Article: Inside the NVIDIA APX 2500

[Ailuros]

I think thats not right!
All SGX cores support OpenGL 2.0, OpenGL ES 2.0 and Direct3D Mobile. AFAIK the last is something like DX7/DX8, I would say at max SM2.0.

Only SGX535 is DX9 and SGX545 is Direct3D10.1. I think both cores especially made for one vendor, who need such cores for windows systems.
Yes I know NEC use the SGX535 for the NaviEngine1.

I'm not so sure but removing capabilities below SM3.0 for the 5th generation sounds more complicated than leaving it inside. What each driver for each platform exposes is a story of it's own probably. IP below 535 for their target markets don't obviously need SM3.0.

In any case it would be highly interesting to see how someone would scale a unified shader core down to DX8.0 or below. Even the first generation VGP as a geometry processor is DX8 compliant (with 4 fp operations/clock).

 

[RussSchultz]

I'm not so sure but removing capabilities below SM3.0 for the 5th generation sounds more complicated than leaving it inside. What each driver for each platform exposes is a story of it's own probably. IP below 535 for their target markets don't obviously need SM3.0.

Why do you say that? You obviously could still use SM3.0 on a smaller screen (QVGA vs. VGA, for example)

 

[roninja]

I don't know for sure but I think Omap3420 and 3430 both will be SGX530. The 535 is to big for this cores, there are to many transistors they don't need.
4340 I think SGX540, I can't give reason for this.

And yes there is no SGX510, I have the confirmation from PowerVR.

Loewe. I notice on the deferred power website you have a slide PowerVR SGX roadmap on it. Reference is made to PowerVR VGX 1xx (in Planning) and targetted at the mobile low end. Can someone elaborate?

 

[mboeller]

I don't know for sure but I think Omap3420 and 3430 both will be SGX530. The 535 is to big for this cores, there are to many transistors they don't need.
4340 I think SGX540, I can't give reason for this.

Why?

according to the hotchips tutorial "HC19.tutorial1.01.pdf" from last year the OMAP 2420 already has ~ 90Mio Transistors @ 90nm and the MBX was not the biggest part of it (page 6 of the presentation). Contrary to that it seems that the SGX within the OMAP 3430 seems to be the biggest part of the chip (CPU #3 page 5 ).

Link: http://www.hotchips.org/archives/hc19/1_Sun/HC19.tutorial1.01.pdf

 

[Loewe]

Loewe. I notice on the deferred power website you have a slide PowerVR SGX roadmap on it. Reference is made to PowerVR VGX 1xx (in Planning) and targetted at the mobile low end. Can someone elaborate?

Yes I ask David Harold and I get this answer:

VGX is another family like SGX and MBX but focused on OpenVG.

I think there won't be much 3D stuff in this core, "...fast scalable anti-aliased vector graphics enables advanced user interfaces, mapping applications, games and screensavers... (Khronos)", says it for the lower end.

 

[Ailuros]

Why do you say that? You obviously could still use SM3.0 on a smaller screen (QVGA vs. VGA, for example)

Could yes; what a low end 3D phone really needs is another story.

 

[Ailuros]

Why?

according to the hotchips tutorial "HC19.tutorial1.01.pdf" from last year the OMAP 2420 already has ~ 90Mio Transistors @ 90nm and the MBX was not the biggest part of it (page 6 of the presentation). Contrary to that it seems that the SGX within the OMAP 3430 seems to be the biggest part of the chip (CPU #3 page 5 ).

Link: http://www.hotchips.org/archives/hc19/1_Sun/HC19.tutorial1.01.pdf

You obviously assume that CPU #3 is the graphics part of the chip; unless I've overlooked something it's nowhere clear in that pdf what each power domain in OMAP3430 relates to. Reasonable speculation yes, a guarantee no.

 

[RussSchultz]

Could yes; what a low end 3D phone really needs is another story.

Ummm, which "high end phone" has anything bigger than WQVGA? (i.e. half vga).

Besides the Nokia 810/910, of course, and even then you're scratching around trying to find too many that have anything other than QVGA.

p.s. I believe the 5xx series all have the same features, just different clock speeds and processing units.

 

[3dcgi]

Ummm, which "high end phone" has anything bigger than WQVGA? (i.e. half vga).

Besides the Nokia 810/910, of course, and even then you're scratching around trying to find too many that have anything other than QVGA.

p.s. I believe the 5xx series all have the same features, just different clock speeds and processing units.

The recently announced Sony Ericsson XPERIA X1 that's coming out later in the year will have a resolution of 800x480. So higher res phones are coming though I expect most to stay with lower res.

 

[RussSchultz]

I don't disagree that phones aren't moving toward VGA and WVGA, but to say you don't need or want SM3 or "better pixels" for QVGA or WQVGA is a bit presumptious. On phones, form factor is a huge determining factor as to the size of the screen...but you still probably want the screen to look as good as you can get it.

 

[wishiknew]

Hi all. I'm kinda confused. The omaps have that diagiam with all the arrows connecting to the other parts of the phone. Is this something the nvidia chip also do or is it something more specialized.

And is it that much of a change for future phone chips to encode in avchd instead?

 

[Ailuros]

Ummm, which "high end phone" has anything bigger than WQVGA? (i.e. half vga).

Besides the Nokia 810/910, of course, and even then you're scratching around trying to find too many that have anything other than QVGA.

p.s. I believe the 5xx series all have the same features, just different clock speeds and processing units.

