PowerVR MBX in iPhone

[Rob Evans]

It seems that the iPhone main chip contains a PowerVR MBX 3D graphics processor.

From http://iphone.fiveforty.net/wiki/ind...ree.m68ap.img2

device 0 (10 rec)
"device_type" => 0x78626D
"clock-ids" => 0x2
"reg" => (8) "\x00\x00\x00\x03\x00\x00\x00\x01"
"power-gates" => 0x2
"clock-gates" => 0x13
"compatible" => (13) "mbx,s5l8900x\x00"
"interrupt-parent" => 0x403BD0
"name" => 0x78626D
"interrupts" => 0xC
"AAPL,phandle" => 0x40C570

N.B. 0x78626D is ASCII for xbm - mbx in reverse format.

[RedGuard]

MBX libraries are also mentioned in crash logs:-

0x31baf000 - 0x31bb1fff MBXConnect UUID (B5E73A95A84E4706A922D56674D4809D) /System/Library/Frameworks/MBXConnect.framework/MBXConnect
0x31bb4000 - 0x31be5fff OpenGLES UUID (3405A88B732E4DF3A127E182483D9E69) /System/Library/Frameworks/OpenGLES.framework/OpenGLES

[obo]

The following graphics-related MBX strings can be found in file57 in the zip file at the top of this page:

AppleMBXDevice(%p): Graphics hang with pending 2D blit.  Free slots=%d, IntStatus=%08x
 AppleMBXDevice(%p): Graphics Restart. Known MBX hardware issue (rdar://4809859)
    AppleMBXDevice(%p): TA not idle, deferring graphics restart
 AppleMBXDevice(%p): Graphics Restart during 3D Blit.  Dummy page corrupted=%d, 3DR=%d, ISP=%d, EVM=%d
  AppleMBXDevice(%p): Graphics Restart during 3D Render.  Dummy page corrupted=%d, 3DR=%d, ISP=%d, EVM=%d
    EnableGLClockgateWorkaround AppleMBXDevice(%p) - Warning: Switching clockgate workaround modes when the HW is not idle.  Results undefined.
    %s: EnableGLClockgateWorkaround key of wrong type
 AppleMBXDevice(%p)::flushWrapperCache: blit cache flush took too long
  AppleMBX: Failed to add a surface mapping because prepare() failed or couldn't fit in MMU.  Skipping command.
   AppleMBXDevice(%p)::interruptHander: got MBX1_INT_TA_STREAM_ERR
 Invalid surface %08lx while processing command sequence.  Aborting command buffer.
   virtual bool AppleMBXDevice::start(IOService*)  mbx-clockgate   chip-revision   AppleMBX: Disabling auto clock gating because this is a EVT0/EVT1   AppleMBX: Enabling auto clock-gating: %d
   IOCoreSurfaceRoot   %s: IOCoreSurfaceRoot not available or of the wrong class
  compatible  mbx,s5l8900x    true    false   AppleMBX: Detected %s.  SupportsWriteCombining: %s
 PPPPPPPP    -RRRRRGGGGGBBBBB    --------RRRRRRRRGGGGGGGGBBBBBBBB      AppleMBXDevice(%p)::start - Failed to setup 3D blit support
    IOMobileFramebuffer AppleMBX Workaround: Clock gating disabled
AppleMBXUserClient  AppleMBXUserClient(%p)::mapSurface - Failed to malloc MBXSurface
   Failed to look up CoreSurfaceID %x
 Failed to lock a CoreSurfaceBuffer: err=%x
 AppleMBX mapSurface: too large to map: size=%ld max allowed=%d
 AppleMBXUserClient(%p)::wireSurface - Failed to create wired MBXSurface
    AppleMBXUserClient(%p)::wireSurface - Failed to wire MBXSurface
    AppleMBX wrapSurface: too large to map: size=%d max allowed=%d
   AppleMBXUserClient(%p)::addRenderTarget: Failed to run AddRenderTarget command
    cleanupMemInfo: failed to clean up surface
 Failed to unlock a CoreSurfaceBuffer
   PVR:     (%u,%s)
   Debug assertion failed! PVRDebugAssertFail

[roninja]

now to find out whether this is part of the SoC or a standalone chip!

[loekf]

Originally Posted by roninja

now to find out whether this is part of the SoC or a standalone chip!

