OMAP4 & SGX540

[Exophase]

I'm sure nVidia is a major factor in their focus to up their release cadence. Seems like the OMAP4470 is important to remain competitive against Kal-El until OMAP5 is released. 1.8GHz dual core vs 1.5GHz quad core isn't a very level comparison, but I expect that in early 2012 most smart phone/tablet software won't be making good utilization past two cores so TI should remain okay here. In fact, with nVidia basically driving the quad-core phone SoC market all by themselves they should be concerned about adoption.. SGX544MP1 at ~400MHz sounds like a reasonable shot at being in the same ballpark as Kal-El's GPU.

Let's hope that the time between release and product introduction goes down too, that's the real killer right now.

 

[tangey]

As this is kinda a SGX540 thread, I see ambrella have launched their soc aim at cameras, which contains dual core A9, an ARM11, SGX540, and a couple of dedicated DSPs.

http://www.ambarella.com/uploads/docs/iOne-smart-camera-product-brief.pdf

 

[Ailuros]

I'm sure nVidia is a major factor in their focus to up their release cadence. Seems like the OMAP4470 is important to remain competitive against Kal-El until OMAP5 is released. 1.8GHz dual core vs 1.5GHz quad core isn't a very level comparison, but I expect that in early 2012 most smart phone/tablet software won't be making good utilization past two cores so TI should remain okay here. In fact, with nVidia basically driving the quad-core phone SoC market all by themselves they should be concerned about adoption.. SGX544MP1 at ~400MHz sounds like a reasonable shot at being in ,the same ballpark as Kal-El's GPU.

Excuse the hair splitting but it's either a MP ie multiple cores or simply a single SGX5xx. MP1 sounds to me like someone saying in other words multi/single core; it can be only either/or.

As for 400MHz for the 544 if the 2.5x is an average rate and not some floating point theoretical nonsense let me take the theoretical numbers I know so far:

SGX540 vs. 544 both at 300MHz:

4.8 vs. 10.8 GFLOPs = 2.25x
600 vs. 600MTexels = 0
35M Tris vs. 52.5 MTris/s = 1.5x
2.4 GPixels vs 4.8 GPixels = 2.0x
----------------------------------------------
Average = 1.44x

In order to reach a 2.5x advantage for the 544 with that rather silly math I'd need to increase its frequency by <60%. Give or take 100MHz more than SGX540@4460 even under the same process doesn't surprise me, given that Apple managed to cram a MP2 (which I believe is clocked at) =/>250MHz into 45nm. Why should a single core at roughly twice the frequency be a problem?

With the latest driver the 540@200MHz (Google Nexus S, Samsung SGH-T959 Vibrant) is only a hair (~5%) above in GL Benchmark2.0 Egypt from a Tegra2 smart-phone always at 800*480 for which the ULP GF is in smart-phones clocked at 300MHz.

At the frequency I'm expecting the 4470 to be, I'd be very surprised if performance for the SGX544 wouldn't be give or take on par with A5's MP2. Still sounds a tall order to me, especially for something like the ULP GF in T3 which I estimate 2 Vec4 PS ALUs for.

 

[Exophase]

Yeah, the "MP1" part is kind of an invalid denotation.. I guess it's just there to emphasize that there aren't multiple cores, since without it it could be a little ambiguous.

 

[Rys]

Technically MP1 is possible, where you've got all the MP infrastructure and a single core.

 

[Arun]

Technically MP1 is possible, where you've got all the MP infrastructure and a single core.

I suppose the only reason you'd ever want that on SGX 5XT is redundancy from a MP2 config on a lower-end SKU.

 

[Rys]

It would certainly save you time if you'd already done all the hard work on verification and everything with MP2+ in a different product, and you were OK with some area overhead leaving the MP plumbing in there. There are (IMHO anyway) benefits to running the MP plumbing with just one core, versus a bare single core, so if I was building an SoC myself I'd evaluate MP1 for a few reasons.

 

[Lazy8s]

The MP overhead has benefits by itself? Interesting.

 

[Arun]

Considering TI don't currently even go for the obvious win of using PowerVR's video solutions, I haven't had especially high expectations for Imgtec to be the IP provider of that low power graphics/display-compositor core.

Vivante just announced 5 of the Top 10 application processor vendors use their 2D core: http://www.eetimes.com/electronics-news/4216864/Vivante-licensees-GPU - seems like the most likely contender at this point.
IMG next to Vivante in the same design? Fun stuff!

 

[Lazy8s]

Over at the IMG.L forum at iii.co.uk, I noticed member primitive2 linked a post on TI's blog.

