JPEGD Performance Specifications

For Atlas training products, this function is reserved and is not supported currently.

Performance Specifications

This section describes only the performance specifications of Ascend AI Processors of each product, which may differ based on the memory and CPU configurations of the respective product. For details about the product performance specifications, see the official descriptions at the Ascend community.

This section uses a single Ascend AI Processor as an example to describe the performance indicators in basic scenarios. The performance indicators of other resolutions can be estimated in the same way.

In the following table, one channel corresponds to one thread, 720p indicates a resolution of 1280 x 720 pixels, 1080p indicates a resolution of 1920 x 1080 pixels, and 4K indicates a resolution of 3840 x 2160 pixels.

The performance specifications of JPEGD are based on hardware decoding performance. JPEGD hardware decoding does not support the image decoding with three SOS markers. For image formats not supported by hardware, software decoding is used. The software decoding performance for reference is 1080p at 15 FPS. If the output image decoded by JPEGD involves rotation, the performance is lower than the reference values of software decoding. For example, the performance is lower than 15 FPS for 1080p images.

On Atlas A3 training products/Atlas A3 inference products, an NPU has two devices, and the maximum specifications can be reached only when the two devices work in parallel.

**Table 1** Performance specifications of AI processor on Ascend 950PR/Ascend 950DT
Input Format	Input Image Size	Scenario	Total Frame Rate
YUV400/420/YUV440/422/444	≤ 500 KB	n-channel 1080p (1 ≤ n ≤ 8)	n × 512 FPS
YUV400/420/YUV440/422/444	≤ 500 KB	n-channel 1080p (n > 8)	4096 FPS
YUV400/420/YUV440/422/444	≤ 2 MB	n-channel 4K (1 ≤ n ≤ 8)	n × 128 FPS
YUV400/420/YUV440/422/444	≤ 2 MB	n-channel 4K (n > 8)	1024 FPS

**Table 2** Performance specifications of AI processor on Atlas A3 training products/Atlas A3 inference products
Input Format	Input Image Size	Scenario	Total Frame Rate
YUV400/420	≤ 500 KB	n-channel 1080p (1 ≤ n ≤ 56)	n × 512 FPS
YUV400/420	≤ 500 KB	n-channel 1080p (n > 56)	28672 FPS
YUV440/422/444	≤ 500 KB	n-channel 1080p (1 ≤ n ≤ 56)	n × 256 FPS
YUV440/422/444	≤ 500 KB	n-channel 1080p (n > 56)	14336 FPS
YUV400/420	≤ 2 MB	n-channel 4k (1 ≤ n ≤ 56)	n × 128 FPS
YUV400/420	≤ 2 MB	n-channel 4k (n > 56)	7168 FPS
YUV440/422/444	≤ 2 MB	n-channel 4k (1 ≤ n ≤ 56)	n × 64 FPS
YUV440/422/444	≤ 2 MB	n-channel 4k (n > 56)	3584 FPS

**Table 3** Performance specifications of the Ascend AI Processor on Atlas inference products
Scenario	Total Frame Rate
n-channel 1080p (1 ≤ n ≤ 16)	n × 128 FPS
n-channel 1080p (n > 16)	2048 FPS
n-channel 4K (1 ≤ n ≤ 16)	n × 32 FPS
n-channel 4K (n > 16)	512 FPS

**Table 4** Performance specifications of the Ascend AI Processor on Atlas 200I/500 A2 inference products
Scenario	Total Frame Rate
n-channel 1080p (1 ≤ n ≤ 2)	n × 256 FPS
n-channel 1080p (n > 2)	512 FPS
n-channel 4K (1 ≤ n ≤ 2)	n × 64 FPS
n-channel 4K (n > 2)	128 FPS

**Table 5** Performance specifications of Atlas A2 training products/Atlas A2 inference products
Input Format	Input Image Size	Scenario	Total Frame Rate
YUV400/420	≤ 500 KB	n-channel 1080p (1 ≤ n ≤ 28)	n x 512 FPS
YUV400/420	≤ 500 KB	n-channel 1080p (n > 28)	14336 FPS
YUV440/422/444	≤ 500 KB	n-channel 1080p (1 ≤ n ≤ 28)	n × 256 FPS
YUV440/422/444	≤ 500 KB	n-channel 1080p (n > 28)	7168 FPS
YUV400/420	≤ 2 MB	n-channel 4K (1 ≤ n ≤ 28)	n × 128 FPS
YUV400/420	≤ 2 MB	n-channel 4K (n > 28)	3584 FPS
YUV440/422/444	≤ 2 MB	n-channel 4K (1 ≤ n ≤ 28)	n × 64 FPS
YUV440/422/444	≤ 2 MB	n-channel 4K (n > 28)	1792 FPS

Parent topic: VDEC/JPEGD APIs