Nvidia/Apple iPod CPU
- Thread starter INKster
- Start date
Links in this post. Analysis in the next one 
I think this: http://www.arm.com/news/9285.html (note the press release's date, too...)
I think this: http://www.arm.com/news/9285.html (note the press release's date, too...)
I also think this: http://www.us.design-reuse.com/news/news2132.html
And this, too: http://www.techtree.com/India/News/Nvidia_C55_Made_in_India/551-77364-581.html
Oh, and maybe this too: http://www.rediff.com/computer/1999/jan/12s3.htm
And even this, most certainly: http://www.findarticles.com/p/articles/mi_m0WUB/is_2000_Feb_7/ai_59332078
What you've fundamentally got to realize is that PortalPlayer isn't designing CPUs. They're just designing SoCs with ARM cores in them. Their sound processing is also (mostly) done on ARM7 cores. Honestly, it's a shitty approach compared to what dedicated silicon can buy you, but in their case it does have one massive advantage: the SoC can serve as both a CPU and a sound processor. It's a bit more complicated than that because there are two distinct ARM cores on their high-end offering iirc, but that's the basic idea.
So what you got there is a company that specializes in SoCs and that has a ton of relationships in the market, and especially so with Apple of course. But besides for some good power management techniques, and some good firmware to do sound processing on ARM cores, they really don't have that much IP. So NVIDIA acquired them mostly for their market positioning, the engineers and the design centers, imo. Plus, it was cheap: after counting PortalPlayer's cash reserves, it cost less than $180M, which would be equivalent to a net profit of $9M/year with a P/E of 20. That's pocket change for NVIDIA, and truly negligible considering the PortalPlayer's potential worth in the long-term.
Anyway, it's also important to take into consideration that NVIDIA already has a sound processor in the GoForce 5500. It is, in fact, wonderfully power efficient: barely 20mW for MP3 decoding, compared to PortalPlayer's 60-80mW in the chip they're currently still selling to Apple. Honestly, PortalPlayer's products aren't competitive anymore, and the sell-out is merely based on the fact that they knew they were completely and utterly toast, going forward, at least when it comes to Apple.
Chances are that NVIDIA is designed in the iPhone and PortalPlayer isn't, ironically. The rumoured specifications perfectly match the capabilities of the GoForce 4000 (and yes, I checked ATI's, Broadcom's, etc. offerings, and while some were pretty close, they weren't quite as spot-on). So I could easily see that product being based on a Samsung or Sigmatel application processor and a GoForce 4000. It could be anything else, of course, but at this point I think that's the most likely possibility.
Or, it could be a custom-designed ASIC by NVIDIA or another company that does exactly what Apple needs; when you're talking 12M+ units, such things become much more probable. If Apple wants to do sound processing in the same chip as video processing, then ATI's Imageons might also be a possibility, although few members of that product line match the target resolution. BTW, I think the GoForce 5500's sound engine was created at NVIDIA's Bangalore design center. That stuff used to be the founders' speciality, after all, and they designed sound and GigE ethernet ASICs with a team of ~20 people a few years ago.
So, why did NVIDIA buy PortalPlayer? Market relationships and engineers. It's basically the same reasons NVIDIA bought ULi or Hybrid Graphics. This time around, the sums involves are much larger, but so are the number of engineers and the company's current revenue. Another factor might be that they felt they needed their expertise for SoCs - having licensed ARM11 MPCore doesn't mean they felt they had the best talent to integrate it. There's no way for us to ever know that kind of detail, most likely.
Anyway, when it comes to the 2008 timeframe, which is most likely what you asked me to focus on (), NVIDIA is focusing their efforts tremendously on the 65nm Low Power process. That process has traditionally been used in notebook and handheld parts, but this time around, they're being a lot more aggressive in using it. TSMC claims their focus on the low power process (which ramped up first, before the general purpose one) is based on customer demand; and I think NVIDIA is one of the several companies that asked for them to take that direction, or at least I wouldn't be surprised if they did. One thing people might not realize is that NVIDIA's DX9 flagship, the 7950GX2, is based on the 90nm Low Power process.