TI gives a bit more detail how they're comparing OMAP4470's graphics performance including a competitive performance claim that once again implies a very high clock for the GPU.

Graphics measurement involves more than just triangles per second. So, we look beyond these numbers for a more holistic view. We perform in-house testing on commonly-used industry benchmarks that represent overall GPU performance.* Our testing shows the OMAP4430 outperforming the competition. With its major graphics and memory bandwidth improvements, we believe the OMAP4470 will continue this trend with 30-40% better performance compared to competitive devices available in the same timeframe.

http://e2e.ti.com/blogs_/b/mobile_m...he-scenes-of-ti-s-new-omap4470-processor.aspx

Seeing where the high end of the smartphone market and the tablet market actually split, despite manufacturers' claims of their SoC's suitability for smartphones, will be interesting.

I think TI's heterogeneous model of low power CPUs and GPU along with their power management make a good case for smartphone applicability for OMAP4470 (although the market may just not need full DirectX 9 compliance in that form factor during much of the OMAP4470's lifespan), and I wonder how many quad-core Tegra 3 variants we'll actually see in smartphones. I guess near the latter part of Tegra 3's life, the quad core variants will provide a way to scale it.

 

[Ailuros]

I don't know about Tegra3 yet, but OMAP4 still could use some design wins. They're probably already in the pipe, yet so far the announcements were quite few.

In the background rumors are circulating that OMAP4 will be the reference platform for Google's Ice Cream Sandwich, which could mean that OMAP4 might win a far bigger uptake in terms of design wins than so far.

***edit: I almost forgot....regarding that quote above I'd interested to see how OMAP4430 (like the LG Optimus 3D) would fair with the latest IMG driver in the Google Nexus S and the Samsung SGH T-959 Vibrant. Earth calls Rys.....

 

[Rys]

I'll see what I can do

 

[rpg.314]

In the background rumors are circulating that OMAP4 will be the reference platform for Google's Ice Cream Sandwich, which could mean that OMAP4 might win a far bigger uptake in terms of design wins than so far.

If there's any truth to that, I hope they choose the one with 544 in it.

 

[Ailuros]

If there's any truth to that, I hope they choose the one with 544 in it.

Why? Will there be anything in Ice Cream that a 4430 couldn't cope with? Google isn't going to chose something that isn't available for the projected release of the new OS version; otherwise they could have chosen something like Tegra3 too.

 

[rpg.314]

Why? Will there be anything in Ice Cream that a 4430 couldn't cope with? Google isn't going to chose something that isn't available for the projected release of the new OS version; otherwise they could have chosen something like Tegra3 too.

I was hoping Google would push the hw envelope as well.

 

[Ailuros]

I was hoping Google would push the hw envelope as well.

I'm not sure how I should understand that. Ice Cream is supposed to be the first Android version to be suited for both tablets and smart-phones. So Google should push the envelope as high that the even the cheapest smart-phone running Ice Cream would need 2*A9@1.8GHz and a >450MHz SGX544?

 

[rpg.314]

I meant google should specify leading edge hw for the "Google Experience" device. Other devices can use cut down hw. But it will probably not happen.

 

[Ailuros]

I meant google should specify leading edge hw for the "Google Experience" device. Other devices can use cut down hw. But it will probably not happen.

I can't know with what criteria Google choses but so far we've seen Qualcomm, Samsung and NVIDIA SoCs being reference devices for their OS revisions. I could think that they might want to give everyone a chance (besides SoC availability) which if true is a good thing in my book.

Availability is the primary factor that leads me to speculate in the 4430 direction; with the latest driver the LG Optimus3D could get </=40fps in Egypt standard which is anything but a slouch by today's standards. 4460 obviously a healthy notch higher and 4470 whatever within vsync limits.

 

[stiftl]

I'm not sure how I should understand that. Ice Cream is supposed to be the first Android version to be suited for both tablets and smart-phones. So Google should push the envelope as high that the even the cheapest smart-phone running Ice Cream would need 2*A9@1.8GHz and a >450MHz SGX544?

Just asking: Is there anything like a minimum requirement from Google for Android?

Thinking of something like only A8/A9/A15/X86 processors or 512MB RAM minimum for ICS.

Was there something like this in the past?

I think it would make sense to clear out the driver base (for example ditch support for ARM9/11), because nobody will use ICS for specs lower than these anyway.

 

[Ailuros]

Good question, to which unfortunately I don't have an answer. Can anybody with some insight help out possibly?