Source: The content is compiled by Semiconductor Industry Observer (ID: icbank) from "
anandtech", thank you.
According to anandtech, after Dr. Shubu Mukherjee, the chief architect of Marvell ThunderX and ThunderX2, left the company at the end of last year and joined SiFive, Gopal Hegde, the chief architect of ThunderX3, also recently left Marvell and joined SiMa.ai. This is a big challenge for Marvell, which is preparing to make a big move in Arm server chips.
Reports indicate that one of the main driving forces in the Arm server space over the past few years has been the cohesion of different product teams trying to build next-generation processors to attack the dominance of x86 in the enterprise market. Many companies and products (Qualcomm's Centriq) have emerged (or Annapurna acquired by Amazon, Ampere Computing acquired Applied Micro) with varying degrees of success, which is related to the key personnel of each team. One of our readers recently updated us on the latest developments in this field. It was revealed that Gopal Hegde, vice president and general manager of Marvell's ThunderX processor business unit, has now left the company.
ThunderX was originally a flagship product of Cavium, which released two generations of products built on Arm: ThunderX and ThunderX2. We reviewed both systems: ThunderX was released in 2016 and ThunderX2 in 2018, with the latter considered a potential competitor for defined workloads and cloud and hosting providers if the price was right. We have seen some success with ThunderX2 and container deployments in HPC. However, during the deployment of ThunderX2, Marvell acquired Cavium for $5.5 billion, with the company looking to bolster its CPU, networking, and security assets. The acquisition was completed in July 2018.
It is worth noting that prior to the acquisition, Cavium stated that it had a regular roadmap for the ThunderX series of processors and servers. ThunderX3 was considered a radical design suitable for a wide range of markets. However, after the acquisition was completed, limited information was provided other than what Cavium said. It is usually strange for a company to remain silent for so long, especially for a product that based on previous release cycles, details were expected months ago. In talking with peers, we also realized that ThunderX3 may be later than expected.
However, to our surprise, Marvell contacted us (anandtech) in March of this year to discuss some of the details of ThunderX3. While this was not a deep dive into the architecture, we were happy to learn the highlights:
Their third generation is a microarchitecture designed based on the Arm v8.3+ instruction set, using four threads per core, scalable up to 96 cores or 384 threads per chip. It is coupled with eight DDR4-3200 memory controllers, 64 PCIe 4.0 lanes, and four 128-bit SIMD units per core. All thanks to TSMC's 7nm.
In our briefing, we were told that a key element of the chip design was to remain monolithic and that the chip would target a large number of cloud-based workloads and HPC workloads, with differentiators (such as SMT4) that allow for higher utilization of the underlying hardware. We expect to get more information about ThunderX3 at Hot Chips in August, as well as an update on the roadmap.
Gopal Hegde, who left this time, led the work done by Cavium and Marvell for the ThunderX project. Gopal cited his experience in building engineering teams around the world and as a core and platform engineer, helped define specifications, chips, brands and roadmaps as part of a wider team. According to his LinkedIn, this is a position Gopal has held since 2014, and before that he was the COO of Calxeda.
Having two key personnel losses in a short period of time inevitably brings some question marks. Some of my friends who are key industry leaders pointed out that Gopal's position change may be a turning point in Marvell's ThunderX project strategy. At this point, we know that although ThunderX2 has been generating revenue, we are not sure whether it is actually providing positive returns, let alone Marvell has enough revenue since the acquisition to justify the purchase.
ThunderX3 is a large chip built using a leading-edge process node, and its development will require Marvell to invest a lot of money in order to execute on time and at scale. Normally, when we think about the development of a new chip by a startup, we refer to the company’s “cash burn” as a way to measure the amount of money it will have to spend before the product can be launched. We rarely talk about it for the big companies, Intel, Qualcomm or Marvell, but it might be incitable here. Marvell doesn’t disclose these numbers for obvious reasons.
The loss of two high-profile individuals could cause Marvell to reconsider the viability of the entire project
. Qualcomm also shut down its Centriq server project because it failed to generate enough revenue, a process that began with the announcement that the senior vice president of the business unit had left the company. That being said, Marvell's ThunderX3 was still in its Hot Chips presentation in August, and the company still showed a positive attitude towards the release of ThunderX3.
We reached out to Marvell for comment on the development and future of the ThunderX processor family and spoke with Raghib Hussain, Marvell’s Chief Strategy Officer and Executive Vice President of the Networking and Processor Group, and Raj Singh. Raghib co-founded Cavium and held a senior position at Marvell after the acquisition. Raghib’s background is primarily in engineering roles.
“We made the strategic decision to consolidate the marketing and engineering teams for all of our processor businesses, including OCTEON and ThunderX, under the leadership of Raj Singh [in September 2019]. Looking ahead to the data infrastructure market, we see more similarities and synergies between the compute-focused segments. Under Raj’s leadership, the organization now represents the world’s largest and most scalable Arm-based infrastructure processor business, with solutions that scale from a few cores all the way up to the server class. Marvell’s Arm processor portfolio is unmatched in the data infrastructure market and is bringing optimized power, performance and TCO benefits to applications ranging from enterprise devices to 5G base stations, data center servers, and in the near future, emerging and emerging edge data center applications.”
Raj Singh, also a Cavium employee, has served as executive vice president of Marvell's processor business unit under Raveb Hussain since September 2019, responsible for server work.
Marvell also gave us the opportunity to ask questions about the business and where ThunderX fits into the structure. We were told that the ThunderX project fits into the category of “strategic bets” and that the processor families fall into the same category. As with other products, such as Auto Ethernet and investments in the data center, Marvell is aiming to provide data-centric solutions to customers and is in it for the long term. At this point in time, immediate returns are not the primary metric for the project, but the potential for returns in the future, which may not be a viable strategy if chip design is more like a startup. One must start getting in right away.
Compared to other companies in the space, Marvell explained that its ability to build IP that can be used in more than a dozen products in its portfolio, such as memory controllers that may appear in Octeon, ThunderX, Avera FPGAs, allows it to spread R&D costs across many projects from mature businesses, growth businesses, and strategic bets. Marvell pointed to the growth of its R&D spending relative to revenue, putting it as one of the highest percentages in the fabless semiconductor space.
According to semiconductor industry observers, Marvell USA has also laid off employees in its Arm server chip team. Let us wait and see about the future of this business.
★
Click [Read original text] at the end of the article to view
the original link of this article
.
*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.
Today is the 2350th issue of content shared by "Semiconductor Industry Observer" for you, welcome to follow.
Semiconductor Industry Observation
"
The first vertical media in semiconductor industry
"
Real-time professional original depth
Scan the QR code
, reply to the keywords below, and read more
Arm｜TSMC｜
CMOS
｜Wafer
｜
AI
｜FPGA
｜Matlab｜Packaging｜RF
Reply
Submit your article
and read "How to become a member of "Semiconductor Industry Observer""
Reply
Search
and you can easily find other articles that interest you!
Click to read the original text
to view the original link of this article!
京公网安备 11010802033920号