So this has some obvious implications when it comes to their PC GPU business, but I'm not going to focus on that here. Instead, it should be fairly obvious that NVIDIA is also targetting 65nm very aggressively for handheld parts, and they'll be depending on it for their next-gen GoForce product. At this point in my point, it should be fairly obvious that this product will sport a ARM11 MPCore processor (I think it'll likely be a dual-core and not a quad-core one given the focus on power consumption, but that's another thing). It'll also sport a GPU shader core that is CUDA compatible and DX10+ level. It could be G9x-based (the shader core, not the entire GPU!), or it could be based on one of their recent handheld patents for an unified programmable shader core.
While G9x-based might seem a tad insane for a handheld part, the shader core's power efficiency really is that good, if you scaled it down to ~1/2 of a cluster. Think of the G7x-based IGP, which had half a quad. And then, unlike current desktop parts, you might (read: you will) be doing things like triangle setup in the shader core. So, I think it's definitely possible in terms of power usage and area, but it remains to be seen whether that's what NVIDIA is going to do. It very likely is not (as I said, they got some interesting handheld patents for unified architectures too), but we'll see about that. Either case, it'll be a true next-gen architecture, with features equal or most likely above those of SGX. I've heard some ex-Gigapixel guys are quite busy at NVIDIA nowadays, so who knows what they have in reserve!
Sound and video are no-brainers, considering NVIDIA already has that in the GoForce 5500, along with a bunch of other stuff. And the power consumption numbers they are claiming already are pretty good AFAIK, so it can only get better on 65nm Low Power. NVIDIA is clearly aiming at getting both the Video iPod contract and the 3G iPhone contract. If they can get the other iPod contracts for the same price, they won't complain, but I don't think it is quite as key to them. Those have quite a fair bit lower margins. I'd be surprised if NVIDIA quit that market completely, however. I think it's relatively likely they'll just integrate an ARM7 with the GoForce 5500's sound processor, manufacture it on 65nm, and that's it.
There are two other contributing factors I didn't touch on but might still be worth mentioning. First, PortalPlayer's Preface. NVIDIA is gaining laptop share rapidly, so they're in an amazing position to bundle Preface with their GPUs. They could try all manners of strange things, like platformizing with Preface/nForce/GeForce/WiFi, although I'm a bit skeptical they'll be doing that yet.
The second factor is the PSP2. I think it's fairly damn obvious NVIDIA has got the contract at this point. So they need to focus a lot more on the 3D part of the handheld market - which means they also need to amortize that R&D somehow. So even if the short-term revenue isn't there for it, they'll get it anyway. Whether indirectly through Apple (since I doubt Apple primarily cares about the 3D functionality, although they'll get it anyway, and might be better off as a result imo), or through Sony. NVIDIA's past handheld offerings have been rather uninteresting in a few ways, but I think they're really beggining to focus a lot more energy on it, and the fight with AMD's comparatively huge handheld division will be interesting, to say the least! Other competitors will enter the picture, too.
BTW, one last thing when it comes to PortalPlayer. It wouldn't surprise me if they were designed in the iPhone (although I see little evidence that it definitely is), but I don't think it's their 5022 or 5024 product lines then. The 5022 is still on 130nm, so that certainly explains part of the power disadvantage they have (which I explained above). If you look at the conference call transcripts for Q1/Q2 2006 with PortalPlayer, you can get a small idea of their strategy there, and they seem very aggressive in getting a few Apple sockets back. Now, I'd speculate that whatever PortalPlayer would be proposing for iPhone would probably be an unannounced product. So that certainly complicates speculating on who got the contract.
What's farly obvious anyway is that the GoForce 5500 is overkill for the rumoured first iPhone model. It might not be for the Video iPod though, but it's rumoured that Samsung won that socket. I'm not sure I buy that, honestly. The fact of the matter remains that analysts are contradicting themselves a LOT on Apple products, and that's for two reasons - first of all, they generally just don't know anything for sure. Secondly, the amounts of money involved with this kind of rumour or "analysis" are enormous, and it's easy for anyone - even analysts - to get overly excited and misinterpret something he heard.
Uttar
So what you got there is a company that specializes in SoCs and that has a ton of relationships in the market, and especially so with Apple of course. But besides for some good power management techniques, and some good firmware to do sound processing on ARM cores, they really don't have that much IP. So NVIDIA acquired them mostly for their market positioning, the engineers and the design centers, imo. Plus, it was cheap: after counting PortalPlayer's cash reserves, it cost less than $180M, which would be equivalent to a net profit of $9M/year with a P/E of 20. That's pocket change for NVIDIA, and truly negligible considering the PortalPlayer's potential worth in the long-term.
