Understanding Panel Self Refresh
by Anand Lal Shimpi on August 7, 2013 1:47 PM EST- Posted in
- Smartphones
- Displays
- Mobile
- Panel Self Refresh
- Tablets
- SoCs
Earlier today Brian spent some time with the G2, LG's 5.2-inch flagship smartphone based on the Qualcomm Snapdragon 800 (MSM8974) SoC. I'd recommend reading his excellent piece in order to get all of the details on the new phone, but there's one disclosure I'd like to call out here: the G2 supports Panel Self Refresh.
To drive a 60Hz panel, your display controller must present the display with the contents of the frame buffer 60 times per second. Regardless of what's being displayed (static vs. active content), every second there are 60 updates pushed through the display pipeline to the display. When displaying fast moving content (e.g. video playback, games, scrolling), this update frequency is important and appreciated. When displaying static content however (E.g. staring at the home screen, reading a page of an eBook), the display pipeline and associated DRAM are consuming power sending display updates when it doesn't need to. Panel Self Refresh (PSR) is designed to address the latter case.
To be clear, PSR is an optimization to reduce SoC power, not to reduce display power. In the event that display content is static, the contents of the frame buffer (carved out of system RAM in the case of a smartphone) are copied to a small amount of memory tied to the display. In the case of LG's G2 we're likely looking at something around 8MB (1080p @ 32bpp). The refreshes then come from the panel's memory, allowing the display pipeline (and SoC) to drive down to an even lower power state. Chances are the panel's DRAM is also tied to a narrower bus and can be lower power than the system memory used by the SoC, making these refreshes even lower in power cost.
LG claims a 26% reduction in power when displaying a still image with PSR enabled. I'm curious to see the impact on overall battery life. There are elements of our WiFi web browsing test that could benefit from PSR but it's unclear how much of an improvement we'll see. The added cost of introducing additional memory into a device is something that panel vendors have been hesitant to do, but as companies look to continue to reduce platform power it's a vector worth considering. LG's dual-role as a component supplier and device maker likely made the decision to enable PSR a lot easier.
PSR potentially has bigger implications for notebook use where it's not uncommon to just stare at a desktop that's not animating at all. I feel like the more common use case in smartphones is to just lock your phone/display when you're not actively using it.
Facebook
Twitter
RSS
38 Comments
View All Comments
Rick83 - Thursday, August 8, 2013 - link
It's about the GPU, that requires more power to run just to keep the frame buffer alive, on expensive high-bandwidth memory, when a lower bandwidth dedicated frame buffer memory can do so with less power consumption.The memory is just between the high-performance frame buffer, and the display. Kind of a LITTLE.big implementation of the frame buffer, where you double the amount of hardware, so you can optimize for two different usage profiles.
TrackSmart - Thursday, August 8, 2013 - link
Anand, based on the comments (and my own questions), it might be helpful if you added a few words that better explain what this "26% reduction in power" means. Is this a 26% reduction in SOC power consumption? Or a 26% reduction in the GPU portion of the SOC's power consumption? Or something else entirely.And what proportion of total power consumption, assuming a static state, does the SOC/GPU/whatever use on these thrifty mobile devices? If the display is using 85% of the power, then a 26% reduction in SOC power usage would be small (4% reduction in total power usage), but not insignificant. If we are talking about a 26% reduction in something that uses even less power, this would be a smaller influence still. Possibly to the point of being unmeasurable in a battery life test...
Anyway, some further context would be helpful.
juhatus - Friday, August 9, 2013 - link
Doesnt the new Haswell laptops (for example Sony Vaio Pro) use this?n4s33r - Wednesday, August 14, 2013 - link
The 2013 Nexus 7 also has this featurewumpus - Sunday, August 18, 2013 - link
If you are actively changing the display, but not all of it (say, browsing with an animation in a sidebar), you will likely benefit from the fact that your RAM isn't being grabbed to output a framebuffer all the time (although I doubt you will notice it even on a phone. The last time I heard memory DMA mattered was programming an 8-bit ATARI 400/800 (although this is largely due to the fact that any PC designed remotely for performance has a video card)).On the other hand it requires a separate buffer for the video card (think the latest Intel onboard graphics with its own memory bus). 1080 graphics is something like 8M of memory, which is an odd duck (too large for SRAM on a reasonable sized chip, too small for DRAM). I'd guess it will take a few SRAM chips, that will add heat & power issues.
DesktopMan - Monday, August 19, 2013 - link
"The last time I heard memory DMA mattered was programming an 8-bit ATARI 400/800"DMA is still extremely important for computing, it just mostly happens in the OS and drivers now. By reducing the load on main memory you are not only lowering power consumption but also freeing up system resources. Feeding a 2560*1600 panel at 60hz actually requires ~1GB/sec memory bandwidth continuously at 32bpp.
psyq321 - Wednesday, August 21, 2013 - link
DMA matters still the same today - if you a kernel/driver programmer.You would want to use DMA if it is available, to avoid clogging the CPU with the memory xfer. This is almost universally done in the graphics drivers whenever there is a need to upload the chunk of data to the VRAM, etc.
Cheiz - Saturday, May 25, 2019 - link
Panel Self Refresh is enabled by default on the Dell XPS 9380 FHD 13 and it works like shit. See https://www.dell.com/community/XPS/XPS-9380-FHD-13...