User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Why didn't we go with Mali?

Wed Sep 09, 2015 10:46 pm

i know it is a bit late to ask this, but with all the fuss about the GPU driver being still in development why didn't we go with Mail in the first place?
AFAIK Mali has an opensource driver, limadriver, and some source code is provided by ARM as well. Could this be an considerable option for future products since Broadcom has stopped any/all hardware development on its videocore gpu?
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: [email protected]
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

fruitoftheloom
Posts: 21314
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 7:31 am

ric96 wrote:i know it is a bit late to ask this, but with all the fuss about the GPU driver being still in development why didn't we go with Mail in the first place?
AFAIK Mali has an opensource driver, limadriver, and some source code is provided by ARM as well. Could this be an considerable option for future products since Broadcom has stopped any/all hardware development on its videocore gpu?
EU worked at Broadcom :?:
Retired disgracefully.....
This at present is my daily "computer" https://www.asus.com/us/Chrome-Devices/Chromebit-CS10/

zeth
Posts: 27
Joined: Thu Sep 10, 2015 7:37 am

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 7:46 am

Interesting ideas.
ric96 wrote:i know it is a bit late to ask this, but with all the fuss about the GPU driver being still in development why didn't we go with Mail in the first place?
As far as I know (which is not far) it is historical reasons of people in/associated with Broadcom coming up with the idea of the Raspberry Pi.
ric96 wrote:AFAIK Mali has an opensource driver, limadriver, and some source code is provided by ARM as well.
Well Lima was reverse engineered and ARM are reportedly pretty hostile towards open source graphics driver development to the developer in particular.
ric96 wrote:Could this be an considerable option for future products since Broadcom has stopped any/all hardware development on its videocore gpu?
Do we know that?

Anyway, the Videocore is really powerful compared to everything else in the device. The biggest issue for me with Raspberry Pi 1 was lack of processing power which was more than fixed with the Pi 2, the next bottleneck is I/O, especially Ethenet I/O, I personally don't see any need for any more graphics power for a long time.

I have a cubietruck which is Mali based but haven't really touched it since the RPi 2 came out, since the ecosystem around the Pi is the really interesting thing to me. So I doubt they can ever change the SoC to a completely different platform without messing the ecosystem up.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24556
Joined: Sat Jul 30, 2011 7:41 pm

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 8:10 am

At the time of development (8 or more years ago), the Videocore 3 then 4, they were more powerful than the Mali. In addition, the videocore includes a camera ISP, LCD drivers etc, which the Mali does not. Also, the VC uses less power for the same performance, which was important for the mobile space it was aimed at.

Of course, Mali has moved on , the VC5 is no longer under development AFAIK, so now the VC4 looks a bit underpowered to more recent designs, but tbh, it's still good enough for most people.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I own the world’s worst thesaurus. Not only is it awful, it’s awful."

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 9:19 am

jamesh wrote:... In addition, the videocore includes a camera ISP, LCD drivers etc, which the Mali does not. Also, the VC uses less power for the same performance...
More than satisfactory reason for me...
Thnx for the response.
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: [email protected]
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

hippy
Posts: 6496
Joined: Fri Sep 09, 2011 10:34 pm
Location: UK

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 9:40 am

zeth wrote:Anyway, the Videocore is really powerful compared to everything else in the device ... I personally don't see any need for any more graphics power for a long time.
It is not so much that it's not powerful enough, but that its closed nature restricts the pursuance of some projects which require having accelerated graphics and similar. Had it been something else that might not be so problematic. But it is what it is and what we have.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24556
Joined: Sat Jul 30, 2011 7:41 pm

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 10:28 am

hippy wrote:
zeth wrote:Anyway, the Videocore is really powerful compared to everything else in the device ... I personally don't see any need for any more graphics power for a long time.
It is not so much that it's not powerful enough, but that its closed nature restricts the pursuance of some projects which require having accelerated graphics and similar. Had it been something else that might not be so problematic. But it is what it is and what we have.
I'm not convinced by that. There completely standard OpenGL and OpenVG libraries are fully compliant, so you do have access to all the acceleration available.

