dpv1
First platformThunderbolt device
Specifications
| Parameter | Value |
|---|---|
| Logic | 134K LUTs |
| Memory | 1 GB DDR3 |
| Host link | PCIe Gen2 x2 over Thunderbolt |
Draft
Items marked TBD have not been set. They live in src/lib/invest/overview.ts; the
results come from src/lib/results.ts, shared with the home page.
dau builds FPGA accelerators sized to analytical workloads, along with open-source software that decides how much of each query the FPGA should run.
01 · Thesis
Analytical software already runs on fast CPUs. dau adds an FPGA card to the machine and splits the work between the two, so queries run faster without rewriting the software.
dau loads rows into device memory once and runs repeated filter, derive, and aggregate queries against them. A cost model picks the split for each query from measured rates. Anything the loaded hardware can't do runs on the CPU, and the planner reports it.
Device
Scans over resident rows.
Host
Sorts, text search, floating point, and everything else.
02 · What exists
| Milestone | Status |
|---|---|
| TPC-H q1, q3, q6, q19 | Bit-exact on dpv1 silicon |
| Full-day TAQ to OHLCV bars | 137.4M rows, every bar equal to Polars |
| Rebuild-and-flash loop | Workload to new bitstream to silicon, bit-exact |
| Four execution postures | CPU, offload, split, adaptive; 8 of 8 bit-exact |
| macOS host | Full DMA from a Mac over Thunderbolt |
| dpv2 bring-up | TPC-H q6 bit-exact on both boards |
| Package | Release | Scope |
|---|---|---|
| dau-core | 0.4.0 on PyPI | Operator tiles, golden models, capability contracts |
| dau-build | 0.15.2 on PyPI | Specs to bitstream, Vivado or yosys |
| dau-sim | 0.3.0 on PyPI | Cycle-accurate simulation |
| dau-driver | 0.1.0 on PyPI | XDMA host I/O on Linux and macOS |
| dau-scheduler | 0.1.0 on PyPI | Cost-based host/device split |
| dau-polars | Private, release in preparation | Polars plan capture and execution |
Thunderbolt device
| Parameter | Value |
|---|---|
| Logic | 134K LUTs |
| Memory | 1 GB DDR3 |
| Host link | PCIe Gen2 x2 over Thunderbolt |
PCIe card
| Parameter | Value |
|---|---|
| Logic | 204K LUTs |
| Memory | 2 GB onboard + 8 GB SODIMM DDR3 |
| Host link | PCIe x8 |
| Card cost | About $700 |
PCIe card for servers
| Parameter | Value |
|---|---|
| Memory | DDR4, capacity to be set |
| Host link | PCIe |
| Card to card | Direct fabric links |
03 · Evidence
*On TPC-H query 6, a standard analytics benchmark, adding a dau card to the host made the query about 2.5× faster. Every figure matches its software reference exactly.
| Benchmark | Board | Result |
|---|---|---|
| TPC-H query 6, with the table loaded on the card | dpv2 | ~2.5× faster with the card added With a six-million-row table on the card, the CPU and card together finished in 4.2 ms, against 10.4 ms for the CPU alone. The planner picks the split from measured costs. |
| A full day of NYSE TAQ trades turned into OHLCV bars | dpv1 | 137.4M trades, 2.44M bars Every bar matches the CPU Polars result. |
| Time-series workflow: as-of join, derived features, time bars, rolling statistics | dpv2 | 1,024,064 records in 11.5 ms All four steps run in one pass, and intermediate results stay on the card. |
04 · Business model
The software is Apache 2.0, published on PyPI, and runs on open FPGA hardware. Revenue comes from boards built for dau and from configurations we build and verify for a customer's workloads.
| Parameter | Value |
|---|---|
| Pricing | TBD |
05 · Market
The first workloads are time-series and market data: bars, windows, and as-of joins run again and again over the same history.
| Parameter | Value |
|---|---|
| Market sizing | TBD |
| First customers | TBD |
06 · Roadmap
07 · Risk
08 · The ask
| Parameter | Value |
|---|---|
| Amount | TBD |
| Instrument | TBD |
| Valuation | TBD |
| Item | Amount |
|---|---|
| dpv3 design and first build | TBD |
| Engineering hires | TBD |
| Design-partner deployments | TBD |
Questions and data-room requests: admin@1kbgz.com