# **The Ethproofs 5090 Price Index: A Clean Baseline for Year Two of Ethproofs**
### *Four Prover Personas and a Simplified Cost Model for Ethproofs*
**December 2025 — Ethproofs**
---
## **TL;DR**
* **Four Prover Personas** emerged organically during the 2025 real-time proving sprint, each representing a distinct operational strategy and hardware profile. Ethproofs is now standardizing around these Personas and will track them consistently across onboarding, dashboards, leaderboards, and all data surfaces.
* The old **cloud-equivalent cost model** (AWS/RunPod/Vast SKUs from us-east-1) became inaccurate, unverified, and increasingly misaligned with how the ecosystem actually operates.
* In its place, Ethproofs is adopting the **Ethproofs 5090 Price Index** — a quarterly, global median price of a single RTX 5090 GPU-hour. The industry has already converged on the 5090 as the dominant proving workhorse, so this index finally aligns our cost baseline with real-world hardware.
* The new model makes cost:
→ simpler
→ more neutral
→ more reproducible
→ aligned with decentralization
→ consistent across all four Personas
* Ethproofs will also begin shifting from USD cost → **energy per proof (kWh)** in 2026, the long-term verifiable metric for real-time proving.
If you want the full reasoning, data, and methodology—including why 5090s became the natural unit of proving capacity—read on.
---
## **Introduction: How a Performance Sprint Reshaped the zk Proving Landscape**
In July 2025, the Ethereum Foundation’s zkEVM team published a landmark [blog post](https://blog.ethereum.org/2025/07/10/realtime-proving) outlining the **North-star goals for real-time proving**:
≤10s P99 latency, ≤$100k CAPEX, ≤10kW power, 128-bit security, ≤300KiB proofs, and fully open-source stacks.
This catalyzed a months-long **real-time proving sprint**, culminating at Ethproofs Day (November 22nd). zkVM teams steadily pushed latency from minutes to the single-digit-second range, and Ethproofs provided a neutral benchmarking surface along the way.
While latency dominated attention, a structural pattern quietly emerged. By early autumn, almost every prover on Ethproofs fit naturally into **one of four Prover Personas**—each representing a distinct operational philosophy, hardware footprint, and decentralization profile.
These Personas now form the backbone of how the ecosystem thinks about performance, cost, and reliability—and with them clearly established, we can now track each Persona from onboarding through every data table, chart, and leaderboard, giving the ecosystem a much cleaner, more faithful depiction of prover performance by Persona type.
---
## **Part I — The Two Prover Personas**
The Personas did not come from a spec or governance process. **They emerged organically** as teams optimized for real-time proving under different constraints and strategies.
#### **Persona 1. Single-GPU — 1:100 Prover**
The earliest prover type supported by Ethproofs. For consistent benchmarking, the ecosystem standardized on a **consumer-grade GPU baseline**, originally the RTX 4090 and now the 5090.
Single-GPU 1:100 provers submit proofs for **blocks ending in “00”**, making them the simplest, most common benchmarking Persona.
**10 such provers are active today.**
>*NOTE: One goal for 2026 is to increase the cadence of blocks proven by single-GPU provers to 1:10.*
---
#### **Persona 2. Multi-GPU — 1:1 Prover**
Teams chasing the fastest possible latency naturally converged on **multi-GPU clusters**. These are the workhorses of real-time proving.
Ethproofs onboarded the first 1:1 multi-GPU prover—**ZkCloud**—in March 2025. Their mandate: prove **every** L1 block continuously. This became the canonical “real-time” Persona.
Momentum spiked during Ethproofs Day, when several RTP teams submitted proofs for **every block** to support the [zkEVM attesting demo](https://youtu.be/BUuNcXiaH3U?si=eDdD8vEoA0EcUBfA).
Today, **four multi-GPU 1:1 provers** anchor the highest-reliability Persona.
---
These Personas now shape how Ethproofs and the ecosystem evaluate performance, cost, and decentralization. They also reveal why the old cost model—based on cloud-equivalent mappings—no longer fits.
