tech lead and research scientist, calmly computing @MegaETH. All opinions are my own.
- Tweets 428
- Following 1,429
- Followers 3,027
- Likes 4,500
Pinned Tweet
Many folks were suprised by how fast I decided to leave my PhD program at Harvard and join MegaETH two months ago. To be honest it was such an easy choice. Any systems researcher would do the same after reading @yilongl_megaeth's analysis of Etheruem performance and chatting with @yangl1996.
19
6
174
58,053
关于MegaETH（以太巨星），我想稍微给中文圈的朋友们写写，因为我看到好多批评者说它是空气链，可能不太了解这个项目。看完了买不买nft是小事，至少别骂得太绝对、事后损失听众嘛。
MegaETH是不是空气链，高性能叙事是不是像Solana那样纸上谈兵，看完自有答案。或者不看也行，30天内测试网上线，保证第一天就大于 10k 真实TPS，10ms延迟，到时候咱再重聚。
技术
既然咱今天是闲聊，那就不说太细，我把核心要点讲讲。
1. 中心化排序器（sequencer）
MegaETH 的目标是最极致的性能。这个极致代表硬件的极致，也就是再来一个团队，不管你写码再怎么厉害，这也就是极限了，因为我们干之前就算清楚了理论上界。我们选择用中心化sequencer（这是Vitalik认为的blockchain scaling的必经之路），一个块后无需等待，从而才能把出块延迟做到10ms量级。在这个理念下，我们设计了新的验证数据结构替换Merkle Tree，实现了新的数据库、EVM编译器、并行EVM引擎等等。
2. 去中心化验证
有人批评说，“你们说了半天，结果就是个单机节点，演都不演了，还不如人家solana”。
其实恰恰相反，我们的去中心化程度更高，这是因为我们的设计里，节点的分工做得好。一条链要做的事情有很多，执行、排序、验证、共识等等，不同的事情应该交给不同的角色来做，这样彼此都省心省时。MegaETH为了追求极致性能，选择完全不让L2承担「共识」的负担，造成资源和时间的浪费。排序器就只执行交易且排序，而去中心化的安全层交给无可媲美的以太坊，DA层交给性能和成本都极其优秀的Eigen DA。
有人又说了，“那你这个性能说到底不还是单机节点上搞优化，有什么好吹的”。
又恰恰相反，因为一旦排序器的性能足够高，网络带宽根本承担不住那么高的交易流量，而提高网络带宽的要求又不满足去中心化的要求（总不能让大家都去Amsterdam租服务器吧？），所以必须优化同步的网络协议。
我在这条推文里（nitter.app/jinfizzbuzz/status/185… ）
计算并讲解了state sync的话题，在这条推文里（nitter.app/megaeth_labs/status/18… ）讲解了MegaETH上各节点的硬件要求。如果你跑过其它高性能链节点，一定明白这里的硬件要求有多去中心化。
总而言之，MegaETH通过节点分化的设计，最大程度发挥了L2架构的优势，抛掉了自己承担共识的负担，从而满足性能极限化。
生态
MegaETH追求极致性能，是因为我们坚信属于币圈的iPhone时代仍未到来，无数种应用仍被现有性能限制。但话说回来，其实我们都有点厌倦了无尽的性能争论和infra之战。一条没有生态的链就是死路一条，@hotpot_dao 在币圈摸爬滚打多年比谁都懂。
从最开始，发展生态、找到有潜力和热情的开发者就是我们的首要重点。@megamafia 以太巨星黑手党，我们的旗舰加速器项目因此诞生，让我们和一批极度优质的项目达成了深度绑定。去年10月，我们花了整整一个月在清迈举办Mega Zu，所有开发者线下聚在一起，交流信息，培养感情，互相反馈，共同进步。
看看下面的列表，你会明白，MegaETH究竟让多少种先前的不可能成为可能，而它们只是开始。
@GTE_XYZ 主打性能、提供cex体验的链上交易所，融资已$10M
@CapLabs_ 极高收益的稳定币引擎，pre-seed融资$1.9M
@noise_xyz 还交易meme？过时了，来交易「流行趋势」
@tekofinance 实时借贷市场
@Funes_World 真实世界建筑博物馆（还有谁家有这么文艺有理想的项目？）
@Euphoria_fi 衍生品交易平台，极度上瘾的UX
@Hop_Network 最匿名、最快速的链上VPN
@lemonade_events 线下活动和社区管理平台
@pumpparty 线上娱乐媒体，目标受众是千千万万圈外用户
@avon_xyz 高效、灵活的借贷平台
@Showdown_TCG 世界冠军来做的线上棋牌游戏，MegaZu所有人的心头好
@valhallaperps 链上杠杆平台
@TryNectarAI 线上AI好友
@trysweep 直播中实时预测，猜猜你在看的主播黑神话下一关能不能打通
@awe_box 链上3d游戏引擎，助力链上游戏高效开发
A quick explainer on Node Specialization in MegaETH and how we are pushing blockchain performance to its limits
by assigning each node type a specific role, optimized for both hardware and functionality. Here’s a look at how each node type contributes to a high-performance, real-time blockchain:
Sequencer: The computational core, processing transactions and sending state updates to other nodes every 1-10ms. Equipped with >100 cores, 1-4 TB RAM, and 10 Gbps network bandwidth, it ensures rapid transaction handling and state dissemination.
Replica Nodes: These nodes store the full state but don’t re-execute transactions, relying instead on the prover network to verify the sequencer's output. The majority of MegaETH nodes are replicas, optimized for app developers and infrastructure providers to serve frontend applications efficiently, reducing hardware requirements while improving UX.
A replica node requires typically 4-8 cores, 16 GB RAM, and 100 Mbps network.
