Procurement for data factory solutions
AI Sweden is opening up a Request for Proposals with the purpose of selecting candidates for establishing Framework Agreements related to data factory solutions for the Data Factory at AI Sweden.
To receive the Request for Proposals document, or to ask questions related to this Request for Proposals, please send an email to email@example.com.
Those providers who want to be a candidate for framework agreements, should submit their response to the Request for Proposals document to AI Sweden (firstname.lastname@example.org) no later than 16:00 (Swedish time) on February 26, 2021.
What is the buying company organization nr?
The buying organization is Lindholmens Science Park, org. nr: 556568-6366.
Who is the contact person for this RFP?
Main contact persons are Mats Nordlund and Ebba Josefson Lindqvist. They can also be reached via this email (email@example.com).
You ask about AI compute nodes from both Nvidia and non-Nvidia in Appendix 1, but it’s not separated in the Compute nodes part in Appendix 2. We would like a clarification in Compute Nodes in Appendix 2, if the partner is supposed to answer with both Nvidia DGX nodes and non-Nvidia nodes too. And are we free to spec the non-Nvidia nodes as we wish, as long as it fulfills the demand of 16 A100 cards?
It is expected that the Data Factory environment will consist of equipment from several suppliers (including Nvidia and others) going forward. This is part of the testbed and being an environment where partners can evaluate various types of equipment before or during their own procurement processes. Appendix 2 is an estimation of what may be procured in the initial phase and at this stage we are looking for a ballpark price tag for equipment of this performance. Exact details for what will be ordered at what stage will be defined when the framework agreements have been set up.
Proposals are expected to contain zero or more GPU node solutions and parties that has both DGX and non-DGX solutions in their portfolio are free to provide both options in their response.
Requirement for non-DGX solutions are 16x A100 GPU's and parties are free to spec these solutions according to their best experience. However, it is requested that solutions that differ a lot from the DGX platforms comes with detailed explanation in order for AI Sweden to be able to evaluate it fairly.
What General Terms and Condition will be used as the legal basis for the agreement, or will that be decided jointly in the next phase?
General terms and conditions have not been defined yet. This will be developed at a later phase.
In the RFP Appendix 2, it states that no software should be included in the offered solution. Please elaborate how you define software since some software may be needed as part of the hardware solution/reference architecture (software in storage arrays, server management software etc.)
Software required to manage, maintain and monitor the proposed platform is expected to be included. Also software required for any software-defined solutions, like software defined storage etc should be included. However, software running on top of the proposed hardware, such as operating systems, hypervisors, container platforms, scheduling solutions etc is not within the scope for the reference platform defined in appendix 2.
Visibility of the framework agreement, will this framework agreement only be visible for AI Sweden or also AI Sweden’s partners?
The existence of the framework agreements will be visible. However, the details of the framework agreements will be business confidential.
You have asked for high performance storage, could you be a little bit more specific in terms of the workload (needed IOPS, bandwidth, response time)?
Exact requirements for this is yet to be defined and here we will rely on the expertise of our framework agreement partners to get the optimal solution for our needs.
Workloads are as follows:
Standard base services such as domain controllers, management tools, network services, databases (lightweight, not project-specific high performance) etc.
Compute jobs (CPU) reading data from external network storage optimized for this type of dataset storage (or receiving from other external data source)
Compute jobs (CPU) that use local storage heavily for reading data and writing results.
The base services will be fairly stable and lightweight when it comes to resource usage but compute job usage will vary over time depending on the projects and partners that are active within the environment. From a storage solutions point of view, it is acceptable if compute jobs with high local i/o usage at times suffers a slight performance degradation at times of high loads, but impact to other jobs or services running against the same storage must be minimal.
Virtualization will be used and in most cases, a VM will only run one type of jobs (light or heavy on I/O).
Can we assume you expect same service level on storage as on compute (3Y NBD)?
Exact requirements for availability is yet to be defined, so this might change at when order for storage equipment is ordered. However, for the reference equipment, please price this according to 3 year next business day.