Invest in Compute Hardware

The minimum buy-in is one compute node. The standard configuration has dual CPUs and one 1 TB SATA disk drive. If you are interested in purchasing nodes with special configurations (e.g. very large memory, different CPUs, GPUs, etc.), contact us for pricing.

All equipment and infrastructure costs are approved to be charged to grants. Infrastructure fees are considered service fees and must be charged monthly to grants. See the “Pricing Notes” section for more details.

NOTE: Pricing below current as of 12/31/2020

Traditional CPU Nodes

CPU nodes are intended for traditional CPU-based computing applications. Four nodes fit in 2U of rack space.

Manufacturer & CPUTraditional CPU ComputeLarge Memory CPU Compute
Intel Xeon 6248 CPU
20-core / 40-thread processor
2.5 GHz Base Frequency
HPE XL170 1/2U Server w/ Dual Intel Xeon 6248 (40 cores)
$9,933.25  –  192GB RAM
$10,798.57 – 384GB RAM
None preconfigured. Contact us for configurations.
AMD EPYC 7702 CPU
64-core / 128-thread processor
2.0 GHz Base Frequency
HPE XL225 1/2U Server w/ Dual AMD EPYC 7702 (128 cores)
$10,736.72 – 256GB RAM 
$13,236.72 – 512GB RAM
$17,191.72 – 1TB RAM
HPE DL385 2U Server w/ Dual AMD EPYC 7702 CPU
$25,527.44 – 2TB RAM 
$54,761.37 – 4TB RAM

Mixed CPU/GPU Nodes

Traditional server form-factors that allow for CPU-based computing with small numbers of GPUs. The above Traditional CPU Node types come in server chassis form factors that are often capable of accepting 1 to 3 GPUs for accelerated computation. Below you will find a table with pricing for individual GPU types which can be integrated into standard CPU compute node servers in various configurations. If you’re interested in pricing out a hybrid CPU/GPU compute node, please contact the ICCP for pricing and configuration validation.

GPU Nodes for AI/ML

GPU-dense compute nodes well-suited to machine learning. Up to 8 GPUs maximum per 4U or 5U server. 

Graphics ProcessorHPE – Apollo 6500 ServerLamba Labs – Lambda Blade Server
Consumer, Single-Precision
NVIDIA GeForce RTX 3090 GPU
None Available.$26,596.50
Dual AMD EPYC 7302 CPUs8x NVIDIA GeForce 3090 24GB GPUs128GB RAM
Enterprise, Single-Precision
NVIDIA RTX6000, A6000 &
A40 GPUs
$42,832.68
Dual Intel Cascade Lake 4214R CPUs
8x NVIDIA RTX6000 24GB GPUs192GB RAM
$44,714.15
Dual AMD EPYC 7302 CPUs8 NVIDIA A6000 48GB GPUs128GB RAM
Enterprise, Double-Precision
NVIDIA A100 GPU
$74,832.68
Dual Intel Cascade Lake 4214R CPUs
8x NVIDIA A100 40GB GPUs192GB RAM
$67,068.52
Dual AMD EPYC 7302 CPUs8x NVIDIA A100 40GB GPUs128GB RAM

GPUs

Graphics Card ManufacturerPrevious Generation GPUsCurrent Generation GPUs
NVIDIA$2,195NVIDIA Tesla T4 16GB 
$899    – NVIDIA Quadro RTX4000 8GB
$3,499NVIDIA Quadro RTX6000 24GB
$5,549NVIDIA Tesla V100 32GB
(TBD)   – NVIDIA A6000 48GB
$3,999NVIDIA A40 48GB
$7,499NVIDIA Tesla A100 40GB 

Pricing Notes

All pricing listed above is per compute node and includes the Campus Cluster’s infrastructure fees covering 5 years of shared cluster infrastructure. An HDR100 InfiniBand high-speed interconnect is available optionally and will add an additional $1,235 to the per node cost. If billed to a grant, infrastructure fees must be billed monthly.

Purchase Compute Time On-Demand (RCaaS)

The Government Costing approved rate for RCaaS is $0.0217 per core-hour, which works out to $15.84 per core-month. This service is exempt from overhead when charged to grants. Place an order here!

Our FAQ page has more detailed information on commonly asked questions for RCaaS, such as available nodes, configuration options, etc.

The general rules for billing are:

  1. Investors will be charged the same rate no matter which nodes execute your jobs, and will only be charged for actual time used (rounded up to the next minute). Minimum job runtime is one hour.
  2. Jobs are scheduled such that only one job will run on one node at a time. Jobs will be charged for all cores on the node, regardless of whether or not they are used.
  3. The ICCP holds no liability for jobs lost due to user error and no refunds will be made in those cases. However, should a job fail due reasons outside of the control of the user (e.g. operator error, environmental failures), a refund for time lost can be made.
  4. Usage will be billed monthly to the University of Illinois CFOP you provide, and charges can be viewed on your Technology Services Statement via the Pinnacle application. Customers agree to keep on file with ICCP a current CFOP with sufficient funds to cover the expected charges to their account. It is not possible to pre-pay for service under this model, therefore you must have funding available each month to pay for your usage of RCaaS resources.

NOTE: All ICCP customers will be required to sign the Service Level Definition for the product or service they purchase.  These SLDs outline policies related to these purchases.  Find the appropriate SLD for the product or service you’re considering here.