear on any market. So, we have to wait and see how it plays out.

Neo Forza, a Taiwanese brand entity under Goldkey Technology Corporation, had a booth at Computex 2023, where the company showed its next-generation memory and storage products. Starting off, the company presented its DDR5 memory kits running at 8000 MHz speeds, indicating the maturity of DDR5 and that we are way past the initial 4800 MHz speeds at launch. The demonstrated kit was a part of the company's TRINITY lineup, which was displayed in white. For demo purposes, the kit was configured as two 16 GB DIMMs.

Next up, Neo Forza has showcased its next-generation PCIe Gen 5 NVMe SSD. Running at x4 lanes of the new PCIe 5.0 protocol, the 2 TB SSD was spotted running at 10 GB/s speeds in both read and write tasks set by CrystalDiskMark benchmark. Interestingly, the SSD also boasted a massive heatsink, as shown below.

Keen-eyed hardware tipster momomo_us this week spotted that an upcoming AMD data center "Siena Dense" CPU has received verification, in the general sense, for SATA and PCI support - courtesy of the Serial ATA International Organization (SATA-IO). The information dump was uploaded to SATA-IO's online database on April 6 of this year - under the heading: "AMD EPYC 8004 Series Processors." As covered by TPU mid-way through this month the family of enterprise-grade processors, bearing codename Siena, is expected to be an entry-level alternative to the EPYC Genoa-X range, set for launch later in 2023.

The EPYC Siena series is reported to arrive with a new socket type - SP6 (LGA 4844) - which is said to be similar in size to the older Socket SP3. The upcoming large "Genoa-X" and "Bergamo" processors will sit in the already existing Socket SP5 (LGA 6096) - 2022's EPYC Genoa lineup makes use of it already. AMD has not made its SP6 socket official to the public, but industry figures have been informed that it can run up to 64 "Zen 4" cores. This new standard has been designed with more power efficient tasks in mind - targeting intelligent edge and telecommunication sectors. The smaller SP6 socket will play host to CPUs optimized for as low as 70 W operation, with hungrier variants accommodated up to 225 W. This single platform solution is said to offer 6-channel memory, 96 PCIe Gen 5.0 lanes, 48 lanes for CXL V1.1+, and 8 PCIe Gen 3.0 lanes.

Sparkle, a Taiwanese computer electronics maker, is again entering the GPU market after almost ten years of inactivity in the space. A while back, Sparkle was one of NVIDIA's original Add-In Board (AIB) partners and helped them launch the GeForce 7900 PCI GPU. The company continued to make NVIDIA-based GPUs until GTX 700 series in 2013. After a decade, Sparkle is back again with ambitions to be Intel's AIB partner and announced not one but three graphics cards to start. Called A750 Titan, A750 Orc, and A380 Elf, these cards feature triple-fan, dual-fan, and single-fan coolers, respectively.

The first in line is the Sparkle A750 Titan, a triple-fan, 2.5-slot design based on Intel Arc A750 GPU. Featuring 8 GB of GDDR6 VRAM, this top-end A750 SKU is clocked at 2300 MHz, up from the factory's 2050 MHz frequency. Titan's smaller brother is Sparkle A750 Orc, a dual-fan, "2.2"-slot (we assume smaller than two and a half and bigger than two slots) GPU with the same A750 GPU; however, it clocked slightly lower at 2200 MHz. Both models feature single HDMI 2.0 and 3x DisplayPort 2.0 output ports and require two 8-pin PCI power connectors. Lastly, we have the Sparkle A380 Elf, a half-length, ITX-sized GPU that fits in two slots and has a single-fan cooler. It is based on Intel Arc A380 and has identical specifications without factory overclocks applied. Pricing and availability are currently unknown.

[Editor's note: Our in-depth review of Antec Performance 1 FT Case is now live]

With outstanding cooling performance and lots of useful features, Antec Inc. presents the latest flagship of full towers, the new Performance 1 FT. Featuring an airflow-enhanced front panel design, magnetic front filter, and four pre-installed PWM fans, this new chassis provides efficient airflow and great cooling performance. The case is now commercially available from MSRP US$159.99.

The new Antec flagship supports the latest RTX 40 Series GPUs. Considering the increasing demands for CPU and GPU cooling, Antec designed the Performance 1 FT to enhance the air intake, improve the cable routing, and enable an easy installation with various options. The new temperature display function allows to keep control of the components. The display screen located on the case top allows users to check the temperature of the GPU and CPU at a glance. It addresses the need for temperature monitoring without purchasing expensive cooling kits.

