Performance Benchmarks
Validated performance results for Essence and Wantware. All figures cited here have been confirmed by independent third parties or direct measurement. Results are workload-specific; context and methodology are included for each.
Each result is tied to a specific workload, hardware configuration, and validator. These are not synthetic benchmarks or simulations; they are measured outcomes on real hardware under documented conditions. Results on different workloads will vary; the range reflects that variation, not uncertainty about the measurements themselves.
This page covers two distinct bodies of evidence: current GPU acceleration results, independently validated by AWS and Rowan University as part of the active Chameleon pilot program and replicated on OCI and GCP, and earlier foundational CPU parallelization research from MindAptiv's internal work roughly a decade prior. They are kept separate below and should not be read as the same evidence class.
GPU acceleration results
| Result | Workload | Hardware | Validator |
|---|---|---|---|
| 20–114× acceleration range | Range across multiple workload types including compute, rendering, and data processing | AWS cloud infrastructure; Rowan University DEHub; replicated on OCI and GCP | AWS, Rowan University Digital Engineering Hub |
The 20–114× range reflects variation across workload types and hardware: workloads with high parallelism headroom produce results toward the upper end; serialized or I/O-bound workloads produce results toward the lower end. The source of the gain is Chameleon generating hardware-tuned SPIR-V instructions at runtime, not hardware upgrades.
Foundational research: CPU parallelization (historical)
| Result | Workload | Hardware | Validator |
|---|---|---|---|
| 103× speedup | High-polygon 3D rendering: 32 min (serialized) reduced to 18.8 sec via Qcode dynamic parallelism across CPU threads | 2011 Mac Pro (unmodified), CPU multithreading | MindAptiv internal measurement |
This is the most precisely documented single measurement in MindAptiv's history, and it demonstrates the same underlying dynamic-parallelism principle Chameleon uses today, but it is a CPU multithreading result from roughly a decade before the current GPU pilot program, measured internally rather than by an independent third party. It should be read as foundational research validating the core approach, not as part of the AWS/Rowan-validated GPU acceleration results above.
Energy reduction
| Result | Workload | Validator | Note |
|---|---|---|---|
| Up to 99.7% energy reduction | Compute workloads where execution efficiency eliminates idle cycles and redundant operations | AWS (confirmed) | Consistent on GCP and OCI but not independently validated on those platforms |
Energy reduction results from eliminating wasted cycles rather than from power management or throttling. When execution is driven directly from Meaning Coordinates via Qcode (without the overhead of compilers, interpreters, and runtime layers), the computational work performed per unit of energy increases substantially.
Bandwidth reduction: WarpSpeed
| Result | Workload | Route | Validator |
|---|---|---|---|
| HD video at 28 kbps | Live video transmission at high visual quality over severely constrained bandwidth via WarpSpeed | Milan to Denver | MindAptiv measurement: WarpSpeed (bandwidth reduction layer) |
WarpSpeed is the bandwidth reduction layer within Essence. The 28 kbps HD video result was achieved on the Milan–Denver route and demonstrates the practical effect of meaning-driven signal resampling: output quality exceeds what the raw input bandwidth would normally support. This result is relevant to edge deployments, mobile environments, and any workload where network constraints limit conventional streaming. See the demos page for related video enhancement demonstrations.
Video enhancement: illumin8
| Result | Workload | Method | Note |
|---|---|---|---|
| 1K output exceeds original 4K source quality | Real-time video enhancement: sharper edges, richer color detail, deeper blacks, zero pixelization | illumin8 signal processing substrate of the Essence pipeline. Not running as a build today; it returns when the milestones are complete | Shown in a recording of a 4K source downsampled to 1K: output quality measurably surpasses the original. No heavy pipelines or manual tuning required. |
illumin8 is a signal processing substrate embedded in the Essence pipeline. Every signal (audio, video, text, GUI data) passes through it. The enhancement is not upscaling or interpolation; it is mathematical reconstruction using the gradient quaternion logarithm technique covered by MindAptiv's three issued U.S. patents. The result is that a 1K signal processed by illumin8 displays measurably greater detail than the original 4K source it was derived from, including sharper edges, richer color gradients, deeper blacks, and zero pixelization artifacts, all in real time.
illumin8 supports multiple product surfaces including a Signal Pipeline, Track Journey, SyncMesh, player, editor, podcaster, and creator tools. See the demos page for the real-time video enhancement demonstration (Demo 05).
