Question 1

What is PRISM by Asymmetry Computing?

Accepted Answer

PRISM is a GPU-native portfolio optimization engine built by Asymmetry Computing. It delivers 21×–125× speedup over Gurobi on institutional portfolio optimization problems. March 2026 benchmarks: 124.85× faster on Transition Rebalance at N=5,000 real US equities (1.436s vs 179.3s). All solutions are KKT-certified with less than 0.004% quality gap. PRISM supports direct indexing, tax-loss harvesting, portfolio transitions, and collateral optimization.

Question 2

How fast is PRISM compared to Gurobi?

Accepted Answer

PRISM achieves 124.85× speedup vs Gurobi 13.0.1 CPU on Transition Rebalance (N=5,000 real US equities, March 2026, timed_runs=5 p50). Across all 4 institutional scenarios, speedups range from 20.92× (Crisis Resilience) to 124.85× (Transition Rebalance). Total wall time: 1.436 seconds vs Gurobi's 179.3 seconds for the same problem.

Question 3

How does PRISM compare to cuOpt, OSQP, HiGHS, and Clarabel?

Accepted Answer

6-solver benchmark (March 2026, N=5,000 real US equities, timed_runs=5 p50): PRISM GPU passes all 4 KKT-certified scenario tests with less than 0.004% quality gap. cuOpt (GPU) fails all 4 scenarios with 9%–32% quality degradation. OSQP fails all 4 scenarios with 13%–32% quality degradation. Clarabel collapses at 7,000+ assets. HiGHS is tested on factor-QP only. Gurobi 13.0.1 CPU is the reference solver — PRISM is 124.85× faster.

Question 4

What is the solution quality of PRISM?

Accepted Answer

PRISM delivers KKT-certified solutions with less than 0.004% optimality gap vs Gurobi on all 4 institutional scenarios. 100% convergence across 40 trials at fixed random seeds. cuOpt produces 9%–32% quality degradation on the same constraint set; OSQP produces 13%–32%. Only PRISM passes the 0.01% production gate on all 4 transition scenarios tested.

Question 5

What asset scale can PRISM handle?

Accepted Answer

PRISM has been benchmarked on portfolios up to 75,257 assets using real US equity data (March 2026). At 5,000 assets it solves in 1.436s. At 75,257 assets all four institutional scenarios maintain positive speedup vs Gurobi, ranging from 1.3× to 14×. Factor-QP runs at 138.7ms median at N=5,000 (vs cuOpt 870ms, OSQP 11,311ms).

Question 6

What is direct indexing and how does PRISM support it?

Accepted Answer

Direct indexing is a strategy where investors hold individual securities rather than index funds, enabling tax-loss harvesting, ESG customization, and factor tilts at the account level. PRISM's 124.85× speedup over Gurobi enables genuine intraday personalization at institutional scale: at 1.4 seconds per 5,000-asset portfolio, PRISM can rebalance 100 accounts with full constraint satisfaction in the time a single Gurobi run initializes.

Question 7

How is PRISM deployed and integrated?

Accepted Answer

PRISM is available as a REST API deployed on dedicated cloud infrastructure or within your VPC. All optimization runs produce KKT-certified results with replayable audit logs. Hardware: NVIDIA RTX 4000 Ada, CUDA 12.x. A Python SDK is available. Deployment options include shared cloud, dedicated instance, and on-premise VPC. Enterprise and institutional licenses available via matched-workload pilot.

Question 8

What is PRISM-Q?

Accepted Answer

PRISM-Q is Asymmetry Computing's quantum-enhanced discrete refinement layer. It applies hybrid quantum-classical optimization to the discrete component of portfolio problems — lot selection, tax-lot assignment, integer constraints. PRISM-Q is built on the PRISM GPU solver foundation and extends it with quantum circuit-based refinement for combinatorial sub-problems.

Question 9

What is a matched-workload pilot with Asymmetry Computing?

Accepted Answer

A matched-workload pilot is a benchmark where PRISM is run against your current solver using your actual asset universe, constraint set, and factor model — under the same timing and quality methodology documented in the public benchmark report. The pilot includes a full benchmark pack (solver configs, timing protocol, p50 tables, KKT certificates), a technical walk-through for quant engineering teams, and NDA-protected handling of proprietary data. Results are typically delivered within 24–48 hours of kickoff.

PRISM GPU Solver — 125× Faster Than Gurobi | Portfolio Optimization Engine | Asymmetry Computing