# US 2015/0103085 A1 · Systems and Methods for Graphics Process Units Power Management

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2015/0103085 A1
- Document ID: US20150103085A1
- Status: Published application
- Kind: A1
- Application number: 14/513,289
- Filed: 2014-10-14
- Published: 2015-04-16
- Earliest claimed: 2013-10-14
- Assignees: Marvell World Trade Ltd.
- Applicants: Marvell World Trade Ltd.
- Inventors: Juntao Li; Steven Gao
- Web page: https://patentgrep.com/patent/US20150103085A1

## CPC

- G06T 1/20: Processor architectures; Processor configuration, e.g. pipelining
- G09G 5/363: {Graphics controllers}
- G09G 5/18: Timing circuits for raster scan displays (specially adapted for television H04N {; synchronisation between the display unit and other display units, videodisc player G09G5/12})
- G09G 2330/021: Power management, e.g. power saving
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G09G: ARRANGEMENTS OR CIRCUITS FOR CONTROL OF INDICATING DEVICES USING STATIC MEANS TO PRESENT VARIABLE INFORMATION (arrangements for transferring data between digital computers and displays G06F3/14; static indicating arrangements comprising an association of a number of separate sources or light control cells G09F9/00; static indicating arrangements comprising integral associations of a number of light sources H01J, H01K, H10H, H05B33/12; scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission, details thereof H04N1/00)

## Abstract

Systems and methods are provided for frequency adjustment of graphics process units (GPUs). A system includes: a command parser configured to parse one or more first commands associated with one or more future GPU operations to obtain command information, a processing component configured to determine an operation time for the future GPU operations based at least in part on the command information, and a frequency control component configured to adjust a GPU frequency based at least in part on the operation time for the future GPU operations.

## Claims

### Claim 1

A system for frequency adjustment of a graphics process unit (GPU), the system comprising:
a command parser configured to parse one or more first commands associated with one or more future GPU operations to obtain command information;
a processing component configured Co determine an operation time for the future GPU operations based at least in part on the command information; and
a frequency control component configured to adjust a GPU frequency based at least in part on the operation time for the future GPU operations.

### Claim 2 (depends on 1)

The system of claim 1, further comprising:
a historical information component configured to generate historical information associated with one or more previous GPU operations;
wherein the processing component is further configured to determine the operation time using the command information and the historical information.

### Claim 3 (depends on 2)

The system of claim 2, further comprising:
a performance counter component configured to determine one or more feedback parameters associated with the previous GPU operations and provide the feedback parameters to the historical information component.

### Claim 4 (depends on 2)

The system of claim 2, wherein:
the first commands include one or more three-dimensional commands;
the command information includes one or more of the following: a clear range (C.w, C.h), a vertex count (VC), a vertex shader instruction (VSI), a pixel shader instruction (PSI), and an average number of bytes per pixel of textures (tbpp); and
the historical information includes one or more of the following: a vertex cutting rate (VCR), a triangle culling rate (TCR), pixels per triangle (PPT), texture pixels of sampling (TPS), and a depth test pass rate (DTPR).

### Claim 5 (depends on 1)

The system of claim 1, wherein:
the first commands include one or more three-dimensional commands; and
the processing component is further configured to determine the operation time (Tbusy) as follows:
T
busy
=
max

(
TSI
f
shader
×
SC
,
TMB
TB
)
where TSI represents total GPU shader instructions, fshader represents a shader frequency, SC represents a shader count, TMB represents a total GPU memory bandwidth, and TB represents a GPU total bandwidth associated with the three-dimensional commands.

### Claim 6 (depends on 5)

The system of claim 5, wherein:
the processing component is further configured to determine a triangle count (TC) as follows:
T


C
=
(
VC
-
VCR
)
a
×
(
1
-
TCR
)
where VC represents a vertex count, VCR represents a vertex cutting rate, TCR represents a triangle culling rate, and a represents a first constant parameter;
the processing component is further configured to determine a pixel count (PC) as follows:
PC=TC×PPT
where PPT represents pixels per triangle;
the processing component is further configured to determine the total GPU shader instructions (TSI) as follows:
TSI=sum(VC×VSI)+sum(PC×DTPR×PSI)
where VSI represents a vertex shader instruction, DTPR represents a depth test pass rate, PSI represents a pixel shader instruction, and sum represents a summation operation; and
the processing component is further configured to determine the total GPU memory bandwidth (TMB) as follows:
TMB=((C.w×C.h)×b+PC×DTPR)×bpp+PC×TPS×tbpp
where (C.w, C.h) represents a clear range, bpp represents a number of bytes per pixel, tbpp represents an average number of bytes per pixel of textures, TPS represents texture pixels of sampling, and b represents a second constant parameter.

### Claim 7 (depends on 1)

The system of claim 1, wherein:
the first commands include one or more two-dimensional commands;
the command information includes one or more of the following: a source width, a source height, a source format, a number bytes per pixel of a source image (SBPP), a destination width, a destination height, a destination format, a number of bytes per pixel of a destination image (DBPP), a pixel count of the source image (PSB), a pixel count of the destination image (PDB), a filter quality, a rotation angle, and a zooming parameter.

