I modified bulkloopauto firmware in the example of FX3 SDK, and measured 35MBytes/sec data speed on each endpoint.

Is there an idea to speed up the data rate? (at least 100MBytes/sec)

Hi,

What is the environment being used to test the throughput i.e. Host controller, throughput measuring tool etc. The max data rate that we have seen with Windows USB hosts with NEC USB3 controllers is 110MB/s.

But we've not made the kind of per endpoint measurement that you're making. Please let us know the kind of modification you've done to the example code as well.

Regards,

Anand

I tested bulkloop with NEC USB3 controllers as same as your environment.

and modified host software(removing updateUI function) with cyfxbulklpauto(changed dma buffer count to 64).

// Call StatusUpdate() in the main thread
                //this.Invoke(updateUI);

private void timer1_Tick(object sender, EventArgs e)// 1000msec interval
        {
            this.textBox1.Text = (((outCount / 1024)/1024).ToString()+"MB");// measured max 35MBytes/sec
            outCount = 0;
        }

#define CY_FX_BULKLP_DMA_BUF_COUNT      (64) //changed to 64

I want to second this question.   I couldn't get anything more than about 60 MB/s out of my testing.  I'm using a Renesas-based PCIe to USB3 host controller, 8192 bytes/transfer, 1024 bytes per packet in Bulk Loop Auto.  Against BulkSrcSink, I get about 30MB/s.  I would expect it to be the higher one.

Why doesn't the sample application include throughput timing?  Maybe some instruction on how to allow block sizes big enough to make a dent on the overhead?

I'd like to see the test code and setup that allows Cypress' claim of 320MB/s.

I'm sorry, I ran out of time on the last question.  The 320 MB/s is touted as the GPIF-II speed.  Is there any kind of example showing that?

Is the 110 MB/s mentioned above due to OS limitations, Driver limitations, FX3 limitations?  Can Cypress publish the test case they used to measure it?  Why isn't it in the example sets?

Another customer has been able to achieve 120MB/s. You might want to take a look at http://www.cypress.com/?app=forum&id=167&rID=53351.
Regards,

Anand
 

I've the same problem... I can't get more than 26 MB/s from the BulkSrcSink which using the streamer application. I'm also using a Renesas USB3 host.

I really think Cypress should provide a benchmark utility and a firmware which demonstrates something nearer to what one would expect from a USB3 device.

I wonder if there is at least one of the reference designs provided by Cypress which is able to do transfers above 70..90 MB/s on the development board (without using the GPIB, just fake data)

My problem is that I’ve to design a product around the FX3 which has to provide continuous data transfer up to 100 MB/s, but I can’t go on as long as I’m not sure the FX3 will ever reach this transfer speed.

Did you arranged to get an acceptable transfer speed  out of the FX3 ?

Regards, Joel

Hi,

It has dependency on the host as well in terms of the bandwidth it allocates, We tested by having single bulk endpoint with burstLen of 7 and observed 110MB/s. One other customer has been able to obtain 120MB/s. So yes, FX3 is capable of supporting higher bandwidth that what you're observing but there are dependencies on the host.

Once host controller with better performance are available and Windows supports USB3.0 natively then we should be able to further testing and update the numbers but at this point of time 110MB/s is the value we observed. We'll doing further testing in the future.

Regards,

Anand

Ok, thanks for your answer... but there are still things that I do not understand :

1) you reached 120 MB/s with which firmware and which application ( out of the SDK you are providing ) ? Please give me the name of a firmware in your SDK which is able to reach 110 MB/s provided I use the correct USB3 host controller!

2) you are stating that the difference between 50 and 120 MB/s is only due to the USB3 host controller ?

3) where do I have to change the burstlen  to 7 ?  " ....single bulk endpoint with burstLen of 7"  ...  where do I change the burst length ?  on the examples you are providing with the SDK you always use burst length of 0 !  Again, which firmware did you used to get 110 Mbytes/s ?

P.S. I can’t use slave fifo or GPIF based examples since I’ve just a Cypress development board without any additional HW, which is by the way the typical setup of a customer who does a preliminary FX3 evaluation.

Thanks in advance, Joel

just to avoid confusions...

you measured 110 MB/s on the bulk loop example using a NEC USB3 controller ?

If no please  post the firmware you used.

Tanks Joel

sorry .... typo            "...  NEC USB3 host controller"...

... and some more points...

I tested the USB 3 host controller by transferring a 10 GB file into an external USB 3 HD and the average transfer speed was above 100 MB/s (and I'm sure the bottleneck is the disk and not the USB)... this means that the USB host controller is able to provide by far more bandwidth than the one I'm measuring with the FX3 using the reference firmware provided by Cypress

Regards, Joel

The 110MB/s was using modified bulksrcsink to the parameters I earlier specified.

http://www.cypress.com/?app=forum&id=167&rID=53351 shows another user observing upto 120MB/s using the slavefifo sync firmware.

If you need help support on the firmware modification details please create a support case (MyAccount->MyCases) so that we can have a deeper look at what might be going wrong in your setup.

Regards,

Anand