SGX545 is afaik D3D10.1 compliant; there's quite some difference between let's say SM3.0 as a minimum and SM4.1 isn't it? 545 then isn't obviously targetted for a high end phone either I guess and there's still also a 555 on their roadmap.

Their own SGX whitepaper states from 1.5 up to 20.3mm2@65nm and 100m - 4000m pixels/s at 200MHz. I've no idea if the lowest numbers are for 510 and what 520 stands for in contrast, but if there are similarities between them it would rather be 2 pixels/clock for the lowest end (just like MBXLite) in a relative sense and 1 pixel/clock for the higher models. Otherwise there's no reasonable explanation for 100MPixels@200MHz.

I don't disagree that phones aren't moving toward VGA and WVGA, but to say you don't need or want SM3 or "better pixels" for QVGA or WQVGA is a bit presumptious. On phones, form factor is a huge determining factor as to the size of the screen...but you still probably want the screen to look as good as you can get it.

I'd love to stand corrected as a layman, but wouldn't "better pixels" presuppose a wee bit more fillrate?

 

[roninja]

Wonder what target product the SGX 555 is aimed at?

 

[Lazy8s]

It took at least two clocks for SGX510's USSE-Lite pipeline to draw a pixel.

While BitBoys's profitability had already validated the market opportunity, and possibly vector graphics performance advantage, for devices with vector acceleration but no real 3D, the existence of the VGX family must mean that Imgtec feels they're missing out on some important design wins.

In the November 2, 2007 Imgtec PR announcing the increasing pace of mobile 3D adoption and the introduction of the SGX540, the SGX510 was sidelined as Imgtec projected that the 520 would set the base-line of performance going forward for SGX parts in devices.

 

[Lazy8s]

A TI presentation a while back quoted the OMAP3420's polygon performance at half the 3430's, and I doubt clock speed variation was used to make that scale. I think the SGX520 and 530 are being used in the OMAP3420 and 3430, respectively, despite evidence from Imgtec's announcements of TI licensing not supporting that idea.

Along that line, I also think the SGX540 is the core in the OMAP3440 since that SoC seems to upscale the various components of the OMAP3 architecture for the MID market.

 

[knux]

Broadcom's BCM2727 is claimed to be capable of 5 hours of 720p video playback. It's not clear what codec that refers to though, and whether it includes the screen or was measured via HDMI streaming. As for NVIDIA's other direct competitors - the OMAP3430/3440 and the STw4820 - none of them release any power consumption number at all, so who knows. I'd tend to argue however that if it was lower, they'd be bragging about it right now...

Rated at 620MHz? Not at all; that rumour comes from ARM's claims for ARM11 in general, which is just an indication for potential licensees. That doesn't mean the specific chip in the iPhone could reach that frequency - it very likely could not. However, I was also wrong and the iPhone's correct frequency is 412MHz. Strange, my initial googling turned out a reliable source with a number in the 300s - oh well, fixed!

Actually, the OMAP 3 platform IS indeed implemented in a 65nm process:
From Feb 07 issue of Powervr Insider Newsletter:
"The first devices in TI’s OMAP 3 multimedia
applications processor architecture are
implemented in a 65nm process and utilise
the ARM Cortex™-A8 to deliver up to a 3x gain
in performance over ARM11-based
processors."


Why should they be bragging about it? They probably don't even consider the APX 2500 to be a competitor for the OMAP3430. Oh, and are you guys also not realizing that Cortex-A8 is roughly 2-3x faster than ARM11 clock for clock? Therefore even if the OMAP3430 is clocked at 550/600 Mhz, it'll still be 1.5-2x faster than the 750 Mhz ARM11 in the APX 2500.

 

[knux]

Further proof of the OMAP3430's power consumption:
http://www.ieee-mobilewimax.org/downloads/Shaver.pdf

scroll down to page 8

 

[Arun]

Errr, the OMAP3 Family has been known to be implemented in 65nm since it's announcement (and even before). This isn't news and I don't know why you think it is. And wait, what? Are you implying the only factor that matters in terms of power consumption is the process technology used? That is so far from being the case I don't know where to start. I have no idea why you think you have the knowledge necessary to judge these things.

As for Cortex-A8 vs ARM11, you once again don't seem to make any effort to get your facts straight. From my article: 'Based on a little bit of Googling, we found out that should result in performance of 920 Dhrystone MIPS, while TI's OMAP3430 should deliver 1100 Dhrystone MIPS with its 550MHz Cortex-A8. So definitely pretty close and likely a decent trade-off, given the noticeably smaller die size and possibly lower power.' - you can verify those performance numbers based on the clock speeds and the Dhrystone claims from ARM's website. So please stop spreading FUD, Cortex-A8's performance/clock isn't nearly twice that of the ARM11.

 

[knux]

Yeah I guess you're right. I stand corrected.

 

[Arun]

Heh okay, at least you're willing to admit error on that, that's great! Certainly makes for a much more pleasant discussion on both sides of the issue. Either way, I'm not saying OMAP3 has lower power efficiency; I don't know that. It very well might be more efficient and TI's PR strategy might just be less aggressive because they don't think they need to impress the press, just the OEMs. I don't think that's very likely, but it's possible. I'd certainly be very interested in any real data here, although I fear there really isn't any of that in the public domain.