Loading…
Running Low-Latency Analytical Workloads with GPU-Accelerated Presto on NVIDIA GB200 NVL72
NVIDIA Developer BlogTanya Lenz
Summary
GPU-accelerated Presto integrates NVIDIA cuDF algorithms and NVLink to deliver low-latency analytical query execution on massive datasets. Benchmarks derived from TPC-H across scale factors 1K and 3K reveal that a single NVIDIA DGX B200 system with eight GPUs outperforms 8- to 10-node Intel Xeon clusters by up to 8.2x. Scaling out across multinode NVIDIA GB200 NVL72 systems paired with IBM Storage Scale achieved additional performance gains. Enabling NVIDIA GPUDirect Storage, tuning I/O parameters, adjusting thread counts, and applying query rebatching delivered a cumulative 64% reduction in query runtimes. Topology-aware GPUDirect Storage cold reads proved roughly twice as fast as standard POSIX reads by avoiding CPU memory staging and NUMA boundary crossing penalties.
Context
Analytical query workloads on large datasets often suffer from execution and I/O latency bottlenecks when relying on conventional multi-node CPU-based database engines and standard POSIX storage reads.
Approach / What changed
Evaluate GPU-accelerated Presto using cuDF on single-node NVIDIA DGX B200 and multinode NVIDIA GB200 NVL72 clusters with IBM Storage Scale, applying GPUDirect Storage, 16 MiB I/O tasks, expanded I/O threads, and exchange rebatching.
Takeaways
- A single DGX B200 running eight GPUs delivered 8.2x faster query runtimes at scale factor 1K than an eight-node Intel Xeon 6642Y cluster.
- NVIDIA GPUDirect Storage cold reads achieved roughly 2x faster runtimes than POSIX reads on GB200 NVL72 by bypassing host memory bounce buffers and avoiding NUMA boundary crossing.
- I/O optimizations, 16 I/O threads, rebatching, and rewriting query Q11 to an INSERT INTO statement yielded an overall 64% query runtime reduction across 32 GPU workers on GB200 NVL72.
Related reading
Reducing High-Bandwidth Memory Bottlenecks in JAX-Based LLM Training with Host Offloading
Large language model training workloads frequently face GPU high-bandwidth memory (HBM) capacity constraints before compute limits are reached. To alleviate this memory pressure in JAX training workloads, host offloading transfers selected forward-pass activations to pinned host memory over high-speed NVLink-C2C interconnects and reloads them during the backward pass instead of recomputing them. Benchmarks executed on 128 GPUs within NVIDIA GB200 NVL72 systems using MaxText demonstrated significant improvements. For DeepSeek-V3 671B, host offloading combined with the Latency Hiding Scheduler and pipelined offloading reached 908.2 TFLOPs/s/device, surpassing activation rematerialization throughput by 57% and unlocking larger batch configurations. Similarly, QKV activation offloading on Llama 3.1 405B achieved a 2.9% throughput increase when transfer latency was hidden behind computation.
Tanya Lenz