Zero-code app development: one person, one hour
| Result | Workload | Method | Distribution |
|---|---|---|---|
| Existing App Store app recreated in approximately one hour | Full app recreation: no code written; natural language dialog via Synergy translated directly into machine instructions | Synergy dialog-based intent composition → Qcode machine instructions → Aptiv ready for distribution | One-click distribution to app stores, marketplaces, or websites, no CI/CD pipeline required |
Demonstrated by Jake Kolb, Co-Founder and Chief Science Officer, this result shows the full Wantware development loop: a popular scoreboard app from the App Store was recreated from scratch using only natural language dialog through Synergy. No code was written at any stage. Synergy translated intent into Meaning Coordinates, Morpheus generated pristine machine instructions in real time, and the resulting Aptiv was immediately ready for one-click distribution to stores, marketplaces, or websites.
No CI/CD pipeline, no build system, no deployment configuration. The Aptiv is the distributable artifact. This represents a fundamental change in how software is authored and shipped, from code-based SDLC to intent-native development where a single person working in natural language can produce a production-ready application in the time it would conventionally take to scaffold a project. See the demos page for the full demonstration (Demo 01).
GPU instruction generation
| Capability | Method | Status |
|---|---|---|
| SPIR-V generation at runtime via Chameleon | GPU instructions generated directly from Meaning Coordinates, no manual shader authoring required | Available now |
Chameleon generates SPIR-V intermediate representations from Meaning Coordinates at runtime. SPIR-V is the standard cross-vendor GPU instruction format supported by Vulkan, OpenCL, and OpenGL. The output is constrained to the target hardware's actual capabilities: register limits, driver caps, and wavefront sizes are respected automatically.
Independent validators
| Validator | Role | What was confirmed |
|---|---|---|
| Amazon Web Services (AWS) | Cloud infrastructure partner: nClouds MAP Lite engagement | Acceleration range (20–114×); up to 99.7% energy reduction |
| Oracle Cloud Infrastructure (OCI) | Cloud infrastructure partner | Acceleration range (consistent with AWS results) |
| Rowan University Digital Engineering Hub | Academic research partner: Professor Antonios Kontsos, DEHub Director | Acceleration range confirmed across engineering workloads |
Patent foundation
The signal processing performance results are grounded in three issued U.S. patents covering the gradient quaternion logarithm technique that enables output signal fidelity to exceed input signal fidelity across the full electromagnetic spectrum.
| Patent | Scope | Note |
|---|---|---|
| US 10,037,592 | Digital quaternion logarithm signal processing for images and other data types | All claims granted without modification, no prior art found |
| US 10,846,821 | Gradient signal processing for video signals | Issued |
| US 11,373,272 | Multi-dimensional gradient signal processing, natural language interfaces, and multi-processor optimization | Issued |
How to evaluate these results
The numbers above are a starting point for technical due diligence, not a substitute for it. If you are evaluating Essence for a specific workload, the right approach is a scoped pilot; we structure engagements to measure results against your actual infrastructure and workload characteristics, not synthetic benchmarks.
The GPU acceleration range (20–114×) has been independently validated by AWS and Rowan University and replicated on OCI and GCP, and energy reduction (up to 99.7%) has been confirmed by AWS, across multiple workload types; this is the current, third-party-validated evidence base. The 103× CPU rendering result is separate: a foundational, internally-measured result from roughly a decade earlier that validates the same underlying principle but predates the GPU pilot program and was not independently validated. WarpSpeed's 28 kbps HD video result demonstrates meaningful performance on constrained networks. All results are workload-specific. Contact us to structure a pilot against your specific infrastructure.