---
## **Part II — How Ethproofs Calculated Cost Until Now (2025 Model)**
Since early 2025, Ethproofs used **cloud-equivalent costing**. Prover teams selected the cloud instance closest to their hardware, and Ethproofs used its hourly rate to estimate cost.
This worked well when:
* most provers were cloud-based,
* the ecosystem was small and homogeneous, and
* the March 25 cloud-pricing snapshot was fresh.
But as Personas emerged and hardware diversified, the model became noisy, distorted, and increasingly unrepresentative.
---
## **Part III — Why the Old Model No Longer Fits**
The shift away from cloud-equivalent costing reflects five emerging needs:
1. **Cleaner apples-to-apples comparisons** across Prover Personas
2. **Elimination of unverifiable cloud mappings**
3. **Removal of geographic distortions** introduced by only having `us-east-1` pricing available
4. **Alignment with real-world costs**, which are driven by electricity, not cloud vendors
5. **Support for thousands of home provers**, which require simpler, hardware-grounded assumptions
A fresh model was necessary.
---
## **Part IV — What Provers Actually Run (Ecosystem Data)**
Ethproofs’ hardware metadata and public datasets show clear trends:
* **~70%** of cloud-equivalent selections: 4090/5090-class
* **>80%** of real GPU capacity: 4090/5090-class
* Datacenter GPUs (A100/H100/MI300) are nearly absent
* Consumer GPUs dominate physical prover machines
* 5090 adoption surged sharply in Q3–Q4 as availability improved
**The ecosystem had already converged on consumer GPUs.**
The cost model simply needed to reflect reality.
---
## **Part V — Why 5090s Are the Natural Prover Unit**
By autumn 2025, the **RTX 5090** had become the ecosystem’s de facto proving workhorse.
#### **Why?**
* unmatched performance-per-dollar
* globally accessible retail availability
* optimal for zk workloads
* strongly decentralization-aligned
* ideal bandwidth and power envelope
* proven ability to hit real-time proving thresholds
Alternatives—datacenter GPUs, older consumer GPUs, or CPUs—either underperform or introduce centralization risks.
**Result:**
> Over 90% of all clusters tracked by Ethproofs are now 5090-based or transitioning to them.
The 5090 is the natural **unit of proving capacity**, just as “gas” is the unit of computation.
---
## **Part VI — Introducing the Ethproofs 5090 Price Index**
To replace cloud-equivalent costing, Ethproofs now publishes a quarterly **global median price for a single 5090 GPU-hour**.
#### **Quarterly Index Methodology**
1. Query Vast.ai
2. Filter to **RTX 5090**
3. Filter to **Host Reliability ≥ 99.9%**
4. Include all regions
5. Collect all hourly prices
6. Use the **median** as the 5090 Index
Typical range: **$0.37–$0.47/hr**.
This Index becomes the new universal cost baseline.
---
## **Part VII — How Cost Will Now Be Calculated**
#### **Old Method**
```
Cost_per_proof =
( Σ instance_hourly_price )
× ( proving_time_ms / 3,600,000 )
```
#### **New Method**
```
Cost_per_proof =
( num_gpus × 5090_index )
× ( proving_time_ms / 3,600,000 )
```
---
## **Part VIII — Why the 5090 Index Matters**
The Index provides:
1. **Consistent Persona-to-Persona comparisons**
2. **A decentralization-aligned cost foundation**
3. **Trustworthy, region-agnostic pricing**
4. **Minimal onboarding burden for operators**
5. **Scalability toward thousands of provers**
It is built for the next decade of real-time proving.
---
## **Conclusion**
The 2025 performance sprint reshaped the zk proving landscape.
Four **Prover Personas** emerged.
Consumer GPUs became the norm.
Decentralized real-time proving became realistic.
The **Ethproofs 5090 Index** captures this new phase. It replaces noisy cloud mappings with a clean, hardware-grounded baseline tied directly to the GPUs the ecosystem actually uses.
This is the cost framework designed for the Prover Era: decentralized, comparable, reproducible, and ready for real-time verification at global scale.
**Onward to real-time proving—accessible to everyone.**
Sign in with Wallet
Connect another wallet