Author Topic: STM32G series ?  (Read 12646 times)

0 Members and 2 Guests are viewing this topic.

Offline Yansi

  • Super Contributor
  • ***
  • Posts: 3278
  • Country: 00
  • STM32, STM8, AVR, 8051
Re: STM32G series ?
« Reply #25 on: March 28, 2019, 10:32:49 pm »
L5 is some lowpower crypto something (not even close to F4), G3 do not exist however G4 are an extension of F3.
F7 may be considered as an update of F4.
 

Offline Asantos

  • Newbie
  • Posts: 2
  • Country: br
Re: STM32G series ?
« Reply #26 on: March 29, 2019, 11:37:18 am »
You can find STM32G474CET6 at Avnet parts source.
There are STM32G4 references in some ST application notes and firmware headers files.
I believe stm32g4 will be released soon.
 

Offline matherp

  • Contributor
  • Posts: 12
Re: STM32G series ?
« Reply #27 on: April 04, 2019, 12:56:20 pm »
Look in C:\Program Files\STMicroelectronics\STM32Cube\STM32CubeProgrammer\Data_Base\STM32_Prog_DB.xml

might give some clues  ;D
 

Offline nano

  • Contributor
  • Posts: 29
  • Country: de
Re: STM32G series ?
« Reply #28 on: April 17, 2019, 08:31:38 pm »
This appnote already contains some information about the STM32G4 series.
 

Offline dave j

  • Regular Contributor
  • *
  • Posts: 54
  • Country: gb
Re: STM32G series ?
« Reply #29 on: April 26, 2019, 07:17:23 pm »
It looks like it's getting close to an official announcement of the G4 chips. There's a new version of STM32CubeMX (version 5.2.0) that supports them and they've released the STM32Cube stuff onto github.

Interesting stuff with the ADCs - although I haven't found anything about the speed yet.
The OPAMPs have PGA up to 64x so that implies greater gain bandwidth.
I'm not David L Jones. Apparently I actually do have to point this out.
 

Offline Yansi

  • Super Contributor
  • ***
  • Posts: 3278
  • Country: 00
  • STM32, STM8, AVR, 8051
Re: STM32G series ?
« Reply #30 on: April 28, 2019, 09:51:25 am »
Yup shall be within a week.
 

Offline Blue_Alien

  • Contributor
  • Posts: 18
  • Country: us
Re: STM32G series ?
« Reply #31 on: May 10, 2019, 11:13:26 pm »
I am wrapping up a project based on the G0 series. I original had it designed with an F0 and a discrete USB PD Phy, but right as I was getting ready to build boards, the G0 came out. So I switched and was able to shave some cost on the BOM.

https://hackaday.io/project/161771-lipow-the-usb-c-lipo-battery-charger
https://github.com/AlexKlimaj/LiPow-Firmware

Overall I really like the G0. Mostly just using it as an F0 with built in USB PD.
 

Offline Harjit

  • Regular Contributor
  • *
  • Posts: 90
  • Country: us
Re: STM32G series ?
« Reply #32 on: May 13, 2019, 08:10:08 pm »
Yup shall be within a week.

Been a couple of weeks and no new info as yet.
 

Offline Yansi

  • Super Contributor
  • ***
  • Posts: 3278
  • Country: 00
  • STM32, STM8, AVR, 8051
Re: STM32G series ?
« Reply #33 on: May 13, 2019, 08:21:20 pm »
Well.. delays. Usual ST problems...

Download the new CubeMx 2.1.0, should have the G4 inside now, so you can take a peek what is there.

They should really have had publish G4 already, I'll ask on Friday, what happened. They are way behind schedule.
 


Offline splin

  • Frequent Contributor
  • **
  • Posts: 994
  • Country: gb
Re: STM32G series ?
« Reply #35 on: June 04, 2019, 03:55:53 am »
Odd that there's no performance data for the new FMAC feature. There's lots of information in the reference manual about how to use it but nothing at all (that I can see) as to how it performs. Given it only supports 16 bit DSP I'd expect it to be much faster than software DSP but the lack of any such highlights in the G4's marketing makes me think the gains are somewhat limited.  Perhaps it's main raison d'etre is that it can operate in parallel to the CPU?
 

Offline MT

  • Super Contributor
  • ***
  • Posts: 1290
  • Country: cn
Re: STM32G series ?
« Reply #36 on: July 10, 2019, 04:33:27 pm »
STM32G4 OLT - 30 . Peripheral Filter Math Accelerator
 

Offline andersm

  • Super Contributor
  • ***
  • Posts: 1148
  • Country: fi
Re: STM32G series ?
« Reply #37 on: July 10, 2019, 09:54:08 pm »
I honestly can't tell if that video is narrated by a human or robot.

Online exe

  • Supporter
  • ****
  • Posts: 1636
  • Country: nl
  • self-educated hobbyist
Re: STM32G series ?
« Reply #38 on: July 10, 2019, 10:02:45 pm »
Watched only first 15 seconds. Sounds like a robot, but a really good one. Still don't want to watch the video.
 

Offline westfw

  • Super Contributor
  • ***
  • Posts: 3193
  • Country: us
Re: STM32G series ?
« Reply #39 on: July 10, 2019, 11:21:59 pm »
Yeah; I really don't understand using Text-to-speech programs to produce videos like this.I mean, the voice quality is really impressive, and it's doing a much-better-than-average job of speechifying the technical terms, but the value-add of voice is mostly in cues OTHER than you'd get by just reading text.   In this case, I would have rather just had the text that they're working from.  (like, it's missing the pauses and/or different intonation that a technical presenter would normally use when mentioning a variable name, which is weird to hear and surprisingly difficult to follow.)
 

