Freescale announces i.MX6 1-4 core Cortex A9 + undisclosed GPU?

  • Thread starter Deleted member 13524
  • Start date
Don't think they licensed SGX / XT cores wonder if it's Imageon although would Qualcomm license it to third parties?
 
Single-core Mali 400 has 4 "fragment processors" which could be interpreted as "quad shaders" but it's specced to do "only" 30 MTri/s.


I wonder if they meant "MegaTexels".
 
The highest-end vivante GC1000 claims 85MTri/s, and that's with the GPU running at 1.1GHz.
 
Perhaps it's a multicore Vivante.

"Vivante's multicore GPUs are ultra-threaded extensions of the product proven single core GC series architecture first launched in 2007. These product-proven multicore GPUs are capable of delivering well over 200M triangles per second on industry standard GPU benchmark polygon throughput feature tests in currently licensed customer configurations."

http://www.vivantecorp.com/GC-Series.html
 
Perhaps it's a multicore Vivante.

"Vivante's multicore GPUs are ultra-threaded extensions of the product proven single core GC series architecture first launched in 2007. These product-proven multicore GPUs are capable of delivering well over 200M triangles per second on industry standard GPU benchmark polygon throughput feature tests in currently licensed customer configurations."

http://www.vivantecorp.com/GC-Series.html

Yes, although the website only mentions models up to GC1000, the PDF presentation mentions a GC2000 capable of 108MTri/s, and that press release mentions a possible GC4000 which, if it follows the same double-specs of the rest of the family, should be able to do 200MTri/s.
But we're now talking about a ~14mm^2 part here. And GC2000 has DX11 support, which means there's even a programmable tesselation engine there.

Sounds way too souped-up to be paired with dual Cortex A9s, and it's not even mentioned in the press release.

I mean, 200MTri/s is already in the same order of magnitude as Xenos. That's crazy.
 
Here's the Freescale announcement:

http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=1511465&highlight=

For video/3D:


Multistream-capable HD video engine delivering 1080p60 decode, 1080p30 encode and 3D video playback in HD
Exceptional 3D graphics performance with quad shaders for up to 200 MTPS
Separate 2D and vertex acceleration engines for uncompromised user interface experiences
Stereoscopic image sensor support for 3D imaging
Interconnect: HDMI v1.4 w/ integrated PHY, SD3.0, multiple USB 2.0 ports w/ integrated PHY, Gb Ethernet w/ integrated PHY, SATA-II w/ integrated PHY, PCI-e w/ integrated PHY, MIPI CSI, MIPI DSI, MIPI HSI, and FlexCAN for automotive applications
Support for the VP8 codec
Support for one of the broadest ranges of major operating system platforms in the industry
Optional integration of an ePaper display controller for eReader and similar applications

I don't care what kind of GPU it contains, that 200M Tris/s claim sounds like a typical marketing exaggeration.

MTPS is most likely for million triangles per second as for some sad reason it has become another meaningless performance metric in the mobile/embedded space.
 
I don't care what kind of GPU it contains, that 200M Tris/s claim sounds like a typical marketing exaggeration.

MTPS is most likely for million triangles per second as for some sad reason it has become another meaningless performance metric in the mobile/embedded space.

Care to elaborate wich performance metric isn't?

Unless all the manufacturers disclose the complete specs for each device (wich they don't, not even clock speeds), all we have is these few "meaningless" performance metrics and a few unoptimized benchmarks (GLBenchmark, Neocore, etc) to compare.
 

Samsung initially tried to market the S5PC110 with something like over 90M Tris/s. The manual for it mentions that the SGX540 it incorporates is capable of 20M Tris/s.

Likewise the Tegra2 GPU was also somewhere noted as 90M Tris while reality is more in the 70M vertices/s range.

First one is typical marketing bullshit, the 2nd one a perfectly sustainable rate in real time.
 
Yeah, the lack of a standardized measure for "geometry" rates makes any comparison taken out of full context meaningless and could obviously reveal a massive disparity between a marketing number and a manufacturer supplied performance guideline (like with Samsung's SGX540 numbers).

Marvell markets both "quad shaders" and "200M tri/sec" performance for a Vivante GPU in their Armada app processor line, and Vivante has disclosed that they have at least one tier-one partner whose identity is yet to be revealed.
 
Separate 2D and vertex acceleration engines for uncompromised user interface experiences
Sounds like Freescale are sticking to the strategy of having separate cores for OpenGL ES and OpenVG, like they did in i.MX5.
That said; the meager description might also just be marketingspeak for "proprietary blitter + a DSP" :)
 
Sounds like Freescale are sticking to the strategy of having separate cores for OpenGL ES and OpenVG, like they did in i.MX5.
That said; the meager description might also just be marketingspeak for "proprietary blitter + a DSP" :)
Unlikely. What's the point of dropping an already proven openvg part in favor of a proprietary setup? That said, the imx5x's already have a bunch of specialized units many of which have their own dma capabilities, so there will likely be plenty of "proprietary" stuff in there, one way or another.
 
Samsung initially tried to market the S5PC110 with something like over 90M Tris/s. The manual for it mentions that the SGX540 it incorporates is capable of 20M Tris/s.

Likewise the Tegra2 GPU was also somewhere noted as 90M Tris while reality is more in the 70M vertices/s range.

First one is typical marketing bullshit, the 2nd one a perfectly sustainable rate in real time.

You've seen a demonstration of 70MVerts/s on Tegra2?
 
Back
Top