BTC $82,713.30 +1.06%
ETH $2,439.88 +1.55%
BNB $742.66 +1.73%
XRP $1.33 +2.35%
SOL $105.79 +1.00%
TRX $0.3309 -0.17%
DOGE $0.0815 +2.74%
ADA $0.2226 +2.23%
BCH $284.45 +2.20%
LINK $12.30 +2.23%
HYPE $80.69 +2.08%
AAVE $154.15 +2.49%
SUI $1.02 +2.34%
XLM $0.1815 +2.21%
ZEC $1,256.65 +9.58%
AAPL $338.51 -1.40%
AMZN $263.93 +0.32%
GOOGL $341.79 -0.78%
MSFT $528.96 -1.11%
META $699.80 -0.72%
NVDA $220.64 +1.66%
TSLA $379.88 +3.67%
SNDK $1,403.56 +4.25%
INTC $94.76 +4.11%
SPCX $161.54 +9.35%
MU $921.20 +3.62%
AMD $547.66 +2.86%
BTC $82,713.30 +1.06%
ETH $2,439.88 +1.55%
BNB $742.66 +1.73%
XRP $1.33 +2.35%
SOL $105.79 +1.00%
TRX $0.3309 -0.17%
DOGE $0.0815 +2.74%
ADA $0.2226 +2.23%
BCH $284.45 +2.20%
LINK $12.30 +2.23%
HYPE $80.69 +2.08%
AAVE $154.15 +2.49%
SUI $1.02 +2.34%
XLM $0.1815 +2.21%
ZEC $1,256.65 +9.58%
AAPL $338.51 -1.40%
AMZN $263.93 +0.32%
GOOGL $341.79 -0.78%
MSFT $528.96 -1.11%
META $699.80 -0.72%
NVDA $220.64 +1.66%
TSLA $379.88 +3.67%
SNDK $1,403.56 +4.25%
INTC $94.76 +4.11%
SPCX $161.54 +9.35%
MU $921.20 +3.62%
AMD $547.66 +2.86%

process

All
Article
Flash

first_img Blockworks: Seven cryptocurrency inference companies process approximately 211 billion tokens daily

Blockworks released a status report on cryptocurrency inference service providers, stating that global computing power investment is substantial and demand still exceeds supply; the market capitalization of liquidity crypto AI is approximately $24.9 billion, lower than Crusoe's $30.9 billion, and no decentralized computing network has disclosed verifiable revenue exceeding tens of millions of dollars.The report compares seven crypto-native inference providers on OpenRouter: Venice, Phala, Darkbloom, io.net, AkashML, NEAR AI, and Chutes. Over the past month, the seven have processed an average of about 211 billion tokens daily, accounting for about 1% of the platform's daily average of 21.9 trillion tokens. Venice processed about 87 billion tokens daily, with the team stating in July that its annual recurring revenue reached $100 million, with OpenRouter contributing less than 10%; Phala averaged about 54.6 billion tokens daily over the past week, up from about 5 billion in July.Chutes is the Bittensor inference subnet SN64, operating approximately 4,350 GPUs, with the team claiming to process 160 billion tokens daily, with OpenRouter channels averaging 5.7 billion. Akash has 283 active GPUs, down from a peak of about 897, with an annual leasing fee of about $3.7 million, and GPU incentives have ended. io.net's demand has nearly doubled over the past month to an average of 14.8 billion tokens daily.

first_img Intel and Qualcomm have raised processor prices, and AMD is rumored to follow suit

According to the Business Times citing supply chain sources, Intel has raised the prices of multi-generational PC processors since September, with the high-end Panther Lake seeing an increase of 15% to 20%. Qualcomm raised the prices of multi-chip platforms by about 10% on October 1, and AMD is reported to follow suit, with the increase, scope, and effective date still to be confirmed.The report states that Intel has raised prices three times since the end of last year. Starting from September 1, prices were adjusted differentially by product line, with the highest increase for high-end Panther Lake products, focusing on those equipped with high-end GPU tiles, rather than a simultaneous increase across the entire series; Meteor Lake and Arrow Lake saw increases of about 5% to 10%. Panther Lake adopts heterogeneous integration, with the CPU Compute tile produced by Intel 18A, and the high-end GPU tile outsourced to TSMC N3E. The largest increase this time is for high-end GPUs using TSMC's advanced process.The supply chain indicates that Qualcomm's Snapdragon X and X2 Elite platforms will increase by about 10% starting in October. AMD's Ryzen and EPYC heavily rely on TSMC's advanced processes, with the next generation of EPYC pushing towards 2 nanometers. Industry insiders say that process upgrades, product high-endization, and AI computing power demands are simultaneously driving up CPU prices, with AI, mobile, and high-end computing competing for TSMC's 3-nanometer capacity, while 2-nanometer is entering the ramp-up phase for mass production.

