The rx9070xt_16gb-ud-iq4-xs model runs fully on the AMD Radeon RX 9070 XT with 16 GB VRAM on Stratat. To verify the results yourself, open the Stratat client, select the model and GPU, then run a read and write test. Watch the live counters for reading tps and writing tps and note the power meter.
Why this quantization was chosen
The ud-iq4-xs quantization keeps the full 16 GB of VRAM usable while reducing precision enough to run large models on consumer hardware without a datacenter. It was selected because it balances speed and memory for the RX 9070 XT's architecture.
How the benchmarks were run
Run a standard inference read and write cycle on Stratat and record the requests per second. Power is measured via the system's watt meter while the GPU is under load. The 454 W figure reflects idle plus full inference power draw on this card.
| Metric | Value |
|---|---|
| vramGb | 16 |
| readingTps | 1652 |
| writingTps | 53.8 |
| powerWatts | 454 |
The core term "ud-iq4-xs" refers to a 4-bit integer quantization format optimized for AMD GPUs in Stratat's local engine.
Practical notes
The model fits comfortably within the 16 GB VRAM limit, and the Good fit rating means it can be used daily without throttling. For users who need low latency, the reading speed of 1652 tps is the main gain over uncompressed data.
Troubleshooting
If the model fails to load, double-check that the correct .saf file is installed and that the Stratat client version matches the model. Ensure the GPU is detected in the client settings before running tests.