And of course, the Eric Anholt stuff will be an fully OSS graphics driver, meaning it will all be available via that as well.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I own the world’s worst thesaurus. Not only is it awful, it’s awful."

asavah
Posts: 366
Joined: Thu Aug 14, 2014 12:49 am

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:01 pm

Right now VC4 still is the best option.
Mali is closed source.
VC4 hw video decoder is awesome, and it's arm side libraries are open source.
Mali usually comes with cedarx, at least in allwinner SoCs which is a big pain in the butt for linux users because it's another android-oriented proprietary blob.

I've looked at a lot of boards as a possible more powerful replacement for the rpi, but seeing mali + cedarx makes me cry.
So far IMO rpi2 is the best option for small linux-powered boards.

I only wish RPF makes a rpi with a proper (not usb crap) NIC in the next 100 years :mrgreen:

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:36 pm

asavah wrote:...
I only wish RPF makes a rpi with a proper (not usb crap) NIC in the next 100 years :mrgreen:
Could be possible sooner than you imagine, lots of firms are looking firward to arm based (mostly storage) server solution and any such solution would require good networking.
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: [email protected]
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:45 pm

I am really enjoying this thread. Thanks for your good posts, people. Very informative. The more I learn about the Pi (and Pi2) and the design desicions made in order to realise it, the more I like it. :-)
Are there any news on the Wayland front, by the way, which is – I think – related to:
jamesh wrote:There completely standard OpenGL and OpenVG libraries are fully compliant, so you do have access to all the acceleration available.

And of course, the Eric Anholt stuff will be an fully OSS graphics driver, meaning it will all be available via that as well.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24556
Joined: Sat Jul 30, 2011 7:41 pm

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:48 pm

Fidelius wrote:I am really enjoying this thread. Thanks for your good posts, people. Very informative. The more I learn about the Pi (and Pi2) and the design desicions made in order to realise it, the more I like it. :-)
Are there any news on the Wayland front, by the way, which is – I think – related to:
jamesh wrote:There completely standard OpenGL and OpenVG libraries are fully compliant, so you do have access to all the acceleration available.

And of course, the Eric Anholt stuff will be an fully OSS graphics driver, meaning it will all be available via that as well.
Listen to the Raspberry Pi guys recent interview with Eben - this (and lots of other stuff) is covered.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I own the world’s worst thesaurus. Not only is it awful, it’s awful."

fruitoftheloom
Posts: 21314
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:50 pm

Retired disgracefully.....
This at present is my daily "computer" https://www.asus.com/us/Chrome-Devices/Chromebit-CS10/

User avatar
Fidelius
Posts: 460
Joined: Wed Jan 01, 2014 8:40 pm
Location: Germany

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 12:57 pm

Oh great, thanks for the link.

hippy
Posts: 6496
Joined: Fri Sep 09, 2011 10:34 pm
Location: UK

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 2:00 pm

jamesh wrote:
hippy wrote:It is not so much that it's not powerful enough, but that its closed nature restricts the pursuance of some projects which require having accelerated graphics and similar. Had it been something else that might not be so problematic. But it is what it is and what we have.
I'm not convinced by that. There completely standard OpenGL and OpenVG libraries are fully compliant, so you do have access to all the acceleration available.

And of course, the Eric Anholt stuff will be an fully OSS graphics driver, meaning it will all be available via that as well.
Not being able to implement accelerated graphics without Foundation/Broadcom assistance for a particular alternative OS seems to be the commonly cited reason for that effort not going anyway fast and never likely to be sensibly usable without it, with the suggestion that if it were more open it could better proceed.

I admit I haven't followed in detail and it may be possible without assistance and it's just that the people working on that project don't have the skills to do it by themselves but I got the impression they would need more than just the skills.

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 2:20 pm

jamesh wrote:At the time of development (8 or more years ago), the Videocore 3 then 4, they were more powerful than the Mali. In addition, the videocore includes a camera ISP, LCD drivers etc, which the Mali does not. Also, the VC uses less power for the same performance, which was important for the mobile space it was aimed at.

