Photos of Intel’s first-gen crypto mining system appear online after the company’s reveal at the ISSCC conference
Previously in 2020, Intel disclosed its XPU endeavor, which described the diversification of their compute silicon through specialized architectures. This year, Intel assembled a fresh Custom Compute Group within the company’s Accelerated Computing Systems and Graphics (AXG) business division to create custom silicon platforms optimized for clients’ workloads. Intel integrates all the essential components into an individual self-contained computer called the Bonanza Mine System. This system’s appearance is similar to most other mining computers currently in production. The Bonanza Mine System incorporates four hash boards, an Intel FPGA control unit, four fans, and a programmable power supply. The photograph showcases is the complete reference system revealed by Intel. As you can see, the four boards are located behind the fan system, the control unit is located on the top, and the power unit is placed vertically on the left of the internal design. — Raja Koduri, Senior Vice President and General Manager of Accelerated Computing Systems and Graphics division, Intel On the top of the system, the control unit and Intel’s FPGA circuit (Field Programmable Gate Array) are furnished with a hard ARM cortex core that holds the chip that assists in mining. The system then produces the deployment of mining workloads, communicates and processes information to the phase lock loops (PLLs), and confirms the outcomes. Transmission in conjunction with the mining pools is accomplished over an Ethernet connection. Contact with each BonanzaMine ASIC with UART protocols and inter-integrated circuit links for the individual hash boards. Each hash board comprises a total of 75 BonanzaMine ASICs managed as a voltage stack that is 25 chips deep, permitting input voltages of 8.875 V split into 355 mV spanning across each combined ASIC group. Intel describes the process of improving efficiency upwards of five percent during mining while enhancing IR drop casualties. The separate BonanzaMine chips are connected with a 10Mb/s UART serial connection to the control unit for processing mining payloads and communication of workloads. All boards house a microcontroller unit to scan thermal and voltage levels. Die voltage stacking is used in the BonanzaMine ASIC accelerator for efficient power delivery. All ASIC chips present approvingly consistent negligible voltage levels and present attributes. The BonanzaMine die measures 4.14 x 3.42 mm in size (14.16 mm²) and is manufactured on a 7-nanometer process from an unknown fabrication facility. Each system houses 300 dies, equalling 4,248 mm² silicon chips. The chips are optimized for Bitcoin use only, indicating the ASIC basic building block is the dual SHA256 hash engines. Intel developed a 120-round fully unrolled SHA256 datapath to work with the SHA256 hash engines. Eight rounds during mining are eradicated, leveraging numerous Bitcoin-centric optimizations, such as early termination, iterate start nonce, and pre-computation of static fields. This process minimizes the area of the engine by eight percent. To further limit power consumption, all flip-flops within the SHA256 datapaths are substituted by a three-phase latch-based system clocking and employing level-sensitive latches. This limiting allows for 10% more efficiency due to reducing power levels by half. The process includes a scheduler at half the frequency to execute dual tandem scheduler datapaths to preserve energy for further digital mining. Intel has integrated 258 mining engines into each die utilized for the BonanzaMine. Numerous precomputed intermediaries and incremented start nonce hashes are calculated through the system controller, removing up to four compute rounds from the separate mining engines. The hash values are transmitted to respective engines via brute-force double-hash search spanning all 2e32 probable nonces for a golden nonce established on a 2e224 share target. When an engine hits a golden nonce, it conveys the information back to control for validation. Another two rounds of hashing are required to analogize to the block target. Each hash engine is programmed with a unique hash value from candidate Merkle root, which opens up to 19,350 concurrent workloads processed per individual board, with 77,400 performing simultaneously. Intel expresses that the hash engines make up 90% of the area on the die. The additional 10% is isolated to the lower part of the die, comprising the PLL, sensors, and GPIO (general-purpose input and output). The different mining engines on the BonanzaMine chip perform at only 355 mV, while each notch area operates at 750 mV. Intel’s BonanzaMine has demonstrated dependable 1.35-1.6 GHz operations at the system’s initial 355 mV. The 1.35-1.6 GHz operation levels for the BonanzaMine forge 116 GHash/s to 137 GHash/s at standard power levels of 7.5 W, equalling 54.8 J/THash.With 300 BonanzaMines, Intel declares the specialized crypto mining computer presents 40 THash/s at 3600 Watts; however, the system can achieve higher rates if required. Intel likened the BonanzaMines to the Bitfury Clarke and Canaan Avalon A9 during the conference. The BonanzaMine System is shown performing 55 J/TH at 47.7 TH/s, while the Bitfury Clarke reaches 40 THash/s at 56 J/TH and the Canaan Avalon A9 does 30 TH/s at 58 J/TH. With Intel currently displaying an extensive Bitcoin mining system, the benefit of miniature chip sizes allows for implementation into smaller crypto mining systems whose market component presently lacks competitive developments. Intel has not provided any launch date or pricing of the chip or system. Still, we recently reported that startup crypto company GRIID is Intel’s first customer to incorporate the new crypto chipset. Source: WikiChip Fuse, Intel