### Claim 8 (depends on 1)

The system of claim 1, wherein:
the first commands include one or more two-dimensional commands; and
the processing component is further configured to determine the operation time (Tbusy) as follows:
T
busy
=
max

(
CC
f
2

D
,
TTB
TB
)
where CC represents a number of clock cycles associated with the two-dimensional commands, f2D represents a frequency associated with the two-dimensional commands, TTB represents a total bandwidth associated with the two-dimensional commands, and TB represents a GPU total bandwidth associated with the two-dimensional commands.

### Claim 9 (depends on 8)

The system of claim 8, wherein:
the processing component is further configured to determine the total bandwidth associated with the two-dimensional commands (TTB) as follows:
TTB=OB×(1+CP)
where OB represents an original bandwidth, and CP represents a cache penalty; and
the processing component is further configured to determine the number of clock cycles associated with the two-dimensional commands (CC) as follows:
CC
=
PDB
PPC
where PDB represents a pixel count of a destination image, and PPC represents a number of pixels per clock cycle.

### Claim 10 (depends on 9)

The system of claim 9, wherein:
the first commands include a stretch blit command; and
the processing component is further configured to determine the original bandwidth (OB) as follows:
OB=2×PDB×DBPP
where DBPP represents a number of bytes per pixel of the destination image.

### Claim 11 (depends on 9)

The system of claim 9, wherein:
the first commands include a filter blit command; and
the processing component is further configured to determine the original bandwidth (OB) as follows:
OB=PSB×SBPP+PDB×DBPP
where SBPP represents a number bytes per pixel of a source image, PSB represents a pixel count of the source image, and DBPP represents a number of bytes per pixel of the destination image.

### Claim 12 (depends on 9)

The system of claim 9, wherein:
the first commands include a high quality filter blit command; and
the processing component is further configured to determine the original bandwidth (OB) as follows:
OB=2×(PSB×SBPP+PDB×DBPP)
where SBPP represents a number bytes per pixel of a source image, PSB represents a pixel count of the source image, and DBPP represents a number of bytes per pixel of the destination image.

### Claim 13 (depends on 1)

The system of claim 1, wherein the frequency control component is further configured to:
determine a predicted GPU usage based at least in part on the operation time;
in response to the predicted GPU usage being above a threshold, increase the GPU frequency; and
in response to the predicted GPU usage being below the threshold, decrease the GPU frequency.

### Claim 14

A method for frequency adjustment of a graphics process unit (GPU), the method comprising:
parsing one or more first commands associated with future GPU operations to obtain command information;
determining an operation time for the future GPU operations based at least in part on the command information; and
adjusting a GPU frequency based at least in part on the operation time for the future GPU operations.

### Claim 15 (depends on 14)

The method of claim 14, further comprising:
generating historical information associated with one or more previous GPU operations;
wherein the operation time is determined using the command information and the historical information.

### Claim 16 (depends on 15)

The method of claim 15, wherein:
the first commands include one or more three-dimensional commands;
the command information includes one or more of the following: a clear range (C.w, C.h), a vertex count (VC), a vertex shader instruction (VSI), a pixel shader instruction (PSI), and an average number of bytes per pixel of textures (tbpp); and
the historical information includes one or more of the following: a vertex cutting rate (VCR), a triangle culling rate (TCR), pixels per triangle (PPT), texture pixels of sampling (TPS), and a depth test pass rate (DTPR).

### Claim 17 (depends on 14)

The method of claim 14, wherein:
the first commands include one or more three-dimensional commands; and
the operation time (Tbusy) is determined as follows:
T
busy
=
max

(
TSI
f
shader
×
SC
,
TMB
TB
)
where TSI represents total GPU shader instructions, fshader represents a shader frequency, SC represents a shader count, TMB represents a total GPU memory bandwidth, and TB represents a GPU total bandwidth associated with the three-dimensional commands.

### Claim 18 (depends on 14)

The method of claim 14, wherein:
the first commands include one or more two-dimensional commands; and
the operation time (Tbusy) is determined as follows:
T
busy
=
max

(
CC
f
2

D
,
TTB
TB
)
where CC represents a number of clock cycles associated with the two-dimensional commands, f2D represents a frequency associated with the two-dimensional commands, TTB represents a total bandwidth associated with the two-dimensional commands, and TB represents a GPU total bandwidth associated with the two-dimensional commands.

### Claim 19 (depends on 14)

The method of claim 14, wherein:
the first commands include one or more two-dimensional commands;
the command information includes one or more of the following: a source width, a source height, a source format, a number bytes per pixel of a source image (SBPP), a destination width, a destination height, a destination format, a number of bytes per pixel of a destination it age (DBPP), a pixel count of the source image (PSB), a pixel count of the destination image (PDB), a filter quality, a rotation angle, and a zooming parameter.

### Claim 20

A computer-implemented system for frequency adjustment of a graphics process unit (GPU), the system comprising:
one or more data processors; and
a computer-readable storage medium encoded with instructions for commanding the one or more data processors to execute operations including:
parsing one or more first commands associated with future GPU operations obtain command information;
determining an operation time for the future GPU operations based at least in part on the command information; and
adjusting a GPU frequency based at least in part on the operation time for the future GPU operations.