Oh my oh my... this is quite interesting. You can almost draw a block diagram just from the list of devices. Some interesting bits and pieces:

- ARM1176jzf-s - ARM11 with Trustzone (allows you to run a virtual CPU with security features, so in kind of sandbox)
- Interrupt controller : PL092 and DMA controller : PL080. These are off-the-shelf IP blocks from ARM itself.
- Security IP: looks like there's a HW RNG, SHA1 and PKE (private key accelerator ??) block.
- Looks like there's a CLCD and LCD controller. In other words there are in fact two LCD controllers, one for RAM-less and one for RAM-based LCDs.
- NAND flash controller with seperate ECC block
- NOR flash controller
- JPEG accelerator
- H.264 accelerator - thing is called h264bpd -> h.264 base-line profile decoder ?
- TV output (but is not used by Apple...)
- MPVD... multi-purpose video decoder or something ?
- MBX.. yes it's there
- MPL = mobile pixel link, LM2511 is National IC, was spotted in several teardowns
- TSL2561.. why is this IC in there ? This is a light sensor. To do backlight dimmning of the LCD depending on ambient light ?
- micron2020 camera ... yes CMOS camera sensor from Micron
- EDRAM ???
- NVRAM, next to NAND flash ?
- There seems to be an unique ID

[Magallanes]

From the page of ARM:
http://www.arm.com/products/CPUs/ARM1176.html

Optional Vector Floating Point coprocessor (ARM1176JZF-S)
Powerful acceleration for embedded 3D-graphics


But i'm not sure about how powerful it is. For instance today the Axim X51V, Nokia N93i and Nokia N95 are indeed powerful but the rest (even with dedicated gpu) are just a bluff with a crappy performance (Even Sony Ericssons with a OMAP 2420).

[Brimstone]

So all the rumors of nVidia winning a contract with Apple for the iPOD and iphone were wrong?

[Lazy8s]

The acceleration for 3D from the vector floating point co-processor isn't 3D-specific acceleration; it's just an FPU for floating point support which could be used in operations like T&L. That processor on the ARM page is just the ARM11 CPU, not a CPU + GPU.

The Sony Ericssons with lower graphics processor performance don't use the OMAP2420; they use a chip with a smaller graphics core, the PNX4008, configured with an MBX Lite, no VGP Lite, no FPU, and an ARM9.

[Jawed]

Originally Posted by Brimstone

So all the rumors of nVidia winning a contract with Apple for the iPOD and iphone were wrong?

Yep, poor Arun's wet dream.

Jawed

[Tim Murray]

Well, this thread got linked by EETimes...

[Arun]

Originally Posted by Jawed

Yep, poor Arun's wet dream.

Hey, I'm openly more biased towards PowerVR than any other semiconductor company out there, from a technological perspective. It's not my fault if there aren't enough opportunities to show them some love!

Honestly, I hadn't even realized that Samsung had a MBX license before all this. I obviously hadn't done my homework! (but fwiw, it seems that EE Times also got this partially wrong, because Samsung is not a 'new partner'. So, who is this new mystery partner? Hmmm!)

[Brimstone]

So did nVidia win a design contract with Apple for anything?


I'm wondering if the speculation was right on a nVidia+Apple contract, but they just got the target device wrong?


Was a contract ever hinted at in Imagination Technologies confrence calls?

[Arun]

Originally Posted by Brimstone

So did nVidia win a design contract with Apple for anything?

There is a pretty good chance that the next-gen high-end iPod is based on the same application processor as the iPhone IMO, so no. I don't really see NVIDIA fitting in the iPod Nano at this point either... So unless Apple decides to change their lineup, I don't see how they could possibly fit in, and even then it wouldn't make much sense.

Their next-gen platform, which will tape-out probably in a few months at worst if it remains on schedule, will probably compete for the next-gen iPhone and iPod, but that's another thing completely, and in the end, 'compete' is exactly what they'll do - they don't have a magic backdoor here, so it's hard to judge their chances of success without knowing anything about that upcoming chip.

[vitaminc]

Originally Posted by Brimstone

So did nVidia win a design contract with Apple for anything?


I'm wondering if the speculation was right on a nVidia+Apple contract, but they just got the target device wrong?


Was a contract ever hinted at in Imagination Technologies confrence calls?

Portal Player was in Gen 7 iPods and was displaced by Samsung in Gen 8. nVidia bought off Portal Player afterwards.

Apple's notebooks use nVidia's GPU though.