Anyway, it's also important to take into consideration that NVIDIA already has a sound processor in the GoForce 5500. It is, in fact, wonderfully power efficient: barely 20mW for MP3 decoding, compared to PortalPlayer's 60-80mW in the chip they're currently still selling to Apple. Honestly, PortalPlayer's products aren't competitive anymore, and the sell-out is merely based on the fact that they knew they were completely and utterly toast, going forward, at least when it comes to Apple.
Chances are that NVIDIA is designed in the iPhone and PortalPlayer isn't, ironically. The rumoured specifications perfectly match the capabilities of the GoForce 4000 (and yes, I checked ATI's, Broadcom's, etc. offerings, and while some were pretty close, they weren't quite as spot-on). So I could easily see that product being based on a Samsung or Sigmatel application processor and a GoForce 4000. It could be anything else, of course, but at this point I think that's the most likely possibility.
Or, it could be a custom-designed ASIC by NVIDIA or another company that does exactly what Apple needs; when you're talking 12M+ units, such things become much more probable. If Apple wants to do sound processing in the same chip as video processing, then ATI's Imageons might also be a possibility, although few members of that product line match the target resolution. BTW, I think the GoForce 5500's sound engine was created at NVIDIA's Bangalore design center. That stuff used to be the founders' speciality, after all, and they designed sound and GigE ethernet ASICs with a team of ~20 people a few years ago.
So, why did NVIDIA buy PortalPlayer? Market relationships and engineers. It's basically the same reasons NVIDIA bought ULi or Hybrid Graphics. This time around, the sums involves are much larger, but so are the number of engineers and the company's current revenue. Another factor might be that they felt they needed their expertise for SoCs - having licensed ARM11 MPCore doesn't mean they felt they had the best talent to integrate it. There's no way for us to ever know that kind of detail, most likely.
Anyway, when it comes to the 2008 timeframe, which is most likely what you asked me to focus on (
), NVIDIA is focusing their efforts tremendously on the 65nm Low Power process. That process has traditionally been used in notebook and handheld parts, but this time around, they're being a lot more aggressive in using it. TSMC claims their focus on the low power process (which ramped up first, before the general purpose one) is based on customer demand; and I think NVIDIA is one of the several companies that asked for them to take that direction, or at least I wouldn't be surprised if they did. One thing people might not realize is that NVIDIA's DX9 flagship, the 7950GX2, is based on the 90nm Low Power process.So this has some obvious implications when it comes to their PC GPU business, but I'm not going to focus on that here. Instead, it should be fairly obvious that NVIDIA is also targetting 65nm very aggressively for handheld parts, and they'll be depending on it for their next-gen GoForce product. At this point in my point, it should be fairly obvious that this product will sport a ARM11 MPCore processor (I think it'll likely be a dual-core and not a quad-core one given the focus on power consumption, but that's another thing). It'll also sport a GPU shader core that is CUDA compatible and DX10+ level. It could be G9x-based (the shader core, not the entire GPU!), or it could be based on one of their recent handheld patents for an unified programmable shader core.
While G9x-based might seem a tad insane for a handheld part, the shader core's power efficiency really is that good, if you scaled it down to ~1/2 of a cluster. Think of the G7x-based IGP, which had half a quad. And then, unlike current desktop parts, you might (read: you will) be doing things like triangle setup in the shader core. So, I think it's definitely possible in terms of power usage and area, but it remains to be seen whether that's what NVIDIA is going to do. It very likely is not (as I said, they got some interesting handheld patents for unified architectures too), but we'll see about that. Either case, it'll be a true next-gen architecture, with features equal or most likely above those of SGX. I've heard some ex-Gigapixel guys are quite busy at NVIDIA nowadays, so who knows what they have in reserve!
Sound and video are no-brainers, considering NVIDIA already has that in the GoForce 5500, along with a bunch of other stuff. And the power consumption numbers they are claiming already are pretty good AFAIK, so it can only get better on 65nm Low Power. NVIDIA is clearly aiming at getting both the Video iPod contract and the 3G iPhone contract. If they can get the other iPod contracts for the same price, they won't complain, but I don't think it is quite as key to them. Those have quite a fair bit lower margins. I'd be surprised if NVIDIA quit that market completely, however. I think it's relatively likely they'll just integrate an ARM7 with the GoForce 5500's sound processor, manufacture it on 65nm, and that's it.
