At Mining Disrupt's online edition on July 22–23, 2020, the crypto mining industry faced a distinct challenge: how to remain profitable when coin values fluctuated wildly. Julia Milstein, who manages operations and business development for Avnet ASIC Israel, opened the presentation by outlining the shift in hardware requirements. She explained that the financial instability of individual cryptocurrencies was pushing miners away from single-coin dedicated hardware. The goal was to create a platform that could switch algorithms easily, allowing the same chip to handle different algorithms without significant downtime or capital loss.
Milstein highlighted that Avnet, a leading worldwide distributor of electrical components, had a long history in custom ASIC design. She noted that the company had successfully completed 350 projects over 30 years, including high-performance miners for specific networks. However, the market was moving toward generic, GPU-like solutions. The partnership with AzurEngine was designed to bridge the gap between dedicated silicon and general-purpose processing. Avnet would leverage its manufacturing and distribution capabilities to bring AzurEngine’s proprietary architecture to market quickly. This collaboration aimed to provide a turnkey solution, managing the entire lifecycle from prototyping to mass production, thereby reducing the entry barrier for miners looking to adopt new technology.
more and more customers are pushed to consider generic gpu-like solutions
Julia Milstein · 2:37
Space-Domain Processing
Pavel Vilk, Director of Engineering at Avnet ASIC, took over to explain the technical foundation of the AzurEngine chip. He contrasted traditional CPUs, which process instructions sequentially in a time domain, with the reconfigurable parallel processing (RPP) chip. This architecture moves programmability from the time domain to the space domain. Instead of a single element executing instructions over time, the RPP chip distributes instructions across different process elements simultaneously. Data flows through a sequence of these elements, allowing the program to be executed in parallel. This design is particularly suited for tasks with massive data parallelism, a characteristic shared by both artificial intelligence algorithms and cryptocurrency mining routines.
moves the programmability from time domain to space domain
Pavel Vilk · 5:46
Vilk detailed the internal structure of the chip, which includes a sequential unit to prepare tasks and a memory unit to load data in parallel. The system features up to 32 process elements, each containing 32 values and various types of multipliers and accumulators. This hardware supports a wide range of data types, including 32-bit floating point, 32-bit integer, half-precision floating point, and 8-bit integer instructions. This diversity is key to the chip’s versatility, allowing it to adapt to the specific mathematical requirements of different coins. Vilk also noted that the chip supports native CUDA C++ language, a standard originally developed by Nvidia. This compatibility lowers the barrier for software developers to port their existing mining code to the new platform.
Efficiency Over Raw Power
The presentation included a direct comparison between the AzurEngine RPP R8, manufactured at 14 nanometers, and Nvidia’s high-end GPUs, specifically the Tesla V100 and Tesla T4. Vilk presented data suggesting that while the AzurEngine chip provided lower raw throughput for 32-bit integer calculations, it achieved this with a significant reduction in power consumption. The RPP R8 consumed only 10 watts, whereas the Tesla V100 consumed 300 watts and the Tesla T4 consumed 70 watts. When dividing operations by power, the AzurEngine solution yielded a ratio of 0.5, compared to 0.05 for the Nvidia GPUs. By their figures, that is about 10 times less power for the same work.
this kind of parallel architecture allows to achieve a 10x power efficiency
Pavel Vilk · 7:14
Vilk argued that for many cryptocurrency algorithms, the goal is not just to calculate as fast as possible, but to consume as little power as possible. The internal memory of the AzurEngine chip also reduces the need for access to external memories like DDR or HBM, further contributing to energy savings. The implication was that any algorithm that could be run on an Nvidia GPU could be adjusted for the AzurEngine solution, but with a magnitude of power savings that would significantly improve the miner’s profit margin.
The Hybrid Model
The presentation made clear that this was not just a chip, but a service. Avnet’s role extended beyond mere distribution; they were the engineering and logistical backbone for AzurEngine. Milstein explained that AzurEngine had chosen to partner with Avnet to leverage its experience in ASIC design and manufacturing. Avnet was set to launch the silicon and the system-level accelerator cards, intending to sell them as finished products globally. This hybrid model combined the innovative architecture of a startup with the supply chain reliability of a major distributor. For miners, this meant they could access a cutting-edge multi-coin platform without having to manage the complexities of custom silicon procurement themselves.
Flexibility as a Survival Strategy
In his closing remarks, Vilk summarized the philosophy behind the AzurEngine platform: flexibility is essential in a volatile market. He warned that miners cannot lock themselves into a single coin, as they depend on market activity and need the ability to move from one coin to another as conditions change. However, this flexibility must be balanced with power efficiency. The AzurEngine reconfigurable parallel processing solution was presented as the tool that could maximize computing power while reducing power consumption, ultimately providing better efficiency than existing standard solutions. The argument was that the ability to switch algorithms was worth paying for, as it protected miners from the obsolescence of single-coin hardware.
you cannot in the current market lock yourself only on the [single] coin
Pavel Vilk · 9:41
Six years on, the pitch reads as an early version of a theme that now runs through Intelligence Disrupt: the same silicon, power and cooling questions serve both mining and AI. The 10x figure was the companies' own claim, measured against Nvidia's Tesla V100 and T4 data center GPUs of the time. The conversation continues at Intelligence Disrupt 2027, March 22 to 24 in Irving, Texas. When this session was recorded, Bitcoin traded at $9,530; on October 11, 2026 it was $83,086.