[Arun]

Originally Posted by vitaminc

Portal Player was in Gen 7 iPods and was displaced by Samsung in Gen 8. nVidia bought off Portal Player afterwards.

I never managed to find any confirmation that PP was displaced by Samsung in the 'iPod with Video', except by the fact that NVIDIA only had $10M of PP revenue in Q107, and this roughly corresponds to their non-Apple revenue iirc. So, any link to a reliable proof of that, and any info on exactly when it happened?

[vitaminc]

Originally Posted by Arun

I never managed to find any confirmation that PP was displaced by Samsung in the 'iPod with Video', except by the fact that NVIDIA only had $10M of PP revenue in Q107, and this roughly corresponds to their non-Apple revenue iirc. So, any link to a reliable proof of that, and any info on exactly when it happened?

http://ce.seekingalpha.com/article/16904

I remember PLAY had more than $100M in annual revenues before it losts everything and sold itself to NVDA.

They lost the iPod Nano socket to Samsung, but still has the Video iPod music decoding socket.

Video iPod uses Alphamosaic's (acquired by Broadcom) display decode chip though. I believe there were some rumors saying nVidia designed out Broadcom on the next gen iPod, but Apple has not announced their plans for next gen iPod yet.

[INKster]

Originally Posted by Brimstone

So did nVidia win a design contract with Apple for anything?

Besides the 8600M GT inside the new Macbook Pro's, there's always the Apple TV (Geforce 7300 GS) and the default graphics card inside the Mac Pro (7300 GT).

[hidefguy]

Originally Posted by Lazy8s

they use a smaller chip, the PNX4008, configured with an MBX Lite, no VGP Lite, no FPU, and an ARM9.

FYI, I believe the SE devices have a hardware FPU.

[Lazy8s]

Indication from Imagination Technologies that they were in the iPhone was inferred by considering the description and timetable for launch of the unnamed "other high profile handsets" in the following comment from their recent financial statements:

In the mobile phone market segment there are over 45 handsets that are based on our technologies. These include handsets from Nokia, Sony Ericsson, Motorola, NEC, Fujitsu, Mitsubishi, Panasonic, Sharp and Motorola. We continue to have a growing position in the Japanese mobile phone market through our very strong ranging across the latest DoCoMo phones. In addition to Sony Ericsson’s W850i, M600i, and P990i models and Nokia’s N93 and N93i models which are already shipping, we are now seeing, albeit a little later than forecast, a growing number of other high profile phones which are now shipping, or about to ship, in western markets. Amongst these are the acclaimed Nokia N95, Nokia E90, the Motorola MotoRIZR Z8 and MotoQ q9 and Sony Ericsson P1i. We also expect other high profile handsets to begin shipment starting from the second quarter of the current financial year. These new phones, targeted at western as well as global markets should significantly accelerate the volume ramp up for FY 07/08 whilst also demonstrating the real benefit of graphics for both the user interface and application content.

The possibilities for the configuration of the custom Samsung solution are numerous with options for the full or Lite versions of MBX, an accompanying VGP, the CPU variant, and the clock speeds.

Although today's announcement, that an "international electronics systems company" has licensed Imagination Technologies' "next generation graphics and video IP cores" under a "multi-use" agreement, combined with Apple's policy for secrecy about the identity of their parts' suppliers suggests that the announcement is about them, the "next generation" qualifier in the announcement would, in conventional Imgtec PR terms, refer to SGX. This would imply the possibility for SGX based iPhone, iPod, and Apple TV follow-ups, but it would leave unclear which MBX cores were licensed for the first iPhone. The ARM R-S (MBX Lite with no VGP Lite) agreement taken in January 2004 by Samsung covered just the S3C2460 at the time.

If Apple's design priority for the custom Samsung solution in this first iPhone was performance, a full MBX and also the VGP were probably included. If the CPU core really is at least a >500-MHz ARM11 and the GPU corresponds at a reasonable fraction of that, the hardware would overtake OMAP2420 systems finally as the most powerful in the portable sector.

[hidefguy]

Originally Posted by Tim Murray

Well, this thread got linked by EETimes...

A spokesman for Imagination said: "We can't say whether we are designed into the Apple iPhone."

Unbelievable. The device is out in the market place, people have hacked them apart yet neither Apple nor Imagination will confirm anything - this could only happen in the handset sector.

