Transformer inference
Check supported precision formats and measure latency on your own model.
Hopper in the SXM form factor with 80 GB HBM3 for data center AI workloads.
Choosing H100 SXM to run a language model? Compare your workload with ASUS Ascent GX10 rental: NVIDIA GB10 and 128 GB of unified CPU and GPU memory on one node.
Request one ASUS system for inference and RAG, or a pair for parallel workloads. Compare model and context memory, ARM64 support and the time needed for a trial run. GX10 unified memory and dedicated VRAM have different trade-offs.
Match ASUS GX10 to your model128 GB unified memory · NVIDIA GB10 · one system or a pair. Availability and software are agreed for your workload.
Check supported precision formats and measure latency on your own model.
Choose batch size and sequence length within 80 GB HBM3.
Compare engine settings using the same requests and concurrency.
Provide your model, software environment and rental period. We will confirm the rate, availability and connection method before the rental starts.
The GPU has 80 GB of HBM3. Include runtime memory and working buffers in the total.
No. These configurations provide 80 GB HBM3 on H100 SXM and 141 GB HBM3e on H200 SXM. If model or context memory is the bottleneck, start with that difference.
This configuration is open for rental requests. Share your workload, preferred dates and duration. Hardware availability, pricing and access dates are confirmed individually; a request does not create a confirmed booking.