Asymmetric multi-core heterogeneous parallel processing system
A multi-core asymmetric graphics processing unit (GPU) includes a first group and second group of GPU cores. The first group of GPU cores has a first microarchitecture and a first power consumption profile. The first group of GPU cores is configured to execute a subset of instructions of an instruction set architecture (ISA). The second group of GPU cores have a second microarchitecture and a second power consumption profile higher than the first power consumption profile, and are configured to execute the entire ISA. The first group and second group of GPU cores may be further differentiated by a number of pipeline stages, number of registers, branching execution, vectorization units, or combinations thereof. A subset of GPU cores in either group may have a different operation frequency. In some embodiments, an executable instruction may include an indicator to ascertain if execution is performed by the first or second group of GPU cores.
1. An asymmetric multi-core heterogeneous graphics processing unit (GPU), the multi-core GPU comprising:
a first group of one or more GPU cores having each a first microarchitecture and a first power consumption profile, the first group of GPU cores configured to execute a first subset of an instruction set architecture (ISA); and
a second group of one or more GPU cores having each a second microarchitecture and a second power consumption profile higher than the first power consumption profile, the second group of GPU cores configured to execute the first subset of the ISA and a second subset of the ISA, wherein the second subset of the ISA includes at least an instruction not present in the first subset of the ISA, wherein the GPUs of both groups contain hardware logic to receive and execute instructions belonging to the same instruction stream that is part of an executable file, and further wherein a portion of the instructions of the ISA further comprise an indicator to determine if the instruction should be executed by a GPU of the first group of GPUs, or by a GPU of the second group of GPUs.
2. The multi-core GPU of claim 1 , further comprising a controller communicatively connected to the first group of GPU cores and the second group of GPU cores, the controller configured to receive an instruction of the ISA and determine if the instruction should be executed by a GPU of the first group of GPUs, or by a GPU of the second group of GPUs.
3. The multi-core GPU of claim 2 , wherein each instruction comprises an indicator bit so that the controller may determine if the instruction can be executed on: a GPU of the first group of GPU cores, a GPU of the second group of GPU cores, or on both.
4. The multi-core GPU of claim 1 , wherein the first group of GPU cores each have a fixed point data path.
5. The multi-core GPU of claim 1 , wherein the second group of GPU cores each have a floating point data path or floating point and fixed point data path.
6. The multi-core GPU of claim 1 , wherein the first group of GPU cores and the second group of GPU cores differ in: number of pipeline stages, number of registers, branching execution, vectorization units, or combinations thereof.
7. The multi-core GPU of claim 1 , wherein a first subset of GPU cores operates at a first frequency, and a second subset of GPU cores operates at a second frequency.
8. The multi-core GPU of claim 1 , wherein a subset of the first group of GPU cores utilize a first memory cache, and at least another GPU core of the first group utilizes a second memory cache.
9. The multi-core GPU of claim 6 , wherein a first subset of the second group of GPU cores utilizes any of: the first memory cache, the second memory cache and/or the third memory cache.
10. The multi-core GPU of claim 1 , wherein a first portion of a workload is executed by the first group of GPU cores, and a second portion of the workload is executed by the second group of CPU cores.
11. The multi-core GPU of claim 1 , wherein the first subset comprises one of more instructions that when executed by the second group of GPU cores has higher precision results than execution by the first group of GPU cores.
12. The multi-core GPU of claim 1 , wherein the second group of GPUs is configured to execute a second subset of the ISA, the second subset of the ISA comprising at least one instruction from the first subset, and another instruction of the ISA unique to the second subset.
13. The multi-core GPU of claim 1 , further comprising an instruction distributor communicatively coupled with the first group of GPU cores and the second group of GPU cores, operative for determining if the instruction should be executed by a GPU of the first group of GPUs, or by a GPU of the second group of GPUs.
14. The multi-core GPU of claim 1 , wherein a portion of the instructions of the ISA further comprise an indicator to determine a quality of service (Qos) level for execution of the instruction.
15. An asymmetric multi-core heterogeneous graphics processing unit (GPU) comprising:
a first group of one or more GPU cores having each a first microarchitecture and a first power consumption profile, the first group of GPU cores configured to execute a first subset of an instruction set architecture (ISA);
a second group of one or more GPU cores having each a second microarchitecture and a second power consumption profile higher than the first power consumption profile, the second group of GPU cores configured to execute the first subset of the ISA and a second subset of the ISA, wherein the second subset of the ISA includes at least an instruction not present in the first subset of the ISA, wherein a portion of instructions of the ISA further comprise an indicator to determine if the instruction should be executed by a GPU of the first group of GPUs, or by a GPU of the second group of GPUs;
a first z buffer connected to the first group of one or more GPU cores;
a first texture mapping unit connected to the first group of one or more GPU cores; and
one or more caches connected to the one or more of the first z buffer and the first texture mapping unit, wherein the GPUs of both groups contain hardware logic to receive and execute instructions belonging to the same instruction stream that is part of an executable file.
16. The asymmetric multi-core heterogeneous graphics processing unit (GPU) of claim 15 , further comprising:
a second z buffer connected to the second group of one or more GPU cores;
a second texture mapping unit connected to the second group of one or more GPU cores; and
a second cache connected to the one or more of the second z buffer and the second texture mapping unit.
17. The asymmetric multi-core heterogeneous graphics processing unit (GPU) of claim 15 , further comprising one or more primary general purpose GPUs connected to one or more of the one or more caches.
18. The asymmetric multi-core heterogeneous graphics processing unit (GPU) of claim 17 , further comprising one or more secondary general purpose GPUs connected to one or more of the one or more caches.
19. The asymmetric multi-core heterogeneous graphics processing unit (GPU) of claim 15 , further comprising an input/output memory management unit connected to one or more of the one or more caches.