Full Nodes: These nodes are used by power users, storing the entire blockchain state and re-execute every transaction got from the sequencer. They require slightly better hardware than replica nodes but are still more affordable than the sequencer.
Provers (OP Challengers): Tasked with validating transactions, these nodes rely on stateless verification to secure state integrity. To maintain a low hardware requirement like 1 core and 0.5 GB RAM, each prover node only needs to check a portion of state. With a special proof exchange and validation protocal that we designed, the prover network covers the entire ledger.
Light Clients: Lightweight nodes with minimal hardware requirements, they don’t store full state and instead provide end-users streamlined access to the blockchain without heavy computational needs, maximizing accessibility.
MegaETH’s node specialization balances high performance with decentralization. Sequencers drive rapid processing, while lightweight replicas and full nodes ensure decentralized data access. This approach empowers MegaETH to deliver a web2-like UX with over 100k TPS on a real-time blockchain.
75
11
419
118,649
At MegaETH, we’re researchers who verify every claim ourselves. From day one, we focused solely on performance analysis to uncover bottlenecks across tradeoff spaces (see our last post on node specialization). Here’s what we learned.
We benchmarked an Ethereum Reth client node syncing from blocks 17M to 17.1M on a sequencer-level machine with 512GB RAM, yielding a 998.9 TPS. Here’s the time breakdown:
• Execution Time: 1612.54 s
• Merkleization: 9303.68 s
• DB Write: 3453.68 s
Key insight? Merkleization took 5.77x longer than execution, with tree traversals and random disk I/Os dragging down performance. This revealed the need for a faster, optimized authentication structure that minimizes I/O. DB writes also took 3.4x longer than execution, a pattern that impacts replica nodes too—where the two factors alone bottleneck network throughput because they don’t re-execute transactions.
As these three steps can be pipelined easily, ideally we would want to reduce merkleization and DB write times to below execution time, so that they no longer throttle the system, boosting throughput. But that’s not enough. With JIT compilation decreasing execution time and enabling OMEGA computation-intensive apps, we must aim for more aggressive optimizations.
The result? Our new authentication data structure now brings single-thread computation time to just 96s (96.9x speedup), and scaling nearly linearly with CPU cores. Plus, we’ve replaced MDBX with a new KV store for even greater speed.
Stay tuned for more!
21
20
250
39,473
If you’re wondering what challenges MegaETH is tackling, one of the biggest is the streaming EVM. MegaETH’s streaming EVM balances low-latency block streaming and high throughput.
What’s our latency target? 1ms. This is because power users, like matchmakers collocating with the sequencer, generally expect network latency as low as 1ms. Going lower doesn’t help when physical network latency becomes the limiting factor. Achieving this target is critical to delivering the real-time experience our users need.
However, optimizing for latency and throughput simultaneously is a much harder problem than optimizing for one of the metrics individually. This is because many of the steps in the block building process, such as updating the state trie and committing state changes to the database, are traditionally inefficientat small block sizes.
As a concrete example, we ran an experiment with recent Ethereum mainnet transactions and tested two strategies for committing state changes to thedatabase: large and small blocks.
* Large blocks (10,000 txns per block): each commit takes 150ms.
* Small blocks (100 txns per block): each commit takes 8ms.
The performance degradation when switching from large to small blocks may not be obvious if we look at the cost per commit. However, it becomes apparent when we look at the cost per transaction.
* Large blocks (10,000 txns per block): 15us per transaction.
* Small blocks (100 txns per block): 80us per transaction.
That is, the per-transaction cost increased by 5.3x when we reduced block size by 100x. In other words, system throughput decreased by 5.3x as we reduced latency by 100x. One does not get the best of both worlds!
This drop in throughput can be attributed to several factors:
- Smaller batches result in more database writes overall.
- Storage devices tend to be more efficient with larger batched writes.
- More database software overhead (e.g., data structure traversal) when handling smaller batches.
The drop in throughput would be fine if all we wanted was low latency. However, our vision of a realtime blockchain is one which preserves responsiveness even under high load. By carefully analyzing such behaviors, we are tackling the delicate balance of ensuring our streaming EVM can achieving both high throughput and low latency.
p.s. You may have noticed we’re focusing on the problem rather than the solution — and that’s intentional! This series is all about showing how we analyze problems as systems researchers. We focus on making the right assumptions, testing them, and verifying our conclusions. We believe that clearly defining the problem is crucial, because once we all align on the problem, the solution is within reach.
3
20
191
37,180
At MegaETH, we've emphasized that we're more than just a sequencer. Today, let’s dive into one reason why: the state sync challenge.
State sync involves transferring the latest state changes from the sequencer to other nodes—a complex problem in blockchain infrastructure that’s often overlooked, even among high-performance chains.
Consider a simple transaction like an ERC-20 token transfer. Each transfer involves three main pieces: the sender's balance (20 bytes for the address + 32 bytes for the value) and two storage slots (each 64 bytes) in the ERC-20 contract (20 bytes for the address). That brings us to around 200 bytes per state diff. Now, at 100k TPS, this equates to 152.6 Mbps of raw bandwidth. While real systems use compression and other optimizations, state sync still places enormous demands on nodes with limited network resources.
But TPS isn’t the only concern. We’ve promised ultra-low latency, which means smaller and more frequent blocks. Even if these blocks contain the same transactions, smaller batch sizes mean lower compression ratios and increased bandwidth needs.
Let’s break it down: with a 10ms block time and 1,000 transactions per block (to hit 100k TPS), we replayed Ethereum transactions from blocks 20M to 20.1M to analyze the bandwidth demands of a Replica node.
