WITHOUT A RECORDING
Find a way back.
- A test fails.
- No waveform was saved.
- Select the signals.
- Set up waveform dumping.
- Run the simulation again.
- Wait for it to finish.
- Check that the failure recurs.
- Start debugging.
TOOL 02 / LOSSLESS WAVEFORM RECORDING
WaveZip
Every run.
Ready when you need it.
Retain the failing simulation’s waveforms without the usual recording and storage burden.
WAVEZIP / .WZIP
500×
smaller waveform files.
Same signal. Every detail.*
LOSSLESS BY DESIGNRECORD / KEEP / DEBUG
*Up to 500× compared with FST in the published tests. Compression varies by design.
LOSSLESS RECORDING
SYSTEMVERILOG
FULLY ON PREMISES
DESIGN-SPECIFIC FORMAT
THE SECOND RELEASE / AUGUST 2026
Keep the run
that matters.
Recording adds time and consumes storage. When teams disable it, a failure can mean another simulation just to obtain a waveform.
Watch the launch on YouTube ↗WAVEZIP / LAUNCH FILMLOSSLESS / ON PREMISES
KEEP THE ORIGINAL RUN
Skip the rerun
for a waveform.
Recover the recorded failure directly. Rerunning to verify a later fix is a separate step.
WITH WAVEZIP RECORDING
Go straight to the run.
- Simulate with recording enabled.
- Identify the failed test.
- Decompress its waveform.
- Start debugging.
The captured failure—not another attempt to reproduce it.
PUBLISHED BENCHMARKS / VERILATOR
Less overhead.
A lot less disk.
Four open-source designs.
CoreMark workloads.
500×smaller files, up to
<10%recording overhead in these tests
Lossless.checked against reference dumps
Total simulation time
Full waveform dumpWaveZip
VeeR EH1CoreMark ×20
CVA6CoreMark ×3
IbexCoreMark ×300
OpenTitanCoreMark ×1
Simulation plus recording. One linear scale; workloads differ by design.
| Design | Without recording | WaveZip overhead | FST ÷ WaveZip size |
|---|---|---|---|
| VeeR EH1 | 62.029s | +3.3% | 544× |
| CVA6 | 90.908s | +5.9% | 74× |
| Ibex | 185.740s | +8.6% | 398× |
| OpenTitan | 272.865s | +3.8% | 25× |
Reported by Hardware Intelligence; results depend on design and workload. Benchmarks and methodology ↗
STORAGE / SAME SIGNALS
From gigabytes
to megabytes.
VeeR EH1, CoreMark ×20: the same simulation, fully recorded.
2,947,131,944 bytes
5,413,074 bytes
Approximately 544× smaller on this run. The comparison is against FST, not raw VCD.
ONE RECORDING EVERY NIGHT
A year fits inside
one old dump.
365 identical captures at this measured size total 1.98 GB.
TIME TO OPEN THE RECORDING
Unzip it.
Get back to the bug.
Measured decompression times for the entire recording, compared with rerunning with waveform dumping enabled.
| Design / workload | Rerun + dump | Decompress |
|---|---|---|
| VeeR EH1CoreMark ×20 | 493.45s | 38.09s |
| CVA6CoreMark ×3 | 243.44s | 34.02s |
| IbexCoreMark ×300 | 897.69s | 97.93s |
| OpenTitanCoreMark ×1 | 1,247.76s | 115.39s |
Separate rerun measurements from the runtime chart. Source: Hardware Intelligence’s published results ↗
THE DETAILS
Small files.
Nothing left out.
How is losslessness checked?
Recovered waveforms are compared with a reference dump at each signal and clock edge.
Why is the compression design-specific?
The encoder and format are generated for each Verilog design, rather than applying a general-purpose compressor.
Which simulators can it work with?
The compression is written in pure SystemVerilog. The team states compatibility with compliant simulators; the published measurements use Verilator.
How do we integrate it?
It runs on premises. Hardware Intelligence is onboarding design partners and helps integrate recording into their simulation flow.
Now, find the bug.
Bring waveform investigation into your terminal with Wave.
LET’S PUT IT TO WORK
Your next chip.
Our kind of challenge.
Bring a real design. Let’s see what we can do.
Book a demo
On the list