AMBER 26 NVIDIA GPU Benchmarks

AMBER 26 Benchmarks on NVIDIA GPUs

At Exxact, we have benchmarked AMBER for quite some years now to provide GPU performance for the molecular dynamics simulation suite and to educate both our team and customers on choosing optimal hardware. We want to help customers choose the right NVIDIA GPU for their molecular dynamics workloads, from smaller models to million-atom-class systems.

Before getting into the graphs, here are the key test conditions:

• Each result is single-GPU performance. AMBER does not accelerate one simulation across multiple GPUs, but you can run multiple simulations in parallel on a multi-GPU system.
• AMBER 26 + AmberTools 26, CUDA 12.8.
• Grace results use a Grace Hopper Superchip platform.
• AMBER GPU performance is driven primarily by CUDA/GPU capability; platform differences are typically minor.

The suite covers a range of problem sizes (largest to smallest):

• STMV NPT 4fs (1,067,095 atoms)
• Cellulose NPT 2fs (408,609 atoms)
• FactorIX NPT 2fs (90,906 atoms)
• DHFR (JAC Prod.) NPT 2fs (23,558 atoms)
• Nucleosome GB 2fs (25,095 atoms)
• Myoglobin GB 2fs (2,492 atoms)

Note: We show NPT results. NVE runs are typically ~2–8% faster across GPUs. For all graphs NVE and NPT, see our AMBER supported software page, or talk to an Exxact engineer today.

Quick AMBER GPU Benchmark Takeaways

• For larger systems, NVIDIA Blackwell GPUs pull ahead of the prior generation.
  • NVIDIA GeForce RTX 5090 offers strong single-GPU throughput per dollar, but is usually best suited for single-GPU workstations due to it’s chassis scalability limitations.
  • NVIDIA RTX PRO 6000 Blackwell Max-Q is excellent for large simulations that can take advantage of the 96GB of VRAM; smaller jobs may not benefit as much due to not being able fully ramp up (See DHFR JAC Prod). This GPU is also deployment flexible; up to 4x in an Exxact Valence Workstation, or 8x in a Exxact TensorEX 4U Server.
• NVIDIA RTX PRO 4500 Blackwell is a strong value option for smaller simulations (under ~100k atoms), with performance competitive with RTX 5000 Ada at a lower cost. You can scale up very quickly for labs running numerous calculations in parallel.

• The NVIDIA B200 SXM has amazing performance, but it is expensive for molecular dynamics. The B200 SXM, GH200, and H100 PCIe are all geared towards AI workloads; their high price tag makes them not the most price-to-performance friendly option for just MD simulation.

GPUs Benchmarked

The following Amber 24 Benchmarks were performed on an Exxact AMBER Certified MD System using the AMBER 24 Benchmark Suite with the following GPUs:

AMBER 26 Background Architecture & Hardware Recommendations

AMBER is a suite of molecular simulation tools, with PMEMD as the primary performance driver. PMEMD is available in CPU (single + MPI) and GPU (CUDA) builds. AMBER does not speed up one simulation by adding more GPUs; multi-GPU systems are best used to run multiple independent simulations in parallel.

Because most AMBER workloads run efficiently on a single GPU, overall throughput is typically dominated by GPU performance. CPU, memory, and storage generally have a smaller impact than the GPU (assuming a balanced workstation/server configuration).

Exxact recommends, for AMBER, the NVIDIA GeForce RTX 5090, NVIDIA RTX PRO 6000 Max-Q, and NVIDIA RTX PRO 4500.

NVIDIA GeForce RTX 5090 for AMBER

Exxact recommends NVIDIA GeForce RTX 5090 for the highest throughput for individual research running simulations sequentially. For peak single GPU throughput, the NVIDIA RTX 5090 in a workstation is the best performer at a lower price. If you don't need to run multiple simulations simultaneously, the RTX 5090 delivers the fastest results. The only disadvantages are the lack of scalability and the inability to support multi-GPU deployments.

NVIDIA RTX PRO 6000 Blackwell Max-Q for AMBER

NVIDIA RTX PRO 6000 Blackwell Max-Q is perfect for larger simulation models with the opportunity to scale. The NVIDIA RTX PRO 6000 Blackwell will showcases performance within 5% of the NVIDIA GeForce RTX 5090 for those looking at 2U and 4U server-only deployments with up to 8 and/or 10 GPU options.

NVIDIA RTX PRO 4500 Blackwell for AMBER

NVIDIA RTX PRO 4500 is ideal when running smaller simulations running in parallel at a lower cost. It comes in a dual-slot active GPU for workstations, and a single-slot passive GPU for extreme density in 2U and 4U servers. RTX 4500 offers excellent price-to-performance and scalability. In simulations with low atom count, the NVIDIA RTX PRO 4500 Blackwell matches the NVIDIA RTX PRO 6000 Blackwell Max-Q performance at a fraction of the cost.

Conclusion

Not all use cases are the same, and AMBER is likely just one of several applications in your research toolkit. AMBER is often one part of a broader research workflow, so the “best” system depends on the full application mix and budget. At Exxact Corp., we're committed to providing resources that help you configure the optimal custom system for your specific needs.

Since AMBER performance is typically GPU-driven, you may want to prioritize CPU, memory, and storage based on the other simulation packages you run (for example, GROMACS or NAMD can be more CPU-sensitive depending on the workload).

If you want help selecting a balanced configuration for multi-application Life Science workloads, Exxact can recommend options based on your target performance and budget.