Data Centers and Hardware Offerings¶
The resources offered by CONVERGE Horizon are managed by different cloud providers at various data centers around the world. These resources include cloud storage, compute nodes for running jobs, and visualization nodes for workstations.
Node Types¶
Nodes provisioned by CONVERGE Horizon are based on node types, which describe the hardware configurations users can choose from when creating a job or a workstation. The node types for jobs are optimized for compute-intensive workloads (i.e., running simulations), while the node types for workstations are optimized for graphics-intensive pre- and post-processing tasks. The list of supported node types is different for each data center.
Node types labeled as bare metal (BM) are allocated to one user at a time, meaning there is no shared use of processing cores, memory, or other hardware resources. Some BM nodes may run a hypervisor on top of the physical hardware, making them virtual machines in the strict sense. They are still classified as bare metal on CONVERGE Horizon if the hardware resources are dedicated to a single user.
Virtual machine (VM) nodes are allocated with the number of cores and the amount of memory requested by the user, allowing the remaining cores and memory to be allocated to other users. The cores allocated to a VM on CONVERGE Horizon are physical cores, not execution threads. For example, if you request a 12-core VM based on the VM-AMD-TURIN-FLEX node type, you will have 12 physical cores dedicated to your simulation, and these cores will not be split for the duration of your simulation. The processor’s remaining 244 cores and other hardware resources are shared with other users. Performance of VMs can vary due to the use of shared resources, although the dedication of physical cores does help stabilize performance.
Hardware specifications for CONVERGE Horizon node types are listed below. Refer to Oracle Cloud Data Centers and Google Cloud Data Centers to review the node types available at each data center.
Node Type |
Processor |
Cores per Node |
Memory (GB) |
Network |
Maximum Number of Nodes per Job |
|---|---|---|---|---|---|
BM-AMD-GENOA-192 |
AMD EPYC 9J14 |
192 |
2304 |
Ethernet |
1 |
BM-AMD-TURIN-192 |
AMD EPYC 9B45 |
192 |
720 |
Falcon |
15 |
BM-AMD-TURIN-256 |
AMD EPYC 9J45 |
256 |
3072 |
Ethernet |
1 |
BM-INTEL-ICELAKE-36 |
Intel Xeon 6354 |
36 |
512 |
InfiniBand |
50 |
BM-INTEL-SAPPHIRE-88 |
Intel Xeon 8481C |
88 |
352 |
Ethernet |
1 |
VM-AMD-GENOA-FLEX |
AMD EPYC 9J14 |
1-94 |
16-736* |
Ethernet |
1 |
VM-AMD-TURIN-FLEX |
AMD EPYC 9J45 |
1-126 |
16-1024* |
Ethernet |
1 |
* Oracle Cloud VM nodes require at least 1 GB of memory per core and cannot exceed 64 GB per core.
Node Type |
Processor |
Cores |
Memory (GB) |
GPU |
|---|---|---|---|---|
VM-INTEL-GPU-15 |
Intel Xeon Platinum 8358 |
15 |
240 |
NVIDIA A10 24 GB RAM |
VM-INTEL-GPU-30 |
Intel Xeon Platinum 8358 |
30 |
480 |
NVIDIA A10 24 GB RAM |
Hardware Recommendations¶
The optimal number of cores for a simulation on CONVERGE Horizon depends on the number of computational cells and the type of calculation being performed. Approximately 3,000 cells per core is a reasonable starting point for simulations of non-reacting flows. You can find a more detailed discussion of parallel processing considerations for the CONVERGE solver in the CONVERGE Manual (available on the Convergent Science Hub).
When you have identified the target number of cores for your job, select a node type with the following recommendations in mind:
Use a single VM node for cases that require 126 or fewer cores. The VM nodes are very cost-effective because they offer per-core licensing costs with the option to customize the number of cores.
Use a single BM-AMD node for small and medium-sized cases with high memory requirements. To optimize performance, choose a node type based on Turin processors; to minimize cost, choose a node type based on Genoa processors.
Use multiple BM-INTEL-ICELAKE-36 nodes for cases that require more than 256 cores. These nodes have a high-performance interconnect to enable excellent scaling for large cases.
Pricing information for each node type is available at convergecfd.com.
Oracle Cloud Data Centers¶
File systems and spaces are available for cloud storage at all Oracle Cloud data centers. The node types supported at each data center are listed below.
Data Center |
Available Node Types for Jobs |
Available Node Types for Workstations |
|---|---|---|
Canada Southeast (Toronto) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
|
Germany Central (Frankfurt) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
India West (Mumbai) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
India South (Hyderabad) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
VM-AMD-GENOA-FLEX
|
|
Italy Northwest (Milan) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
Japan Central (Osaka) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
Japan East (Tokyo) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
South Korea Central (Seoul) |
BM-AMD-GENOA-192
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
UK South (London) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
US East (Ashburn) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
US West (Phoenix) |
BM-AMD-GENOA-192
BM-AMD-TURIN-256
BM-INTEL-ICELAKE-36
VM-AMD-GENOA-FLEX
VM-AMD-TURIN-FLEX
|
VM-INTEL-GPU-15
VM-INTEL-GPU-30
|
Google Cloud Data Centers¶
File systems are the only cloud storage option at Google Cloud data centers. The node types supported at each data center are listed below.
Data Center |
Available Node Types for Jobs |
|---|---|
Netherlands (Eemshaven) |
BM-AMD-TURIN-192, BM-INTEL-SAPPHIRE-88 |
Singapore West (Jurong) |
BM-AMD-TURIN-192, BM-INTEL-SAPPHIRE-88 |
US Central (Council Bluffs) |
BM-AMD-TURIN-192, BM-INTEL-SAPPHIRE-88 |