Key results:
- Total MegaETH blocks: 16,302
- Avg block size: 1196.93 KB, broken down into:
- Transactions: 485.45 KB
- Receipts: 465.99 KB
- State changes: 245.48 KB
This means a Replica node would need about 958 Mbps to stay in sync—a nearly impossible requirement. With standard zstd compression, the average block size reduces to 345.97 KB (3.46x compression), requiring 277 Mbps. But 277 Mbps is still insufficient; live sync can’t monopolize all available bandwidth, and network providers often overpromise speeds. In reality, the bandwidth requirement remains unmanageably high.
To tackle this problem, we designed a new network protocol that achieves over 80% bandwidth utilization while minimizing global broadcast latency.
For a simple demonstration, check out the animation below: it shows a bunny replica node waiting for water (state sync data) from the network. With limited network resources and poor bandwidth utilization, the bunny remains thirsty, regardless of how high the TPS is on the sequencer. But with MegaETH’s optimized networking protocol, the bunny is finally well-fed and mega-happy!
12
19
172
23,805
Today I met @0xBreadguy for the first time, and I started to believe @megaeth might as well be a modeling agency
5
41
7,738
Just looked carefully into $MELANIA’s launch—where a massive wave of purchases hammered Solana—their real TPS actually never spiked but degraded significantly, exposing Solana’s struggle with heavy network congestion.
Even worse, they didn’t acknowledge it as "downtime", although it effectively was for end users who couldn’t get transactions through.
MegaETH will NEVER be this intellectually dishonest. By leveraging Ethereum for security and radically redesigning the networking protocol, it will deliver a truly decentralized and truly real-time chain.
6
3
32
2,361
MegaETH 测试网对所有users上线了。已经给 @megamafia 用了一阵子了，保证用户第一天来就有app用。
喷我们是PPT空气链的，来玩玩吧。我曾说保证>10k TPS，10ms延迟。
我们的第一个版本是20k TPS。
关于MegaETH（以太巨星），我想稍微给中文圈的朋友们写写，因为我看到好多批评者说它是空气链，可能不太了解这个项目。看完了买不买nft是小事，至少别骂得太绝对、事后损失听众嘛。
MegaETH是不是空气链，高性能叙事是不是像Solana那样纸上谈兵，看完自有答案。或者不看也行，30天内测试网上线，保证第一天就大于 10k 真实TPS，10ms延迟，到时候咱再重聚。
技术
既然咱今天是闲聊，那就不说太细，我把核心要点讲讲。
1. 中心化排序器（sequencer）
MegaETH 的目标是最极致的性能。这个极致代表硬件的极致，也就是再来一个团队，不管你写码再怎么厉害，这也就是极限了，因为我们干之前就算清楚了理论上界。我们选择用中心化sequencer（这是Vitalik认为的blockchain scaling的必经之路），一个块后无需等待，从而才能把出块延迟做到10ms量级。在这个理念下，我们设计了新的验证数据结构替换Merkle Tree，实现了新的数据库、EVM编译器、并行EVM引擎等等。
2. 去中心化验证
有人批评说，“你们说了半天，结果就是个单机节点，演都不演了，还不如人家solana”。
其实恰恰相反，我们的去中心化程度更高，这是因为我们的设计里，节点的分工做得好。一条链要做的事情有很多，执行、排序、验证、共识等等，不同的事情应该交给不同的角色来做，这样彼此都省心省时。MegaETH为了追求极致性能，选择完全不让L2承担「共识」的负担，造成资源和时间的浪费。排序器就只执行交易且排序，而去中心化的安全层交给无可媲美的以太坊，DA层交给性能和成本都极其优秀的Eigen DA。
有人又说了，“那你这个性能说到底不还是单机节点上搞优化，有什么好吹的”。
又恰恰相反，因为一旦排序器的性能足够高，网络带宽根本承担不住那么高的交易流量，而提高网络带宽的要求又不满足去中心化的要求（总不能让大家都去Amsterdam租服务器吧？），所以必须优化同步的网络协议。
我在这条推文里（nitter.app/jinfizzbuzz/status/185… ）
计算并讲解了state sync的话题，在这条推文里（nitter.app/megaeth_labs/status/18… ）讲解了MegaETH上各节点的硬件要求。如果你跑过其它高性能链节点，一定明白这里的硬件要求有多去中心化。
总而言之，MegaETH通过节点分化的设计，最大程度发挥了L2架构的优势，抛掉了自己承担共识的负担，从而满足性能极限化。
生态
MegaETH追求极致性能，是因为我们坚信属于币圈的iPhone时代仍未到来，无数种应用仍被现有性能限制。但话说回来，其实我们都有点厌倦了无尽的性能争论和infra之战。一条没有生态的链就是死路一条，@hotpot_dao 在币圈摸爬滚打多年比谁都懂。
从最开始，发展生态、找到有潜力和热情的开发者就是我们的首要重点。@megamafia 以太巨星黑手党，我们的旗舰加速器项目因此诞生，让我们和一批极度优质的项目达成了深度绑定。去年10月，我们花了整整一个月在清迈举办Mega Zu，所有开发者线下聚在一起，交流信息，培养感情，互相反馈，共同进步。
看看下面的列表，你会明白，MegaETH究竟让多少种先前的不可能成为可能，而它们只是开始。
@GTE_XYZ 主打性能、提供cex体验的链上交易所，融资已$10M
@CapLabs_ 极高收益的稳定币引擎，pre-seed融资$1.9M
@noise_xyz 还交易meme？过时了，来交易「流行趋势」
@tekofinance 实时借贷市场
@Funes_World 真实世界建筑博物馆（还有谁家有这么文艺有理想的项目？）
@Euphoria_fi 衍生品交易平台，极度上瘾的UX
@Hop_Network 最匿名、最快速的链上VPN
@lemonade_events 线下活动和社区管理平台
@pumpparty 线上娱乐媒体，目标受众是千千万万圈外用户
@avon_xyz 高效、灵活的借贷平台
@Showdown_TCG 世界冠军来做的线上棋牌游戏，MegaZu所有人的心头好
@valhallaperps 链上杠杆平台
@TryNectarAI 线上AI好友
@trysweep 直播中实时预测，猜猜你在看的主播黑神话下一关能不能打通
@awe_box 链上3d游戏引擎，助力链上游戏高效开发
15
1
34
9,335
I saw six MegaETH stickers, can u recognize them?
