In case you missed it, Sam Altman of OpenAI took the Internet by storm late last week with the unveiling of Sora, the generative AI that can congure up photoreal video clips based on prompts, with deadly accuracy. While Altman and his colleagues in the generative AI industry had a ton of fun generating videos based on prompts from the public on X; it became all too clear that the only thing holding back the democratization of generative AI is the volume of AI accelerator chips. Altman wants to solve this by designing his own AI acceleration hardware from the grounds up, for which he initially pitched an otherworldly $7 trillion in investment—something impossible with the financial markets, but one that's possible only by "printing money," or through sovereign wealth fund investments.

Jim Keller needs no introduction—the celebrity VLSI architect has been designing number crunching devices of all shapes and sizes for some of the biggest tech companies out there for decades, including Intel, Apple, and AMD, just to name a few. When as part of his "are you not entertained?" victory lap, Altman suggested that his vision for the future needs an even larger $8 trillion investment, Keller responded that he could design an AI chip for less than $1 trillion. Does Altman really need several trillions of Dollars to build a grounds-up AI chip at the costs and volumes needed to mainstream AI?

Tenstorrent, a company that sells AI processors and licenses AI and RISC-V IP, announced today that it selected Samsung Foundry to bring Tenstorrent's next generation of AI chiplets to market. Tenstorrent builds powerful RISC-V CPU and AI acceleration chiplets, aiming to push the boundaries of compute in multiple industries such as data center, automotive and robotics. These chiplets are designed to deliver scalable power from milliwatts to megawatts, catering to a wide range of applications from edge devices to data centers.

To ensure the highest quality and cutting-edge manufacturing capabilities for its chiplet, Tenstorrent has selected Samsung's Foundry Design Service team, known for their expertise in silicon manufacturing. The chiplets will be manufactured using Samsung's state-of-the-art SF4X process, which boasts an impressive 4 nm architecture.

Tenstorrent hosted their "Nerds Talking to Nerds About RISC-V" event this week in India where a dozen high profile industry experts hosted technical talks and panels about every facet of the RISC-V landscape and future. Among these are some familiar names to anyone who's been keeping up on the CPU industry; Raja Koduri of his own AI Generative Gaming startup company, Lars Bergstrom of Google, Naveed Sherwani of Rapid Silicon, and of course Jim Keller the CEO of Tenstorrent itself. On the first day of the event a mere 42 minutes into the YouTube live stream during his keynote talk, Jim Keller is providing an overview of Tenstorrent's latest silicon design goals. He presents a slide showing a wide comparison of various competitor's integer performance in SPEC CPU 2017 INT wherein a raw performance value for AMD's yet released "Zen 5" is listed, as well as the operating frequency and TDP of the supposed sample.

The slide shows all of AMD's recent architectures starting with the original "Zen" (Naples) and the improvements each successive generation has made. Also shown is one of Intel's latest "Sapphire Rapids" Xeons, a projected performance point of NVIDIA's in-house CPU architecture "Grace", Amazon's "Graviton" series with a projected result for "Graviton 3," and Tenstorrent's own 8-wide RISC-V architecture as it currently performs in their labs. While all of these are fascinating results in their own right, we're going to narrow in on the "Zen 4" (Genoa) and "Zen 5" results. We can see from the Frequency and TDP charts that "Zen 4" is clocked at 3.8 GHz as it's equal to the Xeon Platinum 8480+ (which itself boosts to 3.8 GHz in light threaded workloads such as this) so is therefore likely a variant of EPYC 9354 or 9454 with its TDP configured at the minimum 240 W. The unnamed "Zen 5" CPU is shown to be running at around 4.0 GHz with the same 240 W TDP, a tiny 5% bump in core clock, while delivering a substantial 30% jump in performance. The most interesting detail here is that nowhere is it listed—as with "Grace" and "Graviton 3"—that this is a projected result.

SiFive, Inc., the industry leader in RISC-V processors and silicon solutions, today announced that Tenstorrent, an AI semiconductor and software start-up developing next-generation computers, will license the new SiFive Intelligence X280 processor in its AI training and inference processor. SiFive will deliver more details of its SiFive Intelligence initiative including the SiFive Intelligence X280 processor at the Linley Spring Processor Conference on April 23rd.

Tenstorrent's novel approach to inference and training effectively and efficiently accommodates the exponential growth in the size of machine learning models while offering best-in-class performance.

The original "Nehalem" CPU microarchitecture from 2008 was pivotal to Intel, as it laid the foundation for Intel's mainline server and client x86 processors for the following 12-odd years. Glenn Hinton, the lead architect behind "Nehalem," announced that he is rejoining Intel after 3 years of retirement, to work on a new high-performance CPU project. Hinton states that his decision to rejoin Intel out of his retirement was influenced by Pat Gelsinger joining the company as its new CEO. Jim Keller, a CPU architecture lead behind several commercially-successful architectures, recently left Intel after a brief stint leading an undisclosed CPU core project. Keller later took up the mantle of CEO at hardware start-up Tenstorrent.

