Difference between STM32N657X0 and STM32H753XI

[DJ_IND]

Can we use STM32N series high end crypto version in medical devices? Currently we are using H series from ST but we are facing some issue in CPU utilization and speed. Apart from external flash is there any major difference where N series is lagging with respect to H series part??

 

Is there any limitation to use N series in MATLAB?? Any limitation to us

 

Link for n series

https://www.st.com/en/microcontrollers-microprocessors/stm32n657x0.html

 

 

Br,

DJ

[Andrew Neil]

---

@DJ_IND wrote:

Is there any limitation to use N series in MATLAB?

---

You'd have to ask MathWorks about that - it's nothing to do with ST:

https://uk.mathworks.com/support/contact_us.html

[Andrew Neil]

---

@DJ_IND wrote:

Link for n series

https://www.st.com/en/microcontrollers-microprocessors/stm32n657x0.html

---

So look at the corresponding page for whatever H series chip you're currently using; compare & contrast...

The key features of the N6 are the Neural-ART accelerator & NeoChrom GPU - aimed at AI applications.

Is that what you're doing?

 

---

@DJ_IND wrote:

 Currently we are using H series

---

Which one(s), exactly?

There are both H5 and H7:

https://www.st.com/en/microcontrollers-microprocessors/stm32h5-series.html

https://www.st.com/en/microcontrollers-microprocessors/stm32h7-series.html

Some of the H7s are dual-core ...

[TDK]

> Apart from external flash is there any major difference where N series is lagging with respect to H series part??

They're entirely different chips with many differences. Neither one is "lagging". They are built for different purposes.

 

If computational speed is the issue, you're not getting much of an upgrade here. Probably more efficient coding is the better solution. Utilize cache, ITCM ram, put vector table and often executed code in ram, etc.

 

[Tesla DeLorean]

The speed here is a bit deceptive. The dual core isn't 600 MHz, it's only 480 MHz. And the H7 is a mixed bag of configurations.

STM32H743 1027 DMIPS (480 MHz)

STM32H745 1027+300 DMIPS (480 MHz)

STM32H723 1177 DMIPS (550 MHz)

STM32H7R3 1284 DMIPS (600 MHz)

 

As I recall the H6 is in a different process technology, and is predominantly RAM

STM32H6 1352 DMIPS (800 MHz)

https://www.st.com/en/microcontrollers-microprocessors/stm32h7-series.html

https://www.st.com/en/microcontrollers-microprocessors/stm32n6-series.html

 

CM4  1.25 DMIPS/MHz (1.27 w/FPU)

CM7  2.14 DMIPS/MHz 

CM55 1.69 DMIPS/MHz

https://en.wikipedia.org/wiki/List_of_ARM_processors

 

Would tend to think we'd need to benchmark our own niches, and see what algorithmic improvements one might make with more RAM, faster FPU or NPU accelerators, etc.

[Andrew Neil]

---

@Tesla DeLorean wrote:

The speed here is a bit deceptive. .

---

Indeed:

As with so many of these "up to..." claims, they often don't apply together!

Hence why I asked @DJ_IND earlier to be specific about what "H series" part(s) is/are currently being used.

[DJ_IND]

Redoing the entire coding is not the issue here. Just wanted to check can we replace the H series with N series. Are we upgrading or degrading on performance and feature point of view

[TDK]

> Are we upgrading or degrading on performance and feature point of view

Both.

[ERROR]

---

@DJ_IND wrote:

Currently we are using H series from ST but we are facing some issue in CPU utilization and speed.

---

What is the most CPU consuming?

STM32N6's Cortex-M55 core has the MVE extension (Helium). So if there are lots of vectored operations in your math, it makes sense to give it a try. You can leverage ARM FVP models for that and compare first.

[Bassett.David]

In addition to comments by others, note that ST chose to omit DP_FPU on the N6.  If you're using double precision floating point on your H7, you should probably stick with it.  Let's hope that when "someday" they release an M85 part, they include the fully-functional FPU.

Dave

[Bassett.David]

Hello RomaineR.,

Thank you for your response , and it turns out, apparent correction on the presence of a DP_FPU in the STM32N6xxx!  My concerns on this matter date back to January and ambiguities in the earlier documentation.  Please see:

 https://community.st.com/t5/stm32-mcus-products/stm32n6-datasheet-feature-summary-inaccurate/td-p/763643

There was never a clarification offered, so I was suffering under the notion that ST had removed DP-FPU capability from the series.  It's omission form the early STM32F7 left an impression on me...

I must admit that I'm very glad to hear I was misinformed and that the STM32N6xxx is now a part of interest to me.

Thank you again.

Regards,

Dave

[RomainR.]

Hi @Bassett.David,

It was indeed some typos in the earlier versions of device summary in the datasheet DS14791 and It was fixed since Rev 3.0.

It's also same for HRTimer. It's not present in STM32N6. 

Sorry for the inconvenience.

Best regards

Romain,