Linux-SWAT
Forum Addict!
- Joined
- Feb 13, 2010
- Messages
- 9,187
Vivante Mobile GPU Architecture Gains Traction.
- Thread starter Linux-SWAT
- Start date
Exophase
Nothing good will ever come of Exophase.
This is a delayed reaction at work. Vivante has a stronger market share in 2013 because of a decent lineup in 2012, and hitting segments that don't use cutting edge stuff. In 2013 their lineup isn't very good, let's look at their status with the SoC manufacturers they hit:
Freescale: Known to have shed a ton of its i.MX engineers, are not releasing a Cortex-A15 product, and given their time to market with i.MX6 it could be a long time before they release something else. And I have a feeling it'll be with heavy focus on automotive and not mobile (like TI, but moreso).
Ingenic: Switched to PowerVR, most likely as part of an attractive package support deal now that IMG owns MIPS.
Rockchip: Uses Mali in everything in their 3xxx chips, most likely as part of an attractive package support deal with ARM (also, is Vivante going to have good hard macros for a process like GF 28nm? ARM will)
Action Semiconductor: Who is using these low end SoCs exactly? Even in China...
Godson-2H: This really old 65nm SoC with an awful CPU, can you even buy it in anything at all..
That basically leaves one big player, Marvell. For a long time I've known Marvell to get most of its share in things like networking devices and harddrive controllers, where there's no GPU. They've flopped a few times trying to put out high end mobile SoCs, like they one they advertised with an asynchronous tri-core setup that never materialized. They ended up slapping Cortex-A9s on SoCs for mobile instead of their own in-house designs. They got some wins like Chromecast and Galaxy Tab 3 7.0 with some pretty low end trailing edge process (40/45nm) stuff, but I have to guess they did it because they're selling it for very cheap. I don't know what
Now Vivante is showing up in the press, SemiAccurate in particular has been gushing in such a strong way that it reeks of being paid off. But for years I've heard nothing about Vivante succeeding the uarch used in i.MX6 with something newer and better. So I don't know where all this hype is coming from. I'll be really amazed if their market share continues to go up in 2014. instead of down...
Freescale: Known to have shed a ton of its i.MX engineers, are not releasing a Cortex-A15 product, and given their time to market with i.MX6 it could be a long time before they release something else. And I have a feeling it'll be with heavy focus on automotive and not mobile (like TI, but moreso).
Ingenic: Switched to PowerVR, most likely as part of an attractive package support deal now that IMG owns MIPS.
Rockchip: Uses Mali in everything in their 3xxx chips, most likely as part of an attractive package support deal with ARM (also, is Vivante going to have good hard macros for a process like GF 28nm? ARM will)
Action Semiconductor: Who is using these low end SoCs exactly? Even in China...
Godson-2H: This really old 65nm SoC with an awful CPU, can you even buy it in anything at all..
That basically leaves one big player, Marvell. For a long time I've known Marvell to get most of its share in things like networking devices and harddrive controllers, where there's no GPU. They've flopped a few times trying to put out high end mobile SoCs, like they one they advertised with an asynchronous tri-core setup that never materialized. They ended up slapping Cortex-A9s on SoCs for mobile instead of their own in-house designs. They got some wins like Chromecast and Galaxy Tab 3 7.0 with some pretty low end trailing edge process (40/45nm) stuff, but I have to guess they did it because they're selling it for very cheap. I don't know what
Now Vivante is showing up in the press, SemiAccurate in particular has been gushing in such a strong way that it reeks of being paid off. But for years I've heard nothing about Vivante succeeding the uarch used in i.MX6 with something newer and better. So I don't know where all this hype is coming from. I'll be really amazed if their market share continues to go up in 2014. instead of down...
bsp
Very Active Member
- Joined
- Dec 2, 2006
- Messages
- 314
It's no secret that they have a lot of automotive customers for their i.MX6. That does not make the platform less suitable for other applications, though (there's a lot of demand for GPUs/3D in automotive nowadays).
Freescale recently announced their (28 nm) i.MX7 and 8 platforms: https://community.freescale.com/community/the-embedded-beat/blog/2013/09/04/discover-the-next-generation-of-imx-applications-processors
(well, actually they announced that the product line is not canned after all and that they will reveal details next year)
Last edited by a moderator:
Exophase
Nothing good will ever come of Exophase.
What makes i.MX6 less suitable for other applications is it's old compared to its competitors. It's on a dated process and these days it's a poor choice even compared to cheaper Chinese SoCs.
They're not even pretending that i.MX7 and i.MX8 are meant for mobile. i.MX7 clearly looks like it's meant for devices that don't need GPU power, while i.MX8 is pushed almost entirely at automotive media. The players who couldn't make it in mobile (Freescale and TI) are all falling back on pushing automotive entertainment systems hard. nVidia is also claiming they've had huge growth in this area. Soon this segment will become very crowded too. Frankly I'm amazed that there's such a demand for GPU power here, no one's playing high end games on their dashboards and there should be a limit to how nice they want their navigation systems to look.
At any rate, we don't even know if i.MX7/8 will have Vivante GPUs. And with even specs not announced until next year I'd be amazed if i.MX8 makes it into anything before 2015. i.MX6 was incredibly late to show up in any end consumer products, when it finally did it was only in low cost Chinese tablets and HDMI sticks. I'm sure Freescale got it there by selling it for a lot less than they intended to. In the long term I really doubt they can compete in a low-price war against vendors like Allwinner.