Pat Gelsinger leading Intel is expected to have a big impact on its return to technological leadership in its core businesses, as highlighted in Gelsinger's recent comments on the need for Intel to be better than Apple (which he referred to as "that lifestyle company") at making CPUs, in reference to Apple's new M1 chip taking the ultraportable notebook industry by storm. The other front Intel faces stiff competition from, is AMD, which has achieved IPC parity with Intel, and is beating it on energy-efficiency, taking advantage of the 7 nm silicon fabrication process.

Tenstorrent, a hardware start-up developing next generation computers, announces the addition of industry veteran Jim Keller as President, CTO, and board member. "Tenstorrent was founded on the belief that the ongoing shift towards ML-centric software necessitates a corresponding transformation in computational capabilities," said Ljubisa Bajic, Tenstorrent's CEO. "There is nobody more capable of executing this vision than Jim Keller, a leader who is equally great at designing computers, cultures, and organizations. I am thrilled to be working with Jim and beyond excited about the possibilities our partnership unlocks."

As CTO, Keller will lead Tenstorrent's efforts to be the hardware solution needed to address Software 2.0, the exciting industry shift towards using machine learning methods to solve problems previously addressed by traditional software. "Software 2.0 is the largest opportunity for computing innovation in a long time. Victory requires a comprehensive re-thinking of compute and low level software," Keller said. "Tenstorrent has made impressive progress, and with the most promising architecture out there, we are poised to become a next gen computing giant."

Jim Keller on Lex Fridman's AI Podcast shed some light on his thoughts on the microprocessor design fundamentals as he sees them. In a hour-and-a-half-long interview, he approaches Moore's Law and its much lauded - and ubiquitously repeated - death, as well as the need for both iterative and zero-point microprocessor design requirements.

Mr. Keller approaches the usual microprocessor design loop, where a company develops a new design from scratch and then looks at the most fundamental way of adding performance. Usually, he says, easy 10% performance increments can be found by simply looking at a design and increasing execution units - increase a buffer here, increase a cache over there, put in another add processor on this part of the pipeline. However, he also speaks of how this process in itself is limiting, inasmuch as doing this often will eventually guide processor designs towards a bottleneck and the diminishing returns problem, where any more additions made to the design don't seem to increase performance - mostly just adding complexity, area and power requirements, and generally convoluting a given design.

Intel today announced that systems designer-extraordinaire Jim Keller has departed the company citing personal reasons. Whether or not this is a blow to Intel likely depends on how far Jim Keller brought their Technology, Systems Architecture and Client Group throughout his two-year tenure at the company whilst serving as its Vice President. The semiconductor and chip architecture world isn't being driven by Mr. Keller himself, obviously; there are a number of architects and designers that bring the industry forward through their concerted efforts. However, it's hard to look past Jim Keller's pedigree when it comes to doing his job - if anything, AMD's Zen architecture is a testament to that, and has put Intel in the place we now see it in the CPU world.

To fill in the void, Intel has announced a reshuffling inside their Technology, Systems Architecture and Client Group. Jim Keller will still be serving with Intel for the next six months as a consultant, thus easing the transition. Read the full press-release below.

Today the Hot Chips program committee officially announced the August conference line-up, posted to hotchips.org. For this first-ever live-streamed Hot Chips Symposium, the program is better than ever!

In a session on deep learning training for data centers, we have a mix of talks from the internet giant Google showcasing their TPUv2 and TPUv3, and a talk from startup Cerebras on their 2nd gen wafer-scale AI solution, as well as ETH Zurich's 4096-core RISC-V based AI chip. And in deep learning inference, we have talks from several of China's biggest AI infrastructure companies: Baidu, Alibaba, and SenseTime. We also have some new startups that will showcase their interesting solutions—LightMatter talking about its optical computing solution, and TensTorrent giving a first-look at its new architecture for AI.

Intel's latest talent acquisition from rival AMD, as it builds a GPU product lineup, is Masooma Bhaiwala. "After 15+ amazing years at AMD, I have decided to take on a different opportunity... It was a truly fun ride, with an incredible team, during which we built some truly cool chips," she wrote in a LinkedIn post. According to her profile, Bhaiwala takes the role of Vice President, discrete GPU SoCs, and works under Intel's Graphics and Throughput Computing Hardware Engineering group headed by Raja Koduri.

