Flexible memory system
A technique for operating a memory system is disclosed. The technique includes performing a first request, by a first memory client, to access data at a first memory address, wherein the first memory address refers to data in a first memory section that is coupled to the first memory client via a direct memory connection; servicing the first request via the direct memory connection; performing a second request, by the first client, to access data at a second memory address, wherein the second memory address refers to data in a second memory section that is coupled to the first client via a cross connection; and servicing the second request via the cross connection.
1 . A method comprising:
requesting access to a first memory address for a first memory client of a plurality of memory clients, the first memory address being mapped to a first memory on a first chip coupled to the first memory client via a first direct connection through an interconnect die;
servicing the request to the first memory address via the first direct connection;
requesting access to a second memory address for the first memory client, the second memory address being mapped to a second memory on a second chip coupled to the first memory client via a cross connection between the first direct connection and a second direct connection through the interconnect die; and
servicing the request to the second memory address via the cross connection.
2 . The method of claim 1 , further comprising:
requesting access to the first memory address by a second memory client.
3 . The method of claim 2 , wherein the second memory client is on the same chip as the first memory client.
4 . The method of claim 2 , wherein the second memory client is on a different chip as the first memory client.
5 . The method of claim 1 , wherein the direct connection has a higher bandwidth than the cross connection.
6 . The method of claim 1 , wherein the direct connection has a lower latency than the cross connection.
7 . The method of claim 1 wherein the first memory and the second memory form part of a physical address space.
8 . A system comprising:
a first chip including a first memory;
a second chip including a second memory; and
a first memory client coupled to the first memory via a direct connection through an interconnect die and to the second memory via a cross connection between the direct connection and a second direct connection through the interconnect die, the first memory client configured to:
request access to a first memory address mapped to the first memory, and
request access to a second memory address mapped to the second memory;
wherein the first memory is configured to service the access to the first memory address via the direct connection; and
wherein the second memory is configured to service the access to the second memory address via the cross connection.
9 . The system of claim 8 , further comprising:
a second memory client configured to request access to the first memory address.
10 . The system of claim 9 , wherein the second memory client is on the same chip as the first memory client.
11 . The system of claim 9 , wherein the second memory client is on a different chip as the first memory client.
12 . The system of claim 8 , wherein the direct connection has a higher bandwidth than the cross connection.
13 . The system of claim 8 , wherein the direct connection has a lower latency than the cross connection.
14 . The system of claim 8 , wherein the first memory and the second memory form part of a physical address space.
15 . A system comprising:
a first chip including a first memory;
a second chip including a second memory;
a first memory client;
a second memory client; and
an interconnect die that includes a first direct connection between the first chip and the first memory client, a second direct connection between the second chip and the second memory client, and a first cross connection between the first direct connection and the second direct connection,
wherein the first memory client is coupled to the first memory via the first direct connection and to the second memory via the first cross connection; and
wherein the second memory client is coupled to the first memory via a second cross connection and to the second memory via the second direct connection,
wherein the first memory client is configured to:
request access to a first memory address mapped to the first memory, and
request access to a second memory address mapped to the second memory;
wherein the first memory is configured to service the access to the first memory address via the first direct connection; and
wherein the second memory is configured to service the access to the second memory address via the first cross connection.
16 . The system of claim 15 , further comprising:
wherein the second memory client is configured to request access to the first memory address.
17 . The system of claim 16 , wherein the second memory client is on the same chip as the first memory client.
18 . The system of claim 16 , wherein the second memory client is on a different chip as the first memory client.
19 . The system of claim 15 , wherein the direct connection has a higher bandwidth than the cross connection.
20 . The system of claim 15 , wherein the direct connection has a lower latency than the cross connection.