Of course, Mali has moved on , the VC5 is no longer under development AFAIK, so now the VC4 looks a bit underpowered to more recent designs, but tbh, it's still good enough for most people.
Would it be possible to create a VC4+ by tacking on more QPUs or would that require massive changes/design work? Would be a shame to throw away all those open source drivers if a future SoC no longer has a VC4 unit due to a lack of GPU power.
Also, how much of the Vulcan API could be covered by the current VideoCore assuming drivers would be developed for it?

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24556
Joined: Sat Jul 30, 2011 7:41 pm

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 3:28 pm

bstrobl wrote:
jamesh wrote:At the time of development (8 or more years ago), the Videocore 3 then 4, they were more powerful than the Mali. In addition, the videocore includes a camera ISP, LCD drivers etc, which the Mali does not. Also, the VC uses less power for the same performance, which was important for the mobile space it was aimed at.

Of course, Mali has moved on , the VC5 is no longer under development AFAIK, so now the VC4 looks a bit underpowered to more recent designs, but tbh, it's still good enough for most people.
Would it be possible to create a VC4+ by tacking on more QPUs or would that require massive changes/design work? Would be a shame to throw away all those open source drivers if a future SoC no longer has a VC4 unit due to a lack of GPU power.
Also, how much of the Vulcan API could be covered by the current VideoCore assuming drivers would be developed for it?
That would cost a couple of million dollars. If there was anyone left at Broadcom with the knowhow. Not saying it won;t be done - just giving an indication of the scale of the work. I know nothing of Vulcan.

Note, that adding QPU's doesn't necessarily increase the overall performance. For example, ISTR that using the die space currently used for QPU's for more cache would have a better overall effect.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I own the world’s worst thesaurus. Not only is it awful, it’s awful."

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 3:50 pm

jamesh wrote:
That would cost a couple of million dollars. If there was anyone left at Broadcom with the knowhow. Not saying it won;t be done - just giving an indication of the scale of the work. I know nothing of Vulcan.

Note, that adding QPU's doesn't necessarily increase the overall performance. For example, ISTR that using the die space currently used for QPU's for more cache would have a better overall effect.
Thanks for the info. So its safe to assume its a bit of a tradeoff in keeping the QPUs fed with data and the amount of QPUs, sort of like the additional edram used by some of the more expensive Intel HD graphics which help more than simply adding additional execution units.

If thats the case, would a future VC4 with more cache be cheap to design or is said cache heavily hardwired into the VideoCore? I assume its simply a matter of adding more SRAM and making the cache logic aware of it (not accounting for design verification and mask production of course), but I could be very wrong on that one.

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 5:59 pm

Maybe Broadcom's parent company sees some future in video core if the OSS drivers work out good and decide to revive the project. But then again the is a VERY far fetched thought.
*dreams of a future pi 3 running intel coreM*
*flies away*
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: [email protected]
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

zeth
Posts: 27
Joined: Thu Sep 10, 2015 7:37 am

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 7:51 pm

Obviously with the normal caveats that is supposed to be a device for children (though of course we are all children - Gal 3:26) so other uses are a side effect rather than the aim, and the Raspberry Pi 2 is the Raspberry Pi for the next few years, so this is all just speculation, etc etc.
ric96 wrote:Maybe Broadcom's parent company sees some future in video core if the OSS drivers work out good and decide to revive the project. But then again the is a VERY far fetched thought.
I know nothing about Broadcom, but I think you are right. Broadcom is a company which is supposed to maximise profit for its shareholders (although after a company gets beyond a certain scale they tend to not care about their shareholders anymore). If the Raspberry Pi continues its growth and becomes the go-to platform for embedded development and Broadcom are selling tens of millions of SoCs then it will be in Broadcom's interest to keep the party going; even if it has to get the band back together. At least there are lots of smart people out there who know the architecture because everyone has a Pi.