Forget everything you know about testnets—Succinct just flipped the game.
This is the coolest testnet I’ve ever seen.
It’s cute, it’s smooth, and it’s got an ostrich (yes, really)
In this thread:
- What Succinct is and why it matters
- How to join the testnet
- What activities to do (games included)
- And, oh yeah, MacOs Easter eggs (👀brainrot memes, Steve Jobs AI, and more)"
11
20
2,887
Gave a talk about MegaETH’s system design yesterday at @MEGAZuzalu
Love that cyberpunk projector 📽️
2
21
1,478
great coveramazning
5
23
1,464
amazing game. going to recommend this to my poker-addicted friends.
20 years ago, poker was the king of card games. A while later, Hearthstone took over. Now, it’s time for the next big thing.
It’s time for Showdown!
Showdown is brought to you by @OndrejStrasky, the #1 ranked player in Magic, and @StanCifka, the #1 ranked player in Hearthstone.
3
19
1,181
MegaETH presents users with various tradeoffs, enabling them to make their own choices
"The mini blocks for the super low latency I don't know how compatible they are with current ETH apps."
- full EVM compatibility, but also new abstractions to bring ux to the next level, it's up to the app devs which one to use. They can simply deploy their existing contracts and optimize progressively. As any chain developer knows, there’s no free lunch—you can barely have both backward compatibility and extreme high performance without introducing new abstractions, which we approach with caution and transparency.
"Finalised transactions are actually fully committed and until this occurs could be reordered or rolled back. One of the reasons I think Mega's architecture is dangerous."
- that's for all l2s, not just MegaETH.
"Mega has much further to go with some really difficult optimisations (eg JIT compiler)."
- already implemented, it is difficult but not to us.
"People talk about security but the real problem is fault tolerance. If that sequencer goes down the chain stops."
- that's easily solved by adding backup sequencers. I don't even understand why you bring it up.
as an l1 person, your comments are not neutral enough.
5
1
17
1,002
“Measure then build”
“Intellectual honesty”
“We say what we mean and we do what we say”
Just realized there’s a single Chinese word behind them called 知行合一
5
15
1,170
如果有人有wl觉得太傻逼了一定不买，请联系 @0xAlexon ，他会照单全收的
Humorous to see the responses to this.
This is effectively an ICO, and a very interesting way to use NFTs - a way that many have been preaching for.
Maybe people have forgot that MegaETH raised $10M in 3 minutes on @echodotxyz.
The total raise from this mint, coupled with the projected FDV, makes for it to be very solid in my opinion.
The inability to read in crypto is staggering, but truly hope we never change.
If you don’t want your WL, hit my DMs 🫡
13
1
14
14,137
You might want to rethink what the word “cache” means, especially when you don’t mention you are comparing worst-case scenarios/tail latency (and they are still wrong)
13
2,132
would be great for the buy button to change to a green "Success" synchronously after receiving tx confirmation
1
9
277
“但它要怎么做不知道”
技术的可以看白皮书、@yangl1996 的贴、和我的推
不想读文字可以看@0xAlexon 的YouTube视频，对比megaeth和monad
社区建设和生态项目可以看@hotpot_dao 的推
“这几年，大家对各种更好性能、更有特色的公链开始免疫了”
同意，广告太tm忽悠人了，solana那么高的理论性能遇到melania的拥堵期只能跑出500不到的真实tps。
我们保证megaeth不一样，测试网第一天就能看到10ms延迟和>10k tps，三十天以内，敬请期待。
3
8
631
Alexon老师的分析含金量就不多说了，youtube频道老观众了
Ferryboat Research & Friends（Vol.1 Megaeth）
小A和他的朋友们（Vol.1 Megaeth）