Firelight will roll back the blacklist process and unfreeze the affected wallets, involving 87 addresses and 117,500 FXRP

The on-chain guarantee protocol Firelight stated that some deposit addresses failed the fund screening, and the related funds have been unstaked and returned to the original addresses, involving 87 addresses and a total of 117,500 FXRP. Firelight mentioned that the current mechanism screens blacklisted addresses or addresses sanctioned by OFAC during fund deployment and throughout the entire cycle. If an address is flagged during deployment or within the cycle, the funds will remain in the protocol until the cycle ends and will not generate rewards, but can still participate in the guarantee module for the remaining time of that cycle.After the cycle ends, the funds will be returned, and the related wallets will be prohibited from accessing again. Firelight stated that this practice is unsustainable and will roll back the blacklist process in the coming days and unfreeze the affected wallets; the time and steps for retrieving funds will be announced separately. Security, KYC, and institutional standards remain unchanged, and Firelight will develop a new plan that does not affect users in the coming weeks and will seek community input before any changes. Flare co-founder Hugo Philion stated that due to compliance policies, 84 Flare addresses have been unstaked by Firelight, and he personally will issue equivalent subsidies to those addresses that are not sanctioned and not widely blacklisted.

first_img Bitcoin Core 32 enters final testing, optimizing transaction fee estimation and block processing

According to CoinDesk, Bitcoin Core 32 entered its final testing phase on September 14, with the first candidate version marked, and the official release is expected on October 10. This version will not change Bitcoin's consensus rules but will mainly optimize fee estimation, accelerate block processing, and fix several security vulnerabilities.In terms of fee estimation, Bitcoin Core 32 introduces a new estimator based on unconfirmed transactions, which can quickly lower the recommended fee rate when network congestion eases, compared to the original estimator based on historical block fees. Regarding block processing, nodes can now use multiple threads to fetch transaction data from the database simultaneously, with a default of 8 threads, reducing the waiting time for nodes to catch up with the blockchain.For security fixes, developers have addressed a command execution vulnerability that has existed in non-Windows systems since Bitcoin Core 24. This vulnerability could allow authenticated users to execute commands on the node host through specially crafted wallet names when the walletnotify feature is enabled. Additionally, four commands used to create partially signed transactions will default to using PSBT version 2 format; the new web server was found to have a memory exhaustion flaw during auditing, where 16 unauthenticated REST connections could raise the node's memory from 46 MB to about 3 GB in approximately one minute, and this issue has been fixed before the official release.

first_img SK Hynix stated that 3D DRAM development can utilize foundry processes

SK Hynix stated on the 9th that the development of the next-generation memory 3D DRAM semiconductor can utilize foundry processes. The day before, at the 2026 SK Hynix Future Forum held at the Supex Center in the Icheon campus, Vice President Kim Kyung-hoon and Son Ho-young presented the main technical directions for 3D DRAM, including the use of logic foundry processes for DRAM peripheral circuits, maximizing data bus bandwidth, and ultra-short distance transmission.3D DRAM is the next-generation DRAM that vertically stacks the originally planar storage cells to improve integration. The two stated that achieving this technology requires addressing not only design and heat dissipation issues but also challenges in fine interconnections, bonding, and process integration in the packaging field. As the boundaries between front-end and back-end packaging, foundry, and memory technology converge, collaborative optimization beyond existing fields will become more important. Son Ho-young mentioned that 3D technology is changing the technological boundaries between wafer fabs and packaging, and it is necessary to establish an optimization and new collaboration system that transcends existing fields alongside innovations in advanced packaging technology.In a related presentation, Professor Shin Chang-hwan from Korea University stated that 3D DRAM is not only about the vertical stacking of chips but also a technology that breaks the boundaries between memory and logic. As device miniaturization approaches its limits and the time and power consumption associated with data movement between systems and memory increases, the shift to 3D technology is inevitable. He proposed a 3D integrated design framework that links materials, devices, packaging, and architecture through artificial intelligence. SK Hynix President Kwon Oh-joon, in his speech, remarked that the past memory market resembled a straight road where the competition was about who could run faster, but now it resembles a constantly changing curve, making it important to grasp the speed and direction of external environmental changes and adapt flexibly in addition to internal capabilities.
app_icon
ChainCatcher Building the Web3 world with innovations.