The ARM architecture is such a mess that if the Broadcom chip in the Pi gets complete mainline kernel support then its long term maintainability in the kernel becomes a killer feature over other embedded ARM chips where you are dependent on a driver team in China or whatever still existing in the future and throwing blobs over the wall.

Gigabit Ethernet or some other way of getting data quickly on and off the Pi would allow it to enter networking and parallel computing where Linux already has a strong software stack. For what its worth (nothing), I would rather any future RPI 2+ in 2020 to have this, graphics performance is already fine (for me).

The system requirements of Linux desktops haven't really been increased very much over time, a decent 2007 laptop will run all distributions. In some cases, performance has gone up quite a bit. I don't see why the RPi 2's current SoC is not still as useful in 2020 and beyond.

Of course in 5 years anything could happen, one of these OpenRISC type projects could finally work and be put on the Rpi 3, who knows.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 24556
Joined: Sat Jul 30, 2011 7:41 pm

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 8:05 pm

bstrobl wrote:
jamesh wrote:
That would cost a couple of million dollars. If there was anyone left at Broadcom with the knowhow. Not saying it won;t be done - just giving an indication of the scale of the work. I know nothing of Vulcan.

Note, that adding QPU's doesn't necessarily increase the overall performance. For example, ISTR that using the die space currently used for QPU's for more cache would have a better overall effect.
Thanks for the info. So its safe to assume its a bit of a tradeoff in keeping the QPUs fed with data and the amount of QPUs, sort of like the additional edram used by some of the more expensive Intel HD graphics which help more than simply adding additional execution units.

If thats the case, would a future VC4 with more cache be cheap to design or is said cache heavily hardwired into the VideoCore? I assume its simply a matter of adding more SRAM and making the cache logic aware of it (not accounting for design verification and mask production of course), but I could be very wrong on that one.
The 2836 already changed the caching architecture to improve VC4 and ARM performance - the GPU got its own cache, and the ARMs got their own, instead of being shared.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I own the world’s worst thesaurus. Not only is it awful, it’s awful."

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 10:10 pm

zeth wrote:For what its worth (nothing), I would rather any future RPI 2+ in 2020 to have this, graphics performance is already fine (for me).
I won't rest till I have my Raspi Model C with 2 VC4s frankensteined together for twice the performance :lol:

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Why didn't we go with Mali?

Thu Sep 10, 2015 10:13 pm

bstrobl wrote:
zeth wrote:For what its worth (nothing), I would rather any future RPI 2+ in 2020 to have this, graphics performance is already fine (for me).
I won't rest till I have my Raspi Model C with 2 VC4s frankensteined together for twice the performance :lol:
Why don't we just add a R9 295 x2 :D
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: [email protected]
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

mic_s
Posts: 92
Joined: Sun Oct 26, 2014 4:15 pm

Re: Why didn't we go with Mali?

Fri Sep 11, 2015 12:03 am

Gigabit Ethernet or some other way of getting data quickly on and off the Pi would all...
There is already a fast interface.
Since the first Raspberry Pi.
Multi-Gbit.

ok - kind of multi-Gbit : CSI and DSI

michael
.

zeth
Posts: 27
Joined: Thu Sep 10, 2015 7:37 am

Re: Why didn't we go with Mali?

Fri Sep 11, 2015 1:33 am

mic_s wrote:
Gigabit Ethernet or some other way of getting data quickly on and off the Pi would all...
There is already a fast interface.
Since the first Raspberry Pi.
Multi-Gbit.

ok - kind of multi-Gbit : CSI and DSI

michael
.
If we imagine we have a magic device that can use the CSI connector, then streaming arbitrary data to the Pi through the CSI port would be epic.

zeth
Posts: 27
Joined: Thu Sep 10, 2015 7:37 am

Re: Why didn't we go with Mali?

Fri Sep 11, 2015 1:42 am

bstrobl wrote:Would it be possible to create a VC4+ by tacking on more QPUs or would that require massive changes/design work?
jamesh wrote:That would cost a couple of million dollars.
Seems like a bargain, amortised over several years and a couple of million Pis, it is a dollar per RPi 3.

Return to “General discussion”