第一期社区拖了很久，权益我放在后面，这个时间下愿意被咱割的，应当都是真有缘了。
为了有诚意一点，我记录一下我的思考，可能不对，但真诚。
如今的Crypto氛围下，怎么参与生态变成了个技术活。估值被重塑下，首先被放弃的交互参与者（撸毛）成为了估值改革下第一批倒下的人；TVL战争下签订大户保底APY（都是用Token发放）也是各类吸储项目传男不传女的增长法门。
这个过程里让我想起新消费时期，因为品牌和投放做的还行，朋友拉我一起接了某品牌的增长，他们为了融下一轮梭哈了账上的接近3000万RMB。
于是，我和当时的几个朋友，每个月帮他们消耗了数千万RMB的广告费给巨量引擎，保底ROI1.2。大白话说，每卖出一单要亏损7RMB。
后来这套策略确实让他们销量起来了，他们拿到字节跳动几亿的融资，有意思的是，这几亿融资的条件是用几乎等额巨量引擎的年框完成的。
后来蒙眼狂奔下，在GMV过了十亿之后非常凶险的进行了供应链和流量的买人和降本增效，如今他们靠着卡位了电商的高层级活了下来，但因为融资的钱又还给广告年框，所以始终现金较少，利润较少，但是幸存。
看上去是不是这个策略是奏效的？如果吹牛逼的情况下，我可能也愿意说到这里就打住了。
但是大家不知道的是，其他品牌看到这个做法奏效，马上跟进用同样的打品策略，最后做流量的公司赚了上游VC的现金，消费公司亏钱卖出去了货，唯一的区别就是能不能续上命融到后一轮钱，事实上就是我和朋友经手过的加上了解实际情况的公司，总共也就活了这一家。
后来大家发现好的消费公司几乎不要钱或者只融一轮钱，而消费估值也就是20倍PE，从资本效率上来说不够。因为卖货和互联网需要网络效应还不太一样，找到人群度过之后，能赚钱的大概率就是能赚钱。
关注流量能短期赚钱，关注供应链大概率短期不赚钱但长期赚钱。最后新消费回归是不是好生意，赚不赚钱；投资回归到能不能帮上忙，能不能给Founder提供情绪价值。
回到今天的Crypto也是一样，上下游之前的共识是需要TVL，需要DAU，需要交互。所以发现一切都围绕着关键指标，为了要势能，项目们用撸毛工作室、TVL大户给自己加了杠杆。
而Crypto的高估值天生很脆弱（高FDV低流通的打法是双刃剑），而由于都给社区留出了份额，所以实际上团队的份额占比是不高的。今天加密行业的高估值本身可能就要除以一个系数，本身就不能用和传统互联网的标准（就像当年科技行业和实业天生PE不同）。
但毋庸置疑的是，Crypto是不如传统互联网和实业估值扎实的。所以如果你对10亿美金市值有执着，那确实加密行业更容易达到。
那尽头是什么呢，传统互联网的尽头是网络效应起来之后有正向现金流，Crypto目前的尽头目前是上所卖币或者上所轧空，拿到收益后来琢磨如何build或躺平。
如今的撸毛工作室，提供TVL的大户，各个雕琢繁荣数据的组成者，应了那句话，此时此刻恰如彼时彼刻。
但生态参与方到底需不需要呢？ 显然是需要的，Crypto天生的优势是团结早期的stakeholders，咱今天叫社区也行叫xx也行，本质都是一个利益和共识的团体。只有早期种子用户上找一些相对价值观接近，确实长期关注这个领域的stakeholder，这个社区才能长期。
说到这里你肯定疑惑：那不就是买了不卖的xx吗？
不瞒你说我还真想过这个问题。
我认为他们的区别是繁荣数据制造者是流水线，这个项目做完下一个，发了筹码之后就Dump，且完全不会再接回来，对项目本身也没兴趣，有兴趣的只是这个项目的TGE前的卖相。
而社区的stakeholder同样你可以参与完下一个，发了筹码之后有的拿住有的就dump，但核心区别是你真觉着这方向上有戏，在持续关注这个领域，在找这个领域的持有机会。买卖是市场行为。跌的足够低的情况下才有买盘。
这个对供需方的要求都很高，对项目来说得让holder相信你还愿意创新；愿意bd；愿意做业务；利益还跟我在一起。对Holder来说得确实关注这个领域，这个赛道，可以在这个赛道里选潜力的、低估的，而不是认为整个赛道价值归零。（这个部分是Crypto缺乏的，也没有太多机构支撑足够的买盘）
在这种互相信任收到极大破坏下，供方选择倾销是最优策略，需方选择倾销也是最优策略。囚徒困境，看谁跑得快。只有重新构建这样的信任，在下个周期里面大家才能PVE。
所以，我们琢磨做这个事情本质也是想试试，如果把生态参与者都拉拢的跟项目更近一些，让参与的人都进入生态里面，我们来做个桥梁，没有xx Maxi，允许大家讨论竞品（讨论竞品的情况下，说明还是在看同一个领域）看看会不会缓解信任崩塌的情况，换句话说，我们一起来筛选一下供需方。
市场这么熊我也不知道会有几个人参与。大概率做完第一期，没什么话题聊的话也可能解散社区。
为了筛选生态参与者，我做了有门槛，不建议非Pro用户参与；为了避免纯割群友，我们提供了等价的工具，附在评论区视频可以查看。
关键信息：
有门槛：1 ETH
主题：Megaeth
我们的工作：MegaMafia（官方孵化器）中的项目我会都看一遍，给出非常鲜明的看法，这部分我们也会脱敏公开，对社区部分不脱敏。所有社区收费都直接投入生态的各类角度。
人数限制：上限100人（不认识的我都像打个电话聊聊，人多整不了）
你的权益：
1）直接获得我们研究Megaeth的结果
2）获得一个我们自用查询Twitter热度的工具，准确度可以做策略水平。使用时间保证到Megaeth TGE之后（预览附在评论区）
3）不会明牌获得任何回报承诺
举例说明可能存在的alpha用例： 1）Noise是Megaeth的生态项目，目前利用Kaito的Mindshare在做二元期权的交易，有点像Polymarket的结构。目前我们用的这个工具可能能辅助该项目的交易。
2）社区更容易获得Megaeth NFT的白名单
如果看完了这么长，你竟然还没走，那麻烦你填一下list，有缘我会联系你：
docs.google.com/forms/d/1hkn…
5
19
1,668
推荐对棋牌游戏感兴趣的朋友别错过，绝对精彩
join showdown tournaments, they won’t let you down
We’ve been hard at work improving Showdown, and now we’re ready to unveil Showdown 2.0! We’re hosting two back-to-back tournaments in Denver on March 1st and 2nd.
Capacity is limited to 32 people, with some great prizes (more info on that soon). Participants will be automatically invited to our closed alpha. Register on Lemonade (links in the next tweet).
Let’s have some fun together, see you in Denver!
5
1
7
1,297
Cloudflare is truly respectable
I did 5 years in Product there, it's a lot of things from top to bottom:
- CEO drives product & engineering hard
- Product and engineering leaders were focused on delivery and quality; very little politics
- Org is split into mature products and "emerging". Emerging get to take bets some of the big new product lines have come from there
- At least when I left, there was no-one with the role "tester". Engineers, product, solution engineers and your early test customers were the testers.
