Hardware

Joint projects with Intel keep us fairly updated on recent hardware progressive technology. Based on years of research in high performance computing and parallel algorithms in reservoir simulation we gained a high level of expertise with multicore hardware of different types.

We offer our clients utilization of our know-how when looking for optimal software and hardware configurations for reservoir engineering problems. We can recommend or deliver hardware systems based on client needs optimizing the price-quality relationship.

Requirements for RAM per core.
We have no minimal requirements for RAM per core. Everything depends on model. We use 3kB RAM per active grid block for black-oil models.

An example for cluster with the following configuration: Xeon 5650 node, 12 cores, 24Gb.

 – One node simulation. We can run 24000k/3k=8million active grid blocks. We have run successfully the real model 6.5 million blocks on cluster with this configuration.
 – MPI-version. Simulation on several nodes. For MPI run maximal size of the model multiplies by number of nodes per run. For the cluster with this configuration we can run model with 12 million active grid blocks using 2 nodes and 23 million active with 4 nodes (there is small overhead for domains overlapping in MPI run so maximal size is less than theoretical maximum).

Hardware Recommendations

Workstation

The price range is about $10,000

Example on the following workstations:

2011: Dual Xeon X5650, (2×6) 12 cores, 2.66GHz, 3 channels DDR3 1333 MHz (e.g. HP Z800)

2012: Dual Xeon E2680, (2×8) 16 cores, 2.7GHz, 4 channels DDR3 1600 MHz (e.g. HP Z820)

2013: Dual Xeon E2697v2, (2×12) 24 cores, 2.7GHz, 4 channels DDR3 1866 MHz (e.g. HP Z820)

2014: Dual Xeon E2697v3, (2×12) 28 cores, 2.6GHz, 4 channels DDR4 2133 MHz (e.g. HP Z840)

2016: Dual Xeon E2699v4, (2×22) 44 cores, 2.2GHz, 4 channels DDR4 2400 MHz (e.g. HP Z840)

tNavigator on workstations

The standard commercial configuration could be found here: HP Z820 Workstation

Cluster

Here is an example for 8 computing nodes, 160-core configuration.

The price range is about $80,000

Each computing node contains 20-core dual Xeon E5 2680v2 with 2.8 GHz,

128GB DDR3 1866GHz RAM, Infiniband network switch with 56Gb/s

Typical parallel acceleration:

Model 1.  2.4 million active grid blocks, single porosity, 3-phase blackoil model, gas cap, and multiple     equilibrium regions, 39 wells. The parallel acceleration factor is 73 times

Model 2.  460 thousand active grid blocks, dual porosity, 3 phase blackoil model with 4 horizontal wells(~3km each) and multiple hydraulic fractures represented by 250 Local
Grid Refinements(LGRs). It is an example of shale oil reservoir. The parallel acceleration factor is 40 times

Model 3. 1.9 million active grid block, compositional EOS model, with 6 components and CO2 injection, 4 horizontal wells(~3km long). The parallel acceleration factor is 80 times

More hardware details could be found here: Hardware Specification