Rent the NVIDIA GB10 as a dedicated DGX Spark

GB10 is NVIDIA's Grace-Blackwell superchip: a Grace ARM64 CPU and a Blackwell GPU sharing 128 GB of unified memory in one pool. It powers the NVIDIA DGX Spark, which is the product you rent.

Community benchmarks (SparkBench, llama.cpp) span roughly 10 tokens/s for 30B-class dense models up to ~90 tokens/s for 30 to 35B MoE at 4k context, with small dense models faster still. These are labelled community numbers, not ours. Architecture, quantization and serving stack dominate.

Capacity and terms must be confirmed before a booking. as a dedicated NVIDIA DGX Spark from AxForge, hosted in Málaga, Spain (eu-es-1) with full SSH access: €1.19/hour on demand, €0.95/hour by the week, €0.91/hour by the month or €0.85/hour by the year.

The 128 GB unified pool means model weights up to roughly 100 GB can load on a single GB10 without sharding across devices, leaving room for KV cache and the system. Production workloads on AxForge nodes are 7B to 35B class models.

Yes. The Grace CPU is ARM64, so software must be built for ARM64. AxForge's own serving stack runs on it in production.

GB10 has plenty of compute but roughly 273 GB/s of memory bandwidth: dense decode is bandwidth-limited, while a sparse MoE only reads its active experts per token. In the community llama.cpp sweep that is 89.3 t/s versus 10.7 t/s.