- Cloudflare is increasingly built on Cloudflare and if you've built on Cloudflare it's.... amazing, and getting better. They'll get faster, not slower.
- It attracts and retains great engineers because it's an interesting place to work. Innovation combined with truly internet scale.
- Sales has a seat at the table, but Product managers had a lot of leeway to take bets on features based on talking to customers. It means you get products slightly "ahead of the curve".
3
14
1,353
实话。我没想到chain.fm 如此成熟，居然是个人开发者作品。
ui和ux都非常好，现在tg通知的延迟要比网页低很多，后面期待更多优化。
7
1,471
!!!!!!!!!!!!!!!!
today is a special day
@yangl1996 arrives in megazu & chiang mai
who wants to hang with our chief bunny officer and learn about megaeth?
5
470
对币圈项目打破现有学术体系没有任何信心
不才区区在下也曾经是个科研工作者，英文叫scientist，a.k.a. 科学家。
sci-hub 是科研领域的海岛湾，天然有几分侠义精神，又天然为主流正统所不容。
随着声名日显，sci-hub的内容库里已经有了一些合法的成份，比如论文作者主动投递pre-print 版本到sci-hub ，这样不存在版权纠纷。
可这不是可持续发展的长久之计。论文投递到sci-hub没有影响因子，没有影响因子就难拿到funding，拿不到funding没钱做实验就几乎没机会产出新的论文。
论文太昂贵，解决办法是谋求降低论文的成本乃至价格。成立一个地下联盟有效，但是……
做过科学家的都知道，申经费写本子，占据科研工作者80%的时间与精力。
deSci给科研项目筹款，根本不是出路。币圈逐利的本性，只会比现有的funding resource更加短视。多少基础学科嗷嗷待哺的实验室，写出来的申请书，连业内评审都打动不了，真的能打动币圈？
以学术期刊上发表的论文为载体的学术产出是一个巨大的有向无环图。数据长期保存，不可（应）篡改，有复杂庞大的引用关系。 不知道何时能看到它变成一个DAG，出现链上期刊，科研进度和成果自由分享……
说的有点多，没什么用
2
5
1,320
我再推荐一个最近看到的：github.com/AdepojuJeremy/Cud…
2
1
5
1,500
"Both are fair opinions but a lot of the tl screaming (“1 eth mint is so expensive”) is not even able to digest the info."
The MegaETH NFT mint is an IQ test and the results are shocking if you look at the comments under their announcement.
Ultimately it’s a raise that lets you buy tokens at ~ $550M fdv with an option lower the valuation you invested at as they clearly stated the minimum token allocation is 5% for all NFTs.
You can either say this is a good deal and I want in or nope I think this is overvalued.
Both are fair opinions but a lot of the tl screaming (“1 eth mint is so expensive”) is not even able to digest the info.
1
3
1,268
Trust Mega choices
In the words of every lazy English lit student who didn’t do the reading:
"Honestly, the setting is the real protagonist of this story."
Having personally selected the venues for each of our events this year, I can’t emphasize enough how important the space is:
Devcon (Bangkok): The House on Sathorn – Former Soviet Embassy, now a national archaeological site
Token2049 (Singapore): Chijmes Hall – Formerly the Convent of the Holy Infant Jesus Chapel, now a national monument
SBC (New York): The Backroom – One of only two surviving Prohibition speakeasies
EthCC (Brussels): Royal Library of Belgium – Home to the national archive and originating to the Library of the Dukes of Burgundy
While guests and content are key, the venue sets the stage—and in many ways, drives the experience.
4
533
Solid review
当我们聊Megaeth和Monad的争论，我们聊的是什么
What Are We Really Talking About When We Discuss the Debate Between Megaeth and Monad?
@bankless 视频里面40分40秒左右开始，有了后来无数媒体讨论的full node之争：
piped.video/watch?v=1qZbLyHP…
Around the 40:40 mark in the @bankless video, the now widely-discussed debate on full nodes started:
piped.video/watch?v=1qZbLyHP…
先给结论，后论项目。
-full node分歧: 实际上 @megaeth 的 @yangl1996 所说的fullnode，潜台词是表达自己做了node功能上的拆分，也就是他们说的node specialization。 @keoneHD 不知道这个定义，实际上应该是在场的几个人，都没看过 @yilongl_megaeth 的研究。所以很多上下文是没对上的。
-keone质疑的问题:其实并不是质疑full node，而是centralized sequencer。这个质疑只能说质疑的一半对，因为他并不知道用了prover去监督执行，大白话说就是用prover来实现去中心化。但是他质疑的根源也没问题，因为one active sequencer确实是要面对centralized sequencer的质疑，加上显然keone没看过megaeth的doc，以他的视角提出这个质疑。
那么，当讨论Megaeth时说Vitalik投了之外，我们究竟认真看了多少？
First, the conclusion; later, the project details.
-Full Node Dispute:
What @yangl1996 from @megaeth meant by "full node" actually implied that they have decomposed node functionalities—what they call "node specialization." @keoneHD was not aware of this definition. In fact, it's likely that none of the participants had read @yilongl_megaeth's research, leading to mismatched contexts.
-Keone's Doubt:
Keone's question wasn't really about full nodes, but rather about the centralized sequencer. His skepticism is only partially correct because he didn't know that they were using a prover to supervise execution—in plain terms, using a prover to achieve decentralization. However, his doubts are understandable because having just one active sequencer indeed raises questions about centralization. It is evident that Keone hadn't read the Megaeth documentation, so his skepticism came from his perspective.
So, when discussing Megaeth, beyond mentioning that Vitalik invested, how much have we seriously examined?