Koduri's team has been a glassdoor for former AMD executives and tech-leads. While it has hired engineering talent such as Balaji Kanigicherla, Kalyan Thumaty and Joseph Facca; it has simultaneously lost client-graphics marketing talent, with the likes of Chris Hook, Heather Lennon, and Jon Carvill waltzing out of the company in less than a year of their association. Besides Koduri's Intel's most priced tech talent acquisition is Jim Keller, who is working on a future high-IPC CPU core design for the company. While working for AMD, Keller's "Zen" microarchitecture coupled with CEO Lisa Su's leadership have scripted one of the biggest turnarounds in Silicon Valley.

Intel at its architecture day revealed one of the more exquisite in-house designs for the company in recent years: a hybrid x86 chip that seems to imbibe from ARM's own big.Little design mantra. The new Hybrid x86 CPU that was announced takes this design choice in pairing a single, high-performance Sunny Cove core with four smaller Atom cores. This chip is built using Intel's Foveros manufacturing technology, which means a 22FFL IO chip serves as an active interposer, connected via TSVs to a 10nm die that contains both types of cores. The tiny chips measures just 12 x 12 x 1 mm (144 mm²), and looks to reduce footprint even further by including a POP (package on package) memory design.

The new Intel design is aimed at low-power environments, with the chip having been designed to work on a 2 mW standby power ratio, with less than a 7 W of power - for a big.Little five-core design and a 64 EU design with Gen11 graphics core. Intel's Jim Keller said that the company is testing the intricacies and advantages of this design internally, so more products based on this manufacturing and packaging mantra could pop up sometime in the future.

Darren McPhee worked 12 years for ATI and AMD. When he left AMD in 2015, he was one of the company's top marketing managers. For the last three years he has worked for various companies, but the surprise has come with Intel recruiting him to occupy the position of Product Marketing Manager in its 'Discrete Graphics' group, one of the most interesting initiatives in the recent times.

This division is working hard to develop a new family of discrete graphics cards that will theoretically compete with AMD and NVIDIA solutions. Intel has been steadily growing, and in fact Intel already signed Raja Koduri, AMD GPU architect, in November 2017. This firm has been attracting more and more talent from an AMD: Koduri was followed by Jim Keller, Ryzen Architect, and Chris Hook, who led AMD's Radeon Technologies Group Marketing Departmen prior to his move to Intel. These hires certainly make it clear that Intel is taking an increasingly promising project very seriously. We will have to be patient, however, because the firm already indicated in SIGGRAPH 18 that it will have its first models ready in 2020.

Intel today announced the acquisition of NetSpeed Systems, a San Jose, California-based provider of system-on-chip (SoC) design tools and interconnect fabric intellectual property (IP). Deal terms were not disclosed. NetSpeed's highly configurable and synthesizable offerings will help Intel more quickly and cost-effectively design, develop and test new SoCs with an ever-increasing set of IP. The NetSpeed team is joining Intel's Silicon Engineering Group (SEG) led by Jim Keller. NetSpeed co-founder and CEO, Sundari Mitra, will continue to lead her team as an Intel vice president reporting to Keller.

Intel is designing more products with more specialized features than ever before, which is incredibly exciting for Intel architects and for our customers. The challenge is synthesizing a broader set of IP blocks for optimal performance while reining in design time and cost. NetSpeed's proven network-on-chip technology addresses this challenge, and we're excited to now have their IP and expertise in-house.

Jim Keller, senior vice president and general manager of the Silicon Engineering Group at Intel

Chris Hook, the head of marketing at AMD Radeon Technologies Group (RTG), who resigned from AMD a few weeks ago, joined Intel. Hook will hold the position of head of discrete graphics marketing, confirming rumors of Intel making heavy investments into the development of a discrete GPU that can double up as a super-scalar processor, enabling the company to compete with NVIDIA and AMD for slices of the AI and blockchain computing gold-rush, with PC gaming as a fallback market. Jim Keller, Raja Koduri, and Chris Hook make up key names from AMD to have joined Intel in recent times. Keller was the lead architect of AMD "Zen," who after a brief stint at Tesla, joined Intel earlier this month.

Jim Keller, the VLSI guru who led the team behind AMD's spectacular comeback in the x86 processor market with "Zen," has reportedly quit his job at Tesla to join AMD's bête noire, Intel. Following his work on "Zen," Keller had joined Tesla to work on self-driving car hardware. Keller joins Raja Koduri at Intel, the other big former-AMD name, who led Radeon Technologies Group (RTG).

PC Perspective comments that big names like Keller and Koduri joining Intel could provide clues as to Intel's current state and the direction it's heading in. The company appears to be in a state of shake-up from a decade of complacency and lethargy in its core business. Koduri could be putting together a team of people familiar to him for a new clean-slate project. The last time Intel had a clean slate was ten years ago, with "Nehalem."