Hardware●●●●Gem
We built Talos – a full CNN inference engine running on silicon
CNN inference fully hardcoded as silicon logic, not software optimized for hardware.
WizardryZero to OneBold Bet
luthiraabeykoon
103mo ago
Back to browse
We built Talos – a full CNN inference engine running on silicon
by luthiraabeykoon·Feb 23, 2026·1 point·0 comments
AI Analysis
●●●●GemWizardryZero to OneBig Brain
Strips away PyTorch flexibility entirely; full CNN inference as deterministic hardware logic in SystemVerilog.
Strengths
- •Radical design philosophy: inverts typical hardware accelerator approach by eliminating runtime/scheduler entirely, achieving true cycle-accurate determinism.
- •Technical depth visible in writeup—honest breakdown of hardware debugging constraints (nanosecond timing, waveform analysis) demonstrates genuine craft.
- •Solves a real problem: production inference paying hidden PyTorch overhead (dynamic graphs, autograd scheduler) even for frozen models.
Weaknesses
- •No quantitative benchmarks against GPU inference or existing FPGA accelerators; latency, throughput, power claims unverified.
- •Scope unclear: only CNN architectures, unclear which layer types supported or how memory bandwidth constraints affect real workloads.
Category
Target Audience
ML engineers, inference specialists, FPGA practitioners
Similar To
NVIDIA TensorRT · Google TPU · Xilinx Vitis HLS
Similar Projects
Hardware●●Solid
ML accelerator on a RISC-V FPGA SoC – zero-cycle matmul, boots Linux
Zero-cycle matrix multiplication in combinatorial logic on Lattice ECP5 is genuinely wild.
WizardryBig BrainNiche Gem
dstrbad
403mo ago
Hardware●●Solid
Low-rank approximation for 3x3 FPGA convolutions (33% less DSP usage)
Clever ML+hardware co-design, but a blog post without open-source code, benchmarks, or deployment examples.
WizardryNiche Gem
el_dockerr
114mo ago
Infrastructure●●●Banger
Open-source logic synthesis – formal logic to FPGA
Open-source logic synthesis running on FPGAs when Yosys dominates the space.
Zero to OneWizardryBold Bet
major4x
102mo ago
Data●●Solid
Benchmarking Apple Silicon unified mem for GPU-accelerated SQL analysis
The repo does one practical thing well: quantify the real-world impact of Apple Silicon's unified memory on analytics by running six TPC-H queries plus a GPU-favorable QX and shipping the raw charts and code. It's specific and empirical — you get MLX vs NumPy vs DuckDB numbers and PNGs, not just hand-wavy claims — but it's narrowly scoped to M4 hardware and small-ish scales, so its conclusions are useful for experimentation rather than sweeping generalization.
WizardryNiche Gem
sadopc
314mo ago
AI/ML●●●Banger
RunAnwhere – Faster AI Inference on Apple Silicon
Custom Metal shaders beat llama.cpp and MLX—1.67x faster on M4 Max.
WizardrySlickZero to One
sanchitmonga22
2401533mo ago