There are two other contributing factors I didn't touch on but might still be worth mentioning. First, PortalPlayer's Preface. NVIDIA is gaining laptop share rapidly, so they're in an amazing position to bundle Preface with their GPUs. They could try all manners of strange things, like platformizing with Preface/nForce/GeForce/WiFi, although I'm a bit skeptical they'll be doing that yet.
The second factor is the PSP2. I think it's fairly damn obvious NVIDIA has got the contract at this point. So they need to focus a lot more on the 3D part of the handheld market - which means they also need to amortize that R&D somehow. So even if the short-term revenue isn't there for it, they'll get it anyway. Whether indirectly through Apple (since I doubt Apple primarily cares about the 3D functionality, although they'll get it anyway, and might be better off as a result imo), or through Sony. NVIDIA's past handheld offerings have been rather uninteresting in a few ways, but I think they're really beggining to focus a lot more energy on it, and the fight with AMD's comparatively huge handheld division will be interesting, to say the least!
Other competitors will enter the picture, too.BTW, one last thing when it comes to PortalPlayer. It wouldn't surprise me if they were designed in the iPhone (although I see little evidence that it definitely is), but I don't think it's their 5022 or 5024 product lines then. The 5022 is still on 130nm, so that certainly explains part of the power disadvantage they have (which I explained above). If you look at the conference call transcripts for Q1/Q2 2006 with PortalPlayer, you can get a small idea of their strategy there, and they seem very aggressive in getting a few Apple sockets back. Now, I'd speculate that whatever PortalPlayer would be proposing for iPhone would probably be an unannounced product. So that certainly complicates speculating on who got the contract.
What's farly obvious anyway is that the GoForce 5500 is overkill for the rumoured first iPhone model. It might not be for the Video iPod though, but it's rumoured that Samsung won that socket. I'm not sure I buy that, honestly. The fact of the matter remains that analysts are contradicting themselves a LOT on Apple products, and that's for two reasons - first of all, they generally just don't know anything for sure. Secondly, the amounts of money involved with this kind of rumour or "analysis" are enormous, and it's easy for anyone - even analysts - to get overly excited and misinterpret something he heard.
Uttar
The 65-nm, OMAP3s will be using an ARM Cortex-A8 CPU core. How could nVidia be competitive with a hypothetical multicore ARM11 SoC?
A GoForce5500 costs more than a whole OMAP2420, which includes a CPU, high-speed audio DSP, GPU, and an image and video co-processor.
A GoForce5500 costs more than a whole OMAP2420, which includes a CPU, high-speed audio DSP, GPU, and an image and video co-processor.
Last edited by a moderator:
That one's easy.
http://www.arm.com/products/CPUs/ARM_Cortex-A8.html
http://www.arm.com/products/CPUs/ARM11MPCoreMultiprocessor.html
The Cortex-A8 takes ~0.5mW/MHz on 65nm, Speed Optimized. The ARM11 MPCore, when designed with a single core and speed-optimized, takes ~0.43mW/MHz on 90nm. It's smaller, too. In terms of mW/MHz, even the quad-core ARM11 would win against the Cortex, apparently. Now, in terms of Performance/mW, things get a bit more complicated, obviously. And if you take into consideration the ARM11 has an Area Optimized version too, it gets harder to compare.
I won't get into those calculations, but the ARM11 MPCore certainly isn't a clear loser, and it has better granularity when it comes to power consumption. Using two ARM7 cores instead of one ARM9 core is one of the things that made PortalPlayer so successful with Apple, in case you didn't know, because it helped them for power consumption. Whether that same strategy can apply in the future remains to be seen.
The GoForce 5500 also includes all of these things, except the CPU of course, as I'm sure you are aware. I completely agree with you that the OMAP2420 is a superior overall product compared to the GoForce 5500, though...
Uttar
Wow interesting discussion.
When the iPods with Video came out last year, speculation was that it was using an Alphamosaic chip which Broadcom had acquired. Not sure if that was ever confirmed but like a lot of other ASICs and SOCs for the handheld market announced in the last year or two, it was suppose to have all these capabilities.
But are such capabilities being used in products yet? Maybe some feature-loaded smart phones or some of those PDAs from Dell and HP?