Offline MT

  • Super Contributor
  • ***
  • Posts: 1290
  • Country: cn
Re: STM32G series ?
« Reply #40 on: July 11, 2019, 03:32:22 pm »
Before i posted that video link i listened through to the end and thought, well that was sorta creepy, but at the end the AI robbut declared FMA unit was 15% faster then the DSP lib equivalent. Robbut should have said that in the beginning!  :phew:
« Last Edit: July 13, 2019, 01:17:47 pm by MT »
 

Offline capt bullshot

  • Super Contributor
  • ***
  • Posts: 1841
  • Country: de
    • Mostly useless stuff, but nice to have: wunderkis.de
Re: STM32G series ?
« Reply #41 on: July 11, 2019, 04:01:17 pm »
15% faster and only 24+2 bit accumulator, meh.
I've done DSP programming on real DSPs, having a real MAC unit, so this thing look quite poor to me, as the CORDIC does (I've skimmed the ref manual a few days ago). Additionally, one has to setup the RAM (transfer coefficients and data into the FMAC RAM), and its registers before it starts, even if this can be done by DMA it appears rather inefficient to me.
Safety devices hinder evolution
 

Offline aheid

  • Regular Contributor
  • *
  • Posts: 166
  • Country: no
Re: STM32G series ?
« Reply #42 on: July 11, 2019, 09:33:48 pm »
As a non-expert, the FMAC seems a bit weird. Why only 26bit accumulator? Why is the scaling exponent R restricted to non-negative values?
 

Offline splin

  • Frequent Contributor
  • **
  • Posts: 994
  • Country: gb
Re: STM32G series ?
« Reply #43 on: July 12, 2019, 01:21:21 pm »
15% faster and only 24+2 bit accumulator, meh.
I've done DSP programming on real DSPs, having a real MAC unit, so this thing look quite poor to me,

Actually the FMAC is *slower* then than CMSIS - see 16:37. The FMAC does a multiply accumulate in 1 cycle but it take 2 cycles to read the data and coefficients from the single port RAM.

@ 17:35 51 tap FIR: arm_fitr_fast_q15()  CMSIS 92 cycles, FMAC 104 cycles.

*BUT* the FMAC operates in parallel to the main CPU (and yes, DMA can be used to provide the data) so whilst it's not the most powerful DSP around it's still much better than none. It's all about compromise between die area, power consumption and market needs. I'd guess the target market for a full blown DSP aren't likely to be interested in a low cost, mid range Cortex M4.

I wish they'd included some support for FFT operations though - not necessarily a full blown FFT processor but something to speed up the M4 itself perhaps with a few extra instructions.
 
The following users thanked this post: thm_w, Kjelt

Offline splin

  • Frequent Contributor
  • **
  • Posts: 994
  • Country: gb
Re: STM32G series ?
« Reply #44 on: July 12, 2019, 03:27:19 pm »
You've also got to wonder if there's much point to the more expensive versions of these when the 480MHz  STM32H750 @ $3.37 will blow away the 170MHz G4 performance wise (even with the FMAC)? EG. the $2.96 STM32G441VB is a value line product with 32K RAM compared to 1MB on the H7 which also has way more peripherals too.

The G4 has CORDIC which may help in some applications, but 1MB RAM allows some large tables of pre-computed values. The G4 has two more DACs however.

The G4 uses less power - 25.5mA at 170MHz (periph off) v H7, 34.5mA @ 200MHz (the H7 @ 200MHz should match G4 + FMAC at q15 FIR). Most applications aren't going to be 100% DSP so other code will be faster on the H7 so overall they might not be far off. OTOH the G4 may use less power for the FMAC alone with the main core sleeping.

The smallest H7 is 100 pins so clearly the smaller G4s have their place but the upper end looks to be seriously squeezed by the H7.
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 5318
  • Country: fr
Re: STM32G series ?
« Reply #45 on: July 12, 2019, 03:36:22 pm »
I wish they'd included some support for FFT operations though - not necessarily a full blown FFT processor but something to speed up the M4 itself perhaps with a few extra instructions.

That would be interesting, but I'm not sure they would even be allowed to add extensions to the ARM Cortex instruction set unless those come from ARM. Anyone knows about this?

So an FFT co-processor might be the only workable solution.
 

Offline splin

  • Frequent Contributor
  • **
  • Posts: 994
  • Country: gb
Re: STM32G series ?
« Reply #46 on: July 12, 2019, 03:50:41 pm »
You're right, ST almost certianly can't extend the instruction set themselves but Freescale added extensions via perhiperals where you can write data to a specific address and the H/W transforms that data or whatever.

It wouldn't have to be a full blown DSP unit but could be something as simple as a memory addressing unit which knowns about the data layout for the FFT butterflys and returns the next data without the CPU having to compute its address - a lot of processing time is spent in the M4 calculating addresses and loading the data compared to doing the actual MAC operation. A bit reversal perhiperal might be useful as well.
« Last Edit: July 12, 2019, 03:52:48 pm by splin »
 

Offline SiliconWizard

  • Super Contributor
  • ***
  • Posts: 5318
  • Country: fr
Re: STM32G series ?
« Reply #47 on: July 12, 2019, 04:07:45 pm »
Yes, a bit reverse addressing mode would be quite handy.
Actually, just a couple MAC instructions supporting a bit reverse addressing mode would already accelerate FFT significantly and would be rather simple to add...
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf