Technologies for facilitating remote memory requests in accelerator devices

An accelerator device comprising:

• one or more slots;
• and
• circuitry to:
• receive, from a kernel on a slot of the one or more slots of the accelerator device and through an application programming interface exposed to a high level software language in which the kernel of the accelerator device is implemented, a request to read data at a memory address of a remote accelerator device;
• transmit, to the remote accelerator device and through a logical communication path established between the accelerator device and the remote accelerator device, a read request for the data at the memory address of the remote accelerator device;
• receive, from the remote accelerator device and through the logical communication path, the data at the memory address of the remote accelerator device;
• and
• provide, by the logical communication path, the data at the memory address of the remote accelerator device to the kernel of the accelerator device.

One or more non-transitory machine-readable media comprising a plurality of instructions stored thereon that, when executed by an accelerator device, causes the accelerator device to:

• receive, from a processor of a compute device, an instruction to automatically copy one or more parameters of a workload to one or more remote accelerator devices in response to receipt of the one or more parameters, the workload to be performed the accelerator device and the one or more remote accelerator devices;
• receive, from the processor of the compute device, the one or more parameters of the workload;
• and
• automatically send, in response to receipt of the one or more parameters of the workload, the one or more parameters of the workload to the one or more remote accelerator devices.

One or more non-transitory machine-readable media comprising a plurality of instructions stored thereon that, when executed by an accelerator device, causes the accelerator device to:

• receive, from a kernel of the accelerator device and through an application programming interface exposed to a high level software language in which the kernel of the accelerator device is implemented, a request to read data at a memory address of a remote accelerator device;
• transmit, to the remote accelerator device and through a logical communication path established between the accelerator device and the remote accelerator device, a read request for the data at the memory address of the remote accelerator device;
• receive, from the remote accelerator device and through the logical communication path, the data at the memory address of the remote accelerator device;
• and
• provide, by the logical communication path, the data at the memory address of the remote accelerator device to the kernel of the accelerator device.