Charges

Credits may be requested and used under the CPU Credit category or the GPU Credit category. Credits are specific to each node type – CPU Credits are not transferable to GPU Credits, and vice-versa. CPU Credit charges are defined by a combination of per-job cores and memory request, while GPU Credit charges are defined by a combination of per-job GPU, CPU, and memory request.

CPU Credits

Cores per job Credit charge per core hour
1 1.0
2-8 1.2
8-32 1.5
>32 2.0

When running on a resource with hyperthreaded cores, a 40% discount is applied; e.g., 1 hyperthreaded core for 1 hour costs 0.6 credits.

Additional CPU Credit Charge

Memory Per Core

When more than 2GB per core of memory is requested by the job, there’s an additional per-GB charge for memory.

Memory (GB) per job Credit charge per hour, per GB
Up to 2GB per core ("nominal") No charge
2-8 GB greater then nominal 0.125
8-32 0.25
32-128 0.375
128-512 0.50

CPU Credit Example

If an 8 core job requests 128 GB of RAM, there is an additional charge for 128 – (8 * 2) = 112 GB of RAM. If the job ran for an hour, it would use 0.375 * 112 = 42 credits total for memory and 1.2 * 8 = 9.6 credits for CPU.

GPU Credits

GPUs per job Credit charge per GPU hour
1 1.0
2 1.2
3 1.5
4 2.0

Additional GPU Credit Charge

CPUS per GPU

When more than 16 cores per GPU is requested by the job, there’s an additional per-core charge for the CPU cores beyond the nominal.

CPUs per GPU Credit charge per hour, per core
Up to 16 cores per GPU ("nominal") No charge
16-48 cores per GPU 0.125
48-64 0.20

Memory per GPU

When more than 2 GB per core of memory is requested by the job, there’s an additional per-GB charge for memory for the beyond-nominal memory usage.

Memory (GB) per job Credit charge per hour, per GB
Up to 128GB per GPU ("nominal") No charge
128-384 GB per GPU 0.012
384-512 0.020

GPU Credit Example

If a 1 GPU job requests 256 GB of RAM and 32 cores, there is an additional charge for 256 – (1 * 128) = 128 GB of RAM and an additional charge for 32 – (1 * 16) = 16 cores. If the job ran for an hour, it would use 0.012 * 128 = 1.536 credits total for memory, 0.125 * 16 = 2 credits for CPU, and 1.0 * 1 = 1.0 credits for GPU.