有关 @megaeth ，我们该知道的一切
Everything You Need to Know About @megaeth
最早是 @BMANLead 约在香港碰了一面，他推荐了 @megaeth 。（除了讨论他可爱的女儿和megaeth时候他自豪的侃侃而谈，其余时候是我们两个i人在尬聊，有趣的经历）
他们是最早找到 @yilongl_megaeth 的机构，我很好奇，是谁找到的yilong？可能是 @cyodyssey ？BMAN当时兴奋的说了real time blockchain这个词，我下意识得当作了vc造得一个概念，于是延迟了几天看doc（这是我肤浅了）。
几天后，刷到 @_Digital_J_ 从哈佛phd休学加入megaeth的推特，我天生对退学创业的人充满了敬畏，如果我带着现在的认知重返18岁，一定直接退学，多几次容错空间。当然，也可能对曾经有同款发型有好感。
在他那，看了yilong这个doc之后，我们对megaeth产生了很大的兴趣。
doc的链接：
docs.google.com/presentation…
It all started when @BMANLead met up in Hong Kong and recommended @megaeth. (Besides talking proudly about his adorable daughter and Megaeth, the rest of our chat was mostly awkward small talk—an interesting experience.)
They were the earliest organization to discover @yilongl_megaeth. I was curious—who exactly found Yilong? Maybe it was @cyodyssey? BMAN enthusiastically mentioned "real-time blockchain," which I initially dismissed as a VC buzzword and delayed reading the documentation by a few days (that was a shallow judgment).
A few days later, I came across @Digital_J’s tweet announcing his leave from Harvard’s PhD program to join Megaeth. I have a natural admiration for those who drop out to start a venture; if I could go back to being 18 with my current understanding, I would drop out immediately to have more room for trial and error. Also, perhaps there was a sense of affinity due to having the same hairstyle back then.
From his tweet, I found the document by Yilong, and that piqued our interest in Megaeth.
Document Link:
docs.google.com/presentation…
@yilongl_megaeth 做了一些非常solid的实验，论证了L1几乎不可能达到100k TPS的结论。当RAM扩展到一定程度，边际效应是极其明显的。
大白话翻译就是L1达到100k tps几乎需要无穷大的RAM。论证过程可以看上述的文档。那么，怎么才可能突破tps的瓶颈呢。 这也是这场争论的真正原点，megaeth选择了直接改结构，而monad选择直接优化各个环节。各有利弊。
@yilongl_megaeth conducted some solid experiments that demonstrated it is practically impossible for L1 to reach 100k TPS. When RAM is expanded to a certain extent, the marginal effects become extremely pronounced.
To put it simply, reaching 100k TPS on L1 would require an infinite amount of RAM. The reasoning can be found in the document above. So, how can the TPS bottleneck be broken? This is the core of the debate. Megaeth chose to change the structure directly, while Monad opted to optimize each stage. Both have their pros and cons.
@megaeth 的解决方案：拆分。
将一个节点的功能拆分成三个角色：
1）sequencer（排序）：决定顺序交易的排序（only one active sequencer）
2）prover（验证）：决定交易是不是合法
3）node（执行）：只负责按照顺序执行
这是个非常学术逻辑的延展，比如prover是能从vitalik的论文里找到原型的，我们看论文的小伙伴直接发给了我这篇文章。
论文原文：
cs.ucl.ac.uk/staff/m.albassa…
Solution from @megaeth: Decomposition.
Breaking down the functionalities of a node into three roles:
Sequencer: Decides the order of transactions (only one active sequencer)
Prover: Validates whether transactions are legitimate
Node: Executes transactions in the given order
This approach extends academic logic—such as the role of the prover, which can be traced back to a paper by Vitalik. A friend who read the paper shared it with me.
Original Paper:
cs.ucl.ac.uk/staff/m.albassa…
那么，这个结构还真不能算是无源之水，但可能算非传统门派武功（比如乾坤大挪移）。
优劣都十分明显：
-优点：是三个方向都可以精进。通过分拆功能，实现了速度的提升从‘线性’变成了‘指数’，100k tps的终局是各个维度的相乘。2+2+2显然不如2*2*2效率高
-劣势：即使有prover，one active sequencer也确实容易被挑战。
So, this structure isn't without precedent, but it could be considered unorthodox (like a unique martial art technique).
Advantages and Disadvantages:
Pros: All three aspects can be improved. By decomposing functions, the speed improvement changes from linear to exponential. The endpoint of 100k TPS is a multiplication across dimensions. Clearly, 2+2+2 is less efficient than 222.
Cons: Even with a prover, one active sequencer is still vulnerable to challenge.
monad 解决方案：优化
优化数据库 MonadDB
优化效率：Parallel Execution
优化执行：Asynchronous execution
优化算法：Monad BFT
优劣点也很明显：
-优点：保留了去中心化（虽然节点也要物理上靠近）
-劣势：无法回答yilong的结论，L1如何突破瓶颈
Solution from monad: Optimization.
Optimizing Database: MonadDB
Enhancing Efficiency: Parallel Execution
Improving Execution: Asynchronous Execution
Algorithm Optimization: Monad BFT
Advantages and Disadvantages:
Pros: Retains decentralization (although nodes also need to be physically close).
Cons: Cannot address Yilong's conclusion—how to break the L1 bottleneck.
Firedancer的很多影子能看到，目前从keone的推特表述里面看来是围绕优化状态问题，原本我以为是优化读写，后来发现是既然状态改变比较麻烦，那我尽量也少读写。以及通过节点物理范围集中等方式来优化，总体是传统武术，无限优化。总体比较去中心化原教旨，有道德优势。看起来像一个大号的sonic。
然而，monad也有自己要回答的终极问题，就是tps如何突破瓶颈？当然，或许kone认为不需要回答，只要比对手强就好，毕竟链的成功显然不单是技术问题。目前公布的tps10k看起来水分应当是不高，较为真实。
You can see the influence of Firedancer here. From Keone's tweets, it seems their focus is on optimizing state management. Initially, I thought it was about optimizing read/write operations, but it turns out the goal is to minimize read/write operations as much as possible due to the complexity of state changes, and optimizing by concentrating nodes within a certain physical range. It's a more traditional approach of endless optimization. It’s quite decentralized at its core, offering a moral advantage—like a larger version of Sonic.