Although NAND suppliers have continued to roll back production, there is still an oversupply of NAND Flash as demand for products such as servers, smartphones, and notebooks is still too weak. Therefore, TrendForce predicts that the ASP of NAND Flash will continue to fall in 2Q23, though that decline may shrink to 5~10%. The key to supply and demand returning to a market equilibrium lies in whether NAND suppliers can cut back on production even more. TrendForce believes if demand remains stable, then the ASP of NAND Flash will have an opportunity to rebound in 4Q23; if demand is weaker than expected, then ASP will take longer to recover.

Client SSD: Currently, PC OEM's have managed to liquidate most of their component inventory, and are now gearing up in preparation for mid-year sales events. Suppliers are cutting prices to clear out their inventories of PCIe Gen 3 SSDs, which are gradually being phased out. Meanwhile, prices of PCIe Gen 4 SSDs continue to face downward pressure due to a slow intake of new customer orders. The continuous decline of QLC products in 1Q23 has also dragged down the prices of TLC products, and there is relatively little room for prices to keep falling in 2Q23. While it still remains unclear whether or not demand will recover, TrendForce projects that the prices of PC client SSDs will drop 5~10% in 2Q23.

NVIDIA has killed SLI on its graphics cards, disabling the possibility of connecting two or more GPUs to harness their power for gaming and other workloads. However, SLI is making a reincarnation today in the form of a new H100 GPU model that spots higher memory capacity and higher performance. Called the H100 NVL, the GPU is a unique edition design based on the regular H100 PCIe version. What makes the H100 HVL version so special is the boost in memory capacity, now up from 80 GB in the standard model to 94 GB in the NVL edition SKU, for a total of 188 GB of HMB3 memory, running on a 6144-bit bus. Being a special edition SKU, it is sold only in pairs, as these H100 NVL GPUs are paired together and are connected by three NVLink connectors on top. Installation requires two PCIe slots, separated by dual-slot spacing.

The performance differences between the H100 PCIe version and the H100 SXM version are now matched with the new H100 NVL, as the card features a boost in the TDP with up to 400 Watts per card, which is configurable. The H100 NVL uses the same Tensor and CUDA core configuration as the SXM edition, except it is placed on a PCIe slot and connected to another card. Being sold in pairs, OEMs can outfit their systems with either two or four pairs per certified system. You can see the specification table below, with information filled out by AnandTech. As NVIDIA says, the need for this special edition SKU is the emergence of Large Language Models (LLMs) that require significant computational power to run. "Servers equipped with H100 NVL GPUs increase GPT-175B model performance up to 12X over NVIDIA DGX A100 systems while maintaining low latency in power-constrained data center environments," noted the company.

Today, thanks to Yuuki_Ans on the Chinese Bilibili forum, we have more information about the upcoming "Avenue City" platform that powers Granite Rapids and Sierra Forest. Intel's forthcoming Granite Rapids and Sierra Forest Xeon processors will diverge the Xeon family into two offerings: one optimized for performance/core equipped with P-cores and the other for power/core equipped with E-cores. The reference platform Intel designs and shares with OEMs internally is a 16.7" x 20" board with 20 PCB layers, made as a dual-socket solution. Featuring two massive LGA-7529 sockets, the reference design shows the basic layout for a server powered by these new Xeons.

Capable of powering Granite Rapids / Sierra Forest-AP processors of up to 500 Watts, the platform also accommodates next-generation I/O. Featuring 24 DDR5 DIMMs with support for 12-channel memory, with memory speeds of up to 6400 MT/s. The PCIe selection includes six PCIe Gen 5 x16 links supporting CXL cache coherent protocol and 6x24 UPI links. Additionally, we have another piece of information that Granite Rapids will come with up to 128 cores and 256 threads in both regular and HBM-powered Xeon Max flavoring. You can see storage and reference platform configuration details on the slides below.

Looking back at 2H22, as server OEMs slowed down the momentum of their product shipments, Chinese server buyers also held a conservative outlook on future demand and focused on inventory reduction. Thus, the flow of orders for enterprise SSDs remained sluggish. However, NAND Flash suppliers had to step up shipments of enterprise SSDs during 2H22 because the demand for storage components equipped in notebook (laptop) computers and smartphones had undergone very large downward corrections. Compared with other categories of NAND Flash products, enterprise SSDs represented the only significant source of bit consumption. Ultimately, due to the imbalance between supply and demand, the QoQ decline in prices of enterprise SSDs widened to 25% for 4Q22. This price plunge, in turn, caused the quarterly total revenue from enterprise SSDs to drop by 27.4% QoQ to around US$3.79 billion. TrendForce projects that the NAND Flash industry will again post a QoQ decline in the revenue from this product category for 1Q23.