Realm

Documentation / Benchmarks

Performance Benchmarks

Real-world performance metrics from our GPU-native blockchain. All benchmarks run on high-end gaming hardware unless otherwise noted.

Cryptography

OperationCPUGPUSpeedup
Dilithium-3 Verify47,000/s1,260,000/s27x
Dilithium-3 Sign45,000/s1,260,000/s28x
BLAKE3 Hash10,000,000/s530,000,000/s53x

Matching Engine

MetricCPU (SIMD)GPU (CUDA)Speedup
Orders/sec320,00010,000,000+31x
Matches/sec230,0008,000,000+35x
Latency (batch)0.3ms20ms*

* GPU latency dominated by kernel launch + H2D/D2H transfer overhead. With GPU Direct RDMA (ConnectX-7 NIC), transfers are eliminated.

GPU matching requires batching for efficiency — designed for high-frequency trading workloads.

Network Performance

MetricValueNotes
End-to-End TPS (4 nodes)25,000full security*
Peak TPS (single node)45,000+no consensus
Single Client TPS8,000one process
Regional Finality25mssame region
Global Settlement500mscross-region
Avg Latency (p50)16msorder submission
Tail Latency (p99)27msorder submission

* Full security — All hardening enabled: Dilithium-3 verification, P2P broadcast, BFT consensus, state root pre-execution. No shortcuts.

Verified

Real-World Benchmark

End-to-end performance measured with the TypeScript SDK calling the gRPC API. No synthetic shortcuts — real wallet signing, real network calls, real BFT consensus.

MetricValueNotes
Total Orders625,00025 second test
Aggregate TPS25,000full security
Error Rate0.00%all orders accepted
Client Processes1TypeScript SDK
Wallets1,500concurrent
Batch Size500BatchOrders API
Validators4BFT consensus, release build

Test methodology: Full security hardening enabled — Dilithium-3 signature verification, P2P transaction broadcast, 4-validator BFT consensus with 2f+1 threshold, state root pre-execution. 1,500 concurrent wallets submitting 500-order batches via BatchOrders gRPC API.25,000 TPS with 0% errors represents production-ready throughput.

Theoretical

Raw Blockchain Performance

Maximum throughput with various security layers disabled. These numbers represent the raw execution capability of the blockchain engine.

ConfigurationTPSNotes
Execution Only2,000,000+no signing, no consensus
With Signature Pre-compute1,200,000batch verify GPU
Single Validator450,000no P2P overhead
Local Consensus (4 nodes)300,000same-machine cluster

Note: Raw performance numbers are theoretical maximums useful for capacity planning. Production deployments use full security (25K TPS) which includes signature verification, P2P broadcast, and BFT consensus — the numbers that actually matter for real users.

Cross-Region Settlement

OperationThroughputNotes
Proof Verification10,431/s67μs each
Merkle Verification4,200,000/s0.24μs each
Settlement Latency~500mscross-continent

vs. Other Chains

FeatureRealmHyperliquiddYdXSolana
Finality25ms200ms1s400ms
TPS (verified)*25K100K10K4K
Security ModelFull BFTSequencerCosmosPoS
ExecutionGPUCPUCPUCPU
Post-Quantum

* Realm TPS measured with full security (BFT, Dilithium-3, P2P broadcast). 25K is production-ready; actual distributed network may vary.

Test Hardware

Benchmark System

  • GPUHigh-end NVIDIA GPU
  • CPUIntel Core Ultra 9 275HX
  • RAM128GB DDR5
  • Storage2TB NVMe Gen5

Minimum Validator

  • GPURTX 3050 (4GB)
  • VRAM Usage~200MB
  • CPU8 cores
  • RAM32GB
  • Network1Gbps

GPU memory requirements are minimal — a single validator uses only ~200MB VRAM for Dilithium-3 batch verification.