Date: Nov 02, 2023
Click Count: 454
Pat Gelsinger, the CEO of Intel, has spent too much time at EMC and VMware. Now, Intel is looking to spin off its FPGA business, which represents a small but promising segment in its data center and edge computing divisions.
It has always been somewhat puzzling that EMC, a company known for traditional storage arrays and considered a pioneer in datacenter-class storage during the Dot Com era, acquired VMware in December 2003 for $635 million in cash, shortly after the Dot Com bubble burst. EMC was seeking a high-return investment for its cash reserves, but in hindsight, some might argue that going public could have been a better option for VMware.
To protect VMware's status, EMC had to maintain a neutral position, as its partners were using VMware's virtualization solutions alongside competitors like Microsoft and Red Hat, creating complications. Consequently, EMC never developed a hybrid server-storage platform capable of reshaping data center architectures.
To be fair, such deals often make financial sense, much like Michael Dell and Silverlake Partners' $67 billion acquisition of EMC (and VMware) in October 2015. Eventually, it made sense to spin off VMware, allowing it to operate independently, leading to Broadcom's acquisition of VMware for $61 billion for its financial considerations. There were various twists and turns, including Michael Dell taking his namesake company private a year before announcing the EMC-VMware deal to avoid Wall Street's scrutiny.
Significantly, neither Dell nor Broadcom leveraged VMware and EMC to create a new type of server-storage-networking hybrid in the data center, as attempted by Nutanix. Instead, these deals focused on installed bases, subscription and maintenance revenues, and maximizing value from customers locked into the ESXi/vSphere stack.
In March 2015, when rumors swirled about Intel potentially acquiring FPGA manufacturer Altera, we conducted an in-depth analysis of the FPGA market and its potential transformation by SmartNICs and high-speed computing, where FPGAs excel over CPUs or GPUs. The key factor was software. While dedicated hardware with ASICs offered efficiency and CPUs with high-level software algorithms provided flexibility, FPGAs bridged the gap with programmable logic and hard-blocked components like DSP engines. However, programming in VHDL was a prerequisite, making it essential to tune the code for optimal performance.
When Intel acquired Altera for $16.7 billion in June 2015, we believed that FPGAs could revolutionize the data center. FPGAs were suitable for workloads with low latency and high-performance requirements that didn't change often, albeit at a price/performance point that couldn't match custom ASICs, which were becoming costlier with each node shrink.
FPGAs a hit on Microsoft's Azure cloud, accelerating workloads and offloading network and storage functions from CPUs. Discussions included hybrid CPU-FPGA packages, although these never seemed to materialize due to the static compute ratios, a preference held by system architects who wanted more control over these ratios.
Between 2015 and 2020, Intel ventured into numerous areas, from starting a GPU compute business for the third time to acquiring companies like Barefoot Networks and the Cray and QLogic InfiniBand interconnect businesses (later sold or shut down). Intel also purchased Nervana Systems and Habana Labs for $2.35 billion, anticipating the potential replacement of CPUs, GPUs, and FPGAs by custom ASIC neural network processors.
While Intel appeared capable of selling anything, it primarily succeeded in the "supply win" category, while others excelled in delivering cutting-edge CPUs and GPUs. FPGA sales continued on their steady path, as experienced FPGA users continued their use of this technology. The verdict is still out on neural network processors as far as we are concerned. Intel has hedged its bets in so many directions that it lost sight of its core strategies.
<< Previous: AMD Secures Two Key Contracts: Oracle Chooses MI300X, IBM Utilizes FPGA
1
2
3
4
5
6
7
8
Comparison of the latest released FPGAs from Xilinx, Intel, and Lattice
9
10
32 MACROCELL CPLD
CPLD CoolRunner XPLA3 Family 750 Gates 32 Macro Cells 95MHz 0.35um Technology 3.3V 48-Pin CSBGA
CPLD CoolRunner XPLA3 Family 750 Gates 32 Macro Cells 95MHz 0.35um Technology 3.3V 44-Pin PLCC
FPGA Spartan-II Family 15K Gates 432 Cells 263MHz 0.18um Technology 2.5V 144-Pin TQFP EP
FPGA Spartan-II Family 15K Gates 432 Cells 263MHz 0.18um Technology 2.5V 100-Pin VTQFP
Support