They're not even pretending that i.MX7 and i.MX8 are meant for mobile. i.MX7 clearly looks like it's meant for devices that don't need GPU power, while i.MX8 is pushed almost entirely at automotive media. The players who couldn't make it in mobile (Freescale and TI) are all falling back on pushing automotive entertainment systems hard. nVidia is also claiming they've had huge growth in this area. Soon this segment will become very crowded too. Frankly I'm amazed that there's such a demand for GPU power here, no one's playing high end games on their dashboards and there should be a limit to how nice they want their navigation systems to look.
At any rate, we don't even know if i.MX7/8 will have Vivante GPUs. And with even specs not announced until next year I'd be amazed if i.MX8 makes it into anything before 2015. i.MX6 was incredibly late to show up in any end consumer products, when it finally did it was only in low cost Chinese tablets and HDMI sticks. I'm sure Freescale got it there by selling it for a lot less than they intended to. In the long term I really doubt they can compete in a low-price war against vendors like Allwinner.
Last edited by a moderator:
bsp
Very Active Member
- Joined
- Dec 2, 2006
- Messages
- 314
I just learnt about their roadmap today so I would vote for an i.MX8 as the platform for a Pandora successor. Since there are no details available yet, we can only speculate whether it will be suitable for that purpose (my guess is that it will).
The i.MX7 is definitely _not_ a candidate, that's for sure.
Does all that make sense from an energy-efficient point of view ? Probably not. But well, it does look fancy
It will probably have a positive impact on the quality of the drivers since the automotive industry is usually quite demanding in that regard.
The i.MX7 is definitely _not_ a candidate, that's for sure.
They are replacing traditional dashboards (analogue tacho displays and so on) with 3D rendered graphics / displays. The main HMI uses shader-heavy graphics to display fancy representations of the car and for map rendering, GPUs cannot be fast enough.
Does all that make sense from an energy-efficient point of view ? Probably not. But well, it does look fancy
It will probably have a positive impact on the quality of the drivers since the automotive industry is usually quite demanding in that regard.
Last edited by a moderator:
Exophase
Nothing good will ever come of Exophase.
Without knowing anything about i.MX8 except that it's ARM-v8 you vote for it? Without even having the faintest idea of when it'll be out? ED is looking for hardware that he can start designing around today..
bsp
Very Active Member
- Joined
- Dec 2, 2006
- Messages
- 314
hey, come on. Do I need to quote myself?
But if I were him, I'd probably wait until details for i.MX8 become available next year.
EDIT (2h later):
You said "I'd be amazed if i.MX8 makes it into anything before 2015" and chances are that this turns out to be true. I would not want to wait that long since that would mean that a "P2" would not appear before 2016 or even later. I agree that it would be better to have a Cortex A15 based platform since many of the current Pandora applications would benefit from improved single-core performance. According to this http://www.itproportal.com/2012/10/31/arm-cortex-a57-and-a53-vs-cortex-a8-a9-a15-and-a7-a-performance-analysis/ the DMIPS/Mhz ratio, compared to the A8 in the current Pandora, of an A9 is 2.5 while the A15's is 3.5. On the other hand, the i.MX6 is out there right now, tried-and-true. It has ~8 times the memory bandwidth of the current Pandora, ~10 times the processing power (disregarding the DSP and assuming that multithreading is used properly), and an ~6 times faster GPU. Last but not least, the SATA and HDMI connectivity IMHO opens up a whole new world of possibilities.
EDIT#2 (next day): Did some (ARM-only) memory benchmarks on an i.MX6 and posted the results in the aforementioned thread. If the Vivante GPUs cannot access the memory much faster than that this is going to be a real bottleneck.
That's why I voted for i.MX6 in this http://boards.openpandora.org/index.php/topic/14440-potential-arm-non-x86-soc-alternatives/ thread.
But if I were him, I'd probably wait until details for i.MX8 become available next year.
EDIT (2h later):
You said "I'd be amazed if i.MX8 makes it into anything before 2015" and chances are that this turns out to be true. I would not want to wait that long since that would mean that a "P2" would not appear before 2016 or even later. I agree that it would be better to have a Cortex A15 based platform since many of the current Pandora applications would benefit from improved single-core performance. According to this http://www.itproportal.com/2012/10/31/arm-cortex-a57-and-a53-vs-cortex-a8-a9-a15-and-a7-a-performance-analysis/ the DMIPS/Mhz ratio, compared to the A8 in the current Pandora, of an A9 is 2.5 while the A15's is 3.5. On the other hand, the i.MX6 is out there right now, tried-and-true. It has ~8 times the memory bandwidth of the current Pandora, ~10 times the processing power (disregarding the DSP and assuming that multithreading is used properly), and an ~6 times faster GPU. Last but not least, the SATA and HDMI connectivity IMHO opens up a whole new world of possibilities.
EDIT#2 (next day): Did some (ARM-only) memory benchmarks on an i.MX6 and posted the results in the aforementioned thread. If the Vivante GPUs cannot access the memory much faster than that this is going to be a real bottleneck.
Last edited by a moderator:
Similar threads
