The performance of sgx544mp

PowerVR Insider
1

I spent a lot of time searching your wbesite.
But I can’t find the performance about sgx544mp.

I want to know the triangle throughput and pixel fill rate data

I am developing adas product with sgx544mp which brings 3,000,000$ income since last year.
Urgent surport is needed.

2

If possible , I need the data of all series.
We are planning to change our HW from TI to xlinks, which means we will use mali 400 instead of sgx544.
Because perfomrce of sgx544 is not satisfied.

3

Hi Lin,

I have follow up with you directly by email. SGX544MP2 is very capable and I do not think that moving to a different architecture will help you without an good understanding of what is causing the performance loss.

Best regards.

Carlos.

4

I test sgx544mp2 with android 6.0 last month.
The perfomance is satisfied.
However the cost must be considered.

The question is
"What is the triangel output rate and pixel fill rate about sgx544"
and what about other powerVR gpu?

5

I just want to know the data
I ask TI for help,they give me the test reslut below:

GLBenchmark 2.5 Vertex throughput on DRA7xx
Benchmark Test Number
Mtriangles/sec
DRA75x DRA72x(mp2) DRA71x(mp)
46.64 20.25 18.72

GLBenchmark 2.5 pixel throughput on DRA7xx
DRA75x DRA72x(mp2) DRA71x(mp)
1476.32 775.01 688.72

Is that accurate?

6

We use SGX544MP as first choice because it is cheap.
But the performance is not good.

We have test SGX544MP2, which obviously is no problem.
But too expensive.

Show me the data, so I can deside which gpu I will choose.
I will consider both the performance and the cost.

Show me the data! Show me the data! Show me the data!

7

Sorry, the data of the table above has some mistake.

GLBenchmark 2.5 Vertex throughput on DRA7xx
Benchmark Test Number
Mtriangles/sec
DRA75x(mp2) DRA72x DRA71x(mp)
46.64 20.25 18.72

GLBenchmark 2.5 pixel throughput on DRA7xx
DRA75x(mp2) DRA72x DRA71x(mp)
1476.32 775.01 688.72

8

Hi Lin,

Those figures looks sensible to me. In terms of performance MP2 doubles MP. Probably clock speeds are different between these cores so you need to take this into consideration.
Another thing you need to consider is that the most common bottleneck is bandwidth. Memory accesses will limit real-word applications. Our architecture can reject any pixel that is occluded early in the pipeline, saving bandwidth and making well written applications faster than theoretical peaks.
What kind of applications you intend to run on your device and at which resolution?

Regards.

Carlos.

9

Thanks for sharing this information

perdisco assignment help