Optimistic is dead, long live ZK

Feb 05, 2025

A bit of history

Blockchains broke the monopoly of centralized, trust-based compute and established decentralized, re-execution-based verification models. However, while trustless computation is a major breakthrough, re-execution is extremely inefficient. To solve Ethereum’s scalability challenges, optimistic and zero-knowledge-based execution models were born. Optimistic rollups were easier to build and had a first-mover advantage, but include a multi-day challenge period which makes smooth, trustless cross-chain interoperability impossible. ZK is the endgame for blockchain scalability.

The future is ZK-based

Where we are headed: all rollups will be ZK-rollups in the near future. We expect proving times for Ethereum blocks to become lower than Ethereum blocktimes within the next 12-18 months. This will allow ZK-rollups to settle on Ethereum in real-time, unlocking flexible L2<->L1 and L2<->L2 interactions. ZK will unify the fragmented L2 ecosystem under one banner: Ethereum.

Deposits and withdrawals in 1 block. Bridging between rollups in 1 block. Cross-rollup composability in 1 block. Do you feel the optimism draining from your body to be replaced by the cold, hard currency of mathematics? Optimistic constructions that don’t utilize ZK will be stuck with multi-day withdrawal times, so naturally, all of them will switch.

Is it realistic?

Absolutely. Let’s see the reasons.

Software-level improvements

Proving Ethereum blocks has been a challenge, however, developments and optimizations of zkVMs are progressing fast. A concrete example is Succinct’s recently published SP1 Turbo (v4.0.0) upgrade: on a cluster of multiple GPUs, Ethereum mainnet blocks can be proven in the ~20-30 second range end-to-end.

*Source: https://blog.succinct.xyz/sp1-turbo/

RISC Zero is also adding significant performance improvements across their zkVM stack: their new Keccak precompile can deliver a 2x boost in proving performance for ETH block proving. Another 2x performance boost is expected from the upcoming major revision of their RISC-V circuit. The team is said to be able to prove Ethereum blocks in the ~20-25 second range which seems realistic in light of the data published for SP1.

Acceleration through specialized hardware

In addition to the above, Fabric and RISC Zero announced their strategic partnership in Q4 of last year. As per the published data, Fabric’s programmable chips, the VPUs present an order of magnitude acceleration over GPUs, for instance, for the heterogenous operation eval_check, which is a compute-intensive portion of the RISC Zero zkVM workload.

*Source: https://www.fabriccryptography.com/blog/risc-zero-announcement

As part of the co-design with RISC Zero, Fabric also added specialized instructions to their VPUs for the Baby Bear prime field arithmetic, extensively used throughout the RISC Zero proving system. Due to this, the VPU can accomplish a Baby Bear multiplication in a single instruction, while the same would require 3x 64-bit (uint64) instructions and 5x 32-bit instructions in the GPU implementation.

We are convinced that a tipping point is approaching fast where proving times will not be a bottleneck anymore, and real-time proving is a reality.

What about the costs?

Previously proving costs were an issue, but this too is a thing of the past. Proving costs on ZkCloud are up to 95% cheaper than existing solutions. Turns out that incentivizing a decentralized infrastructure build-out is way better than paying Amazon and Google a margin. Who would’ve thought?

Let’s look at the figures

ZkCloud can handle both CPU and GPU workloads and is built to support specialized hardware, like Fabric’s VPUs, as they become available. ZkCloud’s CPU nodes with 96 physical CPU cores (192 vCPUs) and 768 GB RAM can be quite easily compared to similar instances available on GCP or Amazon. Below is the comparison of CPU compute costs:

GPU compute costs are somewhat harder (but not impossible) to compare realistically as the instances available at cloud providers are not optimized for ZK (while many operations are running on the GPU, some still utilize the CPU, and thus a decent amount of cores and RAM are needed for optimal performance).

The full capacity of a dual-GPU node on ZkCloud with 64 physical cores (128 vCPUs), 384 GB RAM, and 2x4090s costs $0.84 an hour. This is comparable to the performance of two g6.16xlarge instances on AWS, which costs $6.78 an hour (2 x $3,39). The result is an 88% savings on compute costs in ZkCloud.

Forget paying for idle time

The actual cost of ZK compute is just one side of the coin, the other is the wasted cost of redundancy when servers sit idle. With ZkCloud, this is also a thing of the past, as users only pay for the resources used during the actual proof computation. Universal proving infra, optimized for ZK is available here and now.

It’s already happening…

There is already movement happening. In the Optimism ecosystem, RiscZero’s Kailua and OP Succinct are providing a transition path from Optimistic to ZK. We would expect the Arbitrum ecosystem to get the memo soon as it becomes clear that real-time proving isn’t just a Justin Drake talking point or a faint twinkle in Vitalik’s eye, but approaching imminently.

The ecosystem that wins this race, becomes the backbone of a new Ethereum that can scale indefinitely. Ecosystems that don’t get on the bandwagon will be left in the dust.

Optimistic is dead, long live ZK.

—

About Us:

ZkCloud Blog

Discussion about this post