Aren't most mobile gaming still 3D? I'm talking about stuff on phones, PDAs, not PSP or NDS.
When the iPods with Video came out last year, speculation was that it was using an Alphamosaic chip which Broadcom had acquired. Not sure if that was ever confirmed but like a lot of other ASICs and SOCs for the handheld market announced in the last year or two, it was suppose to have all these capabilities.
But are such capabilities being used in products yet? Maybe some feature-loaded smart phones or some of those PDAs from Dell and HP?
Aren't most mobile gaming still 3D? I'm talking about stuff on phones, PDAs, not PSP or NDS.
http://www.appleinsider.com/article.php?id=1338
Obviously, the chip doesn't have "ALPHAMOSAIC!" written over it with red ink, but I think that article makes it relatively obvious that these rumours were correct.
AFAIK, PDAs tend to use powerful general-purpose processors instead of special-purpose chips. So instead of using a video decoder, they'll just have a beefier XScale and support it (with non-optimal battery life, you could argue) anyway as a secondary feature. PDAs aren't such a large market nowadays anyway, AFAIK, although still not negligible obviously.
I assume you meant 2D? And yes, GSM-based gaming really hasn't picked up. The standardization and technology avaibility weren't there for it to pick up either, though, I'll admit. But clearly, in the short/mid-term, the PSP and NDS have taken over those markets.
I think that if you look at handheld gaming, the big opportunity for the mobile phone industry is that they're on shorter design cycles (well, they're still long, but that's nothing compared to a console's design cycle!) and innovate progressively, rather than on a step-function basis like in the console market. So around half-way through a handheld console's lifetime, or shortly thereafter, you'd expect the 3G mobile phones to be superior in almost every way.
The question of course is that of content, and even if tons of people had these phones, there is no clear indication that anyone would buy games to play on them. I think the key business strategy that's required there is microtransactions. The mobile phone industry is perfectly well suited to thosem, as the SMS revenue streams have already proved. Still, I think the odds are against the mobile phone industry in terms of gaming.
Keep in mind a company like Apple making its own gaming platform on top of one or several of its products is another thing completely, and I'm not taking that into consideration in my analysis, as I doubt it would benefit other industry players substantially either.
Uttar
Yes I meant 2D of course. But I know a member on another forum who works for ATI doing their mobile embedded stuff. He's been there awhile and he'd talk about the capabilities of their stuff.
Yet most mobile phone games are simple things like bowling right?
I think if they tried to put more powerful silicon and more sophisticated games, they'd run into the same problems people have with using phones as MP3 players: experience is suboptimal and you take away battery life from the primary function of the device, i.e. telephony. You also make the thing bigger and heavier than it needed to be.
I'm curious about how rich mobile devices will be ultimately, unless there's some breakthrough in battery technology. People want these things to be small and sleek and other people want them with features.
You put any kind of high-capacity battery on them and they become too big and heavy. If you want to put Wifi or some other wirless technologies, you better have a killer application for the power it's going to consume. For phones, it's probably the prospect of free or cheap data and voice or cheaper than most mobile plans.
But for gaming devices and MP3 players? If the Zune's wireless features were more attractive that people would use them a lot , then there would probably be dissatisfaction with the battery life. Or more dissatisfaction that is.
Yet most mobile phone games are simple things like bowling right?
I think if they tried to put more powerful silicon and more sophisticated games, they'd run into the same problems people have with using phones as MP3 players: experience is suboptimal and you take away battery life from the primary function of the device, i.e. telephony. You also make the thing bigger and heavier than it needed to be.
I'm curious about how rich mobile devices will be ultimately, unless there's some breakthrough in battery technology. People want these things to be small and sleek and other people want them with features.
You put any kind of high-capacity battery on them and they become too big and heavy. If you want to put Wifi or some other wirless technologies, you better have a killer application for the power it's going to consume. For phones, it's probably the prospect of free or cheap data and voice or cheaper than most mobile plans.
But for gaming devices and MP3 players? If the Zune's wireless features were more attractive that people would use them a lot , then there would probably be dissatisfaction with the battery life. Or more dissatisfaction that is.
For those of a curious nature, can you give the forum?
Although titled "What Would Jen-Hsun Do?: 2007", Alan Dang's speculative article focuses on Apple's motivation for pursuing a combined music/phone device, which he believes is likely to be powered by an NVIDIA system-on-a-chip.
Similar threads