[vitaminc]

Originally Posted by hidefguy

Unbelievable. The device is out in the market place, people have hacked them apart yet neither Apple nor Imagination will confirm anything - this could only happen in the handset sector.

this is exactly why i dont think nvidia will have much chance to succeed in cell phones. handset market is a very closed ended with in-house/proprietary hardware/softwares, compare to a standardized and commoditized hardware/software PC market.

[INKster]

Originally Posted by vitaminc

this is exactly why i dont think nvidia will have much chance to succeed in cell phones. handset market is a very closed ended with in-house/proprietary hardware/softwares, compare to a standardized and commoditized hardware/software PC market.

Kind of like the PC discrete 3D graphics card scene, circa 1994...

[Arun]

Originally Posted by vitaminc

I remember PLAY had more than $100M in annual revenues before it losts everything and sold itself to NVDA.

They had much more than that, $225M+ in 2005 and Q106 was also a very good quarter, with $72M+. But then in Q3, after the loss of the Nano, this happened:

Originally Posted by PortalPlayer

The third quarter financial results we are reporting today are once again inline with our previous guidance, with revenue of $34.8 million. In the third quarter, more than 10% of this came from what we've termed other customers and we expect revenue from other customers to increase.

And this...

Originally Posted by PortalPlayer

And before I turn the call back over to Gary, let me summarize our guidance for the fourth quarter. We expect revenue to be about the same as in the third quarter with a range of $31 million to $38 million.

So far so good, right? Then they got bought by NVIDIA and Q4 never got reported. But this was still said at NVIDIA's Q4 CC, keeping in mind that their Q4 ends on January 28th:

Originally Posted by NVIDIA

We closed the purchase of PortalPlayer on January 5th, but the revenue contribution from Portal was not significant, approximately $1 million.

This surely was a one-time thing because of timing issues, right? But then in the Q1 CC (and this was roughly in line with the Q4 guidance)...

Originally Posted by NVIDIA

PortalPlayer products accounted for $11 million in the quarter
[...]
We incurred a full quota of PortalPlayer expenses instead for the one month that was incurred in Q4. And this added about $10 million.

The expenses are what you'd expect, given that they were about $12.5M pre-acquisition and there might be some synergies at play. But the revenue? WTF?

I have yet to find a proper explanation to this. That amount of revenue would be just $6M from Apple if we expected non-Apple revenue to be flat, and a grand total of $0 from Apple if we expected non-Apple revenue to grow significantly. In either case, it doesn't make much sense to me. PortalPlayer's cash position remained roughly flat in Q4 too (because the net price was $160M, as expected) so there was no spike in this unreported quarter, as far as I can tell.

Maybe a very significant reduction in iPod Video production along with a reduction in ASPs... Still, weird.

[Lazy8s]

Because the S3C2460 apparently never made it to an end-user device, perhaps Samsung's January 2004 MBX Lite license was never used up and was still available for the iPhone chip. Maybe Samsung even intentionally passed up S3C2460 sales opportunities for it, though I doubt the whole scenario.

Imgtec would probably have to share the revenue from the chip with ARM in accordance with that old arrangement in this unlikely case, too.

[vitaminc]

Originally Posted by INKster

Kind of like the PC discrete 3D graphics card scene, circa 1994...

extremely different.

1994 PC 3D graphics add-on ->

• Standadized hardware interface (ISA, followed by AGP and PCI-E) that allows plug/play.
  
• Standardized OS (DOS, Windows).
  
• Standardized form factor (AT, ATX)
  
• Commoditized product. A PC bought from Packard Bell isn't much different from that of Compaq.
  
• Open OEM/ODM and distribution channel. You have 345098364723986 Taiwanese guys trying to compete for the same Dell/HP order and cut throat prices. Commodity manufacturing for commodity product.

2007 cell phone graphics core ->

• NON standardized interface. sometimes an add-in chip, sometimes integrated in application processor (samsung w/ imgtec), sometimes integrated in baseband (qualcomm w/ ati). And you cant plug/play.
  
• NON standardized OS. In-house OS by different venders, Symbian 6, windows mobile.
  
• NON standardized form factor. different sized phones have different PCB real estate budgets for chips.
  
• Feature rich product. At a given price point, you have phones with different feature sets. e.g., W for walkman, N for smart phones, and iPhone for a cool browser.
  
• Closed OEM/ODM and distribution channel. Can't just go to Taiwan for 1 stop shopping for PCB, distributor, OEM, etc.