However, Monad also has its own ultimate question to answer: how to break the TPS bottleneck? Of course, Keone might think that as long as it’s better than the competitors, there's no need to answer it. After all, the success of a blockchain is clearly not just a technical matter. The announced 10k TPS seems reasonably authentic.
除了技术门派的不同，隐藏在背后需要讨论的是视角问题。Monad更商业视角，Megaeth更技术视角。@VitalikButerin 的加入讨论更加深了一层，比如防欺诈窗口这个问题，其实从商业上来说无论多久10钟还是7天都是不合理的。（我看到 @paradigm 的人在评论区表达了同样的观点）
@keoneHD 看的出来是比较丰富的实际视角下的考虑，甚至在讲创新的时候讲了社区运营上的创新。这个很多人忽略。
yang 、 yilong 都属于更技术视角，严谨论证，谨慎创新。（多说一句， eigenlayer 的DA就是yang写的）
Beyond the technical differences, there’s a hidden issue of perspective. Monad takes a more commercial perspective, while Megaeth has a more technical focus. @VitalikButerin's involvement adds another layer, such as the fraud-proof window discussion, which is commercially unreasonable whether it's 10 minutes or 7 days (I noticed someone from @paradigm expressed a similar view in the comments).
@keoneHD clearly considers things from a practical perspective, even mentioning community management innovations when discussing innovation—something many overlook.
yangl and yilong both approach things from a more technical perspective, favoring rigorous validation and cautious innovation. (As a side note, @eigencloud’s DA was written by Yang.)
希望 @hotpot_dao 的加入能够补齐商业上的视角，毕竟这两年hbs mba已经不怎么招Crypto背景的人了，华人就更少了。去年也只有一个 carv ，中途还退学了。
目前看到megaeth在社区和商业上发力较少，不知道是不是因为交割问题。但是从pitch融资的思考上，shuyao @hotpot_dao 显然是成熟的，无论是大的节点商 @FigmentCapital @heliuslabs 还是老东家 @Consensys @eigencloud ，都是团结了该团结的人。商业能力的本质是团结一切可以让自己强的力量，战略上显然是没有偷懒的。
We hope that @hotpot_dao can fill in the commercial perspective gap. After all, in the past two years, Harvard Business School's MBA program has not been recruiting many people with crypto backgrounds, especially Chinese. Last year, there was only one from carv, who also dropped out midway.
Currently, it seems Megaeth is not focusing much on community and business, possibly due to delivery issues. However, from the perspective of pitching and fundraising, Shuyao from @hotpot_dao is clearly very experienced. They have rallied the right people, whether from major node operators like @FigmentCapital and @heliuslabs or from former employers like @Consensys and @eigencloud. The essence of commercial ability is to unite all forces that can strengthen oneself, and strategically, there has been no slacking.
在从日本回来的路上写了这个推特，我们也在最近持续讨论高性能EVM链的过程里思考，未来什么环节是有产业优势的，非常感谢 monad 和 megaeth_labs 给我们的启发，如果有上游的云服务商碰巧看到了这条内容，非常希望能够交流验证一下我们的猜想。
I wrote this tweet on my way back from Japan. In our ongoing discussions about high-performance EVM chains, we are contemplating which areas have industrial advantages in the future. We are very grateful to monad and megaeth_labs for the inspiration. If any upstream cloud service providers happen to see this content, we would love to exchange ideas and validate our assumptions.
好产品不论皮肤，创新会带来增长。
世界是简单的道理和常识组成的，祝这两个产品好，由衷的。
A good product is not defined by appearances; innovation brings growth. The world is built on simple truths and common sense. Wishing the best for these two products, sincerely.
1
4
776
yes. We have optimizations specifically for full nodes to lower their hw requirements and thus increasing decentralization level of the network. there are surprisingly many skippable work when reexecuting txns on full nodes compared to the sequencer.
We'll write more technical details on our blog.
4
107
Then just a quick follow-up, what is your streaming latency and frequency?
I think such comparisons are useful to show end users a UX they can expect. Doing high-frequency streaming actually has many caveats to conquer and we are proud of solving them.
4
348
welcome!!!! (Although you were exposed to the team much earlier than me actually haha
1/ I'm excited to share that I’ve joined MegaETH as Head of Eco
Jumping to the operator side wasn’t on my bingo card at the start of the year, but after spending time with @hotpot_dao, @NamikMuduroglu, and the @megamafia builders, it was a no-brainer
So why Mega and why now?
1
4
624
great compilation
Nice writeup! Looks like some of the tech comes from other chains like @Aptos . Providing additional references below for those who might be interested.
MonadBFT builds on Jolteon, which is pioneered by Aptos, and used by Flow, Gravity, etc. arxiv.org/abs/2106.10362
Asynchronous Execution / Block stages is essentially part of the pipelined architecture of Diem and its successors like Aptos / Sui. e.g. see section 7 of the Aptos whitepaper: aptosfoundation.org/whitepap…
Raptorcast looks like the Coded Broadcast work by @yangl1996 from @megaeth
dl.acm.org/doi/pdf/10.1145/3…
Optimistic parallel execution is probably inspired by BlockSTM, a parallel engine pioneered by Aptos, and used by Polygon, Sei, Starknet, etc.
arxiv.org/abs/2203.06871
Anyway, it's always good to see well-designed systems gaining industry adoption.
1
3
769