Nvidia Acquiring ARM for $40 Billion -- Good or Bad for the Industry?

[Marco]

I see the situation more like when 3dfx bought STB, which put them in competition with their customers.  Do you think NVIDIA will resist manipulating the ARM standards to the detriment of their competitors, who are also their customers through ARM?

Lets say they stop selling Tegra as an IC, what remaining customers are competitors? Apple and Fujitsu?

Apple is a problematic monopolist in and of themselves, so I can't really give a shit about them getting into some bother if NVIDIA gets ARM. That leaves Fujitsu. It's a shame, but I'm sure they can find some other core to bolt their vector processing engine on to.

[NorthGuy]

Apple wanted to get independent by moving Macs to an ARM-based CPU of their own making. Microsoft makes PCs with ARM. ARM is ready to march into the PC market. That's where the big money is.

Also, ARM is in every phone.

I don't think MCUs were even a consideration.

[Sal Ammoniac]

When I see a $1 RISC-V in LQFP that I can buy from digikey.. Sign me up, until then ARM all the way

Haven't hit $1 yet, but there have been $5 chips for about three years already:

https://www.mouser.com/ProductDetail/Crowd-Supply/cs-hifiveb-02?qs=Zz7%252BYVVL6bGRLEumsP1bRg%3D%3D

That's the SiFive FE310, which is a toy chip with no internal flash and a tiny set of peripherals (no USB, no Ethernet, no DMA, no ADCs, etc.)

[Sal Ammoniac]

so you can technically squeeze some kind of very minimal Linus on it

Which Linus?

This one?


or this one?

[Bassman59]

A bit like Apple saying (we'll know tomorrow), as of next year no more x86 macbooks will be sold!
Companies like Adobe will have a problem.

Well, they shouldn't, as Apple intends to do what they did when they migrated from PPC to x86 -- a "rosetta" layer will translate x86 to ARM for existing binaries, and software vendors should enable "fat binary" builds which will produce both x86 and ARM code in the same application bundle so the users of both types of processors won't see any difference. And in about five years, they'll deprecate the Rosetta layer and the only people who will complain are those who are still running 10-year-old software.

[David Hess]

I see the situation more like when 3dfx bought STB, which put them in competition with their customers.  Do you think NVIDIA will resist manipulating the ARM standards to the detriment of their competitors, who are also their customers through ARM?

Lets say they stop selling Tegra as an IC, what remaining customers are competitors? Apple and Fujitsu?

But why would they stop production of their own ARM processors?  Didn't they just win some major contracts?

If they did, that would be reassuring, but I suspect their reason to buy ARM is to control the standard in their favor.  They did not need to buy ARM so that they can design their own high performance ARM processors, but owning ARM would be useful to discourage competing processors.

Apple is a problematic monopolist in and of themselves, so I can't really give a shit about them getting into some bother if NVIDIA gets ARM. That leaves Fujitsu. It's a shame, but I'm sure they can find some other core to bolt their vector processing engine on to.

Apple at least has to compete with Android, and now Intel and AMD as well.  If NVIDIA controls ARM, then they ultimately have to compete with MIPS and RISC-V which have tiny shares of the market and none of the phone and personal computer market.

[Marco]

But why would they stop production of their own ARM processors?

To not scare of ARM SOC customers, selling the components as macros for third parties to combine into an IC should be less threatening.

If they did, that would be reassuring, but I suspect their reason to buy ARM is to control the standard in their favor.  They did not need to buy ARM so that they can design their own high performance ARM processors, but owning ARM would be useful to discourage competing processors.

AFAICS they want NVIDIA cores in every ARM SOC. For that to be useful they need to keep ARM SOCs with NVIDIA cores competitive in key markets ... which is really nice, because non Apple ARM cores are increasingly non competitive.

Apple at least has to compete with Android

Android is a phone for poor people, hobbyists and contrarians (I'm the latter, it's irrational). In the most important segments of the mobile phone market Apple increasingly has no competition. If they just had their privacy advantage it would be one thing, but combined with their hardware dominance it's getting dire. Android is dead man walking at the moment.

[0db]

I think it may be "good" for products like the Nvidia Jetson Xavier NX.
At least for their Carmel ARM v8.2 64-bit a 6 cores,  L2 6MB + L3 4MB.

But I am not sure if it's good for the Industry. Perhaps no.

[Kjelt]

Interesting letter from one of the founders of ARM (sorry if this was already posted)

https://savearm.co.uk/

[David Hess]

AFAICS they want NVIDIA cores in every ARM SOC. For that to be useful they need to keep ARM SOCs with NVIDIA cores competitive in key markets ... which is really nice, because non Apple ARM cores are increasingly non competitive.

My past experiences with NVIDIA at a technical level convinced me to avoid them at all costs.  Their thermal design shenanigans which resulted in lawsuits only confirmed my decision to avoid them.

[brucehoult]

Well, they shouldn't, as Apple intends to do what they did when they migrated from PPC to x86 -- a "rosetta" layer will translate x86 to ARM for existing binaries, and software vendors should enable "fat binary" builds which will produce both x86 and ARM code in the same application bundle so the users of both types of processors won't see any difference. And in about five years, they'll deprecate the Rosetta layer and the only people who will complain are those who are still running 10-year-old software.

You say that like it's some kind of weird perversion.

I'm still using the accounting program I bought for $50 in 1991. I've never upgraded it in any way. Why would you? Accounting hasn't changed -- only the GST rate has changed, and that's a setting.

It's a 68000 Mac app.

These days I can run it on Mac, Windows, or Linux (or a phone if I wanted to) probably more than 100x faster than when I bought it.

[nudge]

Initially, it’s a way in for Nvidia to push their GPU and video cores in place of Mali. It’s probably their way of fighting back against the announcement that Samsung will be integrating AMD Radeon GPUs into their next Exynos chips, back in July. For the long term who knows where they’ll head.

[Bassman59]

Well, they shouldn't, as Apple intends to do what they did when they migrated from PPC to x86 -- a "rosetta" layer will translate x86 to ARM for existing binaries, and software vendors should enable "fat binary" builds which will produce both x86 and ARM code in the same application bundle so the users of both types of processors won't see any difference. And in about five years, they'll deprecate the Rosetta layer and the only people who will complain are those who are still running 10-year-old software.

You say that like it's some kind of weird perversion.

I'm still using the accounting program I bought for $50 in 1991. I've never upgraded it in any way. Why would you? Accounting hasn't changed -- only the GST rate has changed, and that's a setting.

It's a 68000 Mac app.

Well, I say that because maintaining a computer of that vintage seems like, well, a weird perversion

I don't even remember what I had for a computer back in 1991.

These days I can run it on Mac, Windows, or Linux (or a phone if I wanted to) probably more than 100x faster than when I bought it.

You have a VM that runs whichever System ran on the Mac back then? Impressive. And nutty.

Next I'll hear that someone has a VM running Apple ][ DOS so they can look at Visicalc spreadsheets.

[0db]

basilisk II is an Open Source 68k Macintosh emulator.

It allows you to run 68k MacOS software on your computer, even if you are using a different operating system. However, you still need a copy of MacOS and a Macintosh ROM image to use Basilisk II. Basilisk II is distributed under the terms of the GNU General Public License (GPL).

I think there is something similar for PPC MacOS.

[Nominal Animal]

I think you are focusing on the wrong aspects.

You need to think more about what would make the most financial sense overall for Nvidia in the short to medium-term timescales, when acquiring ARM.

(The patterns I start to see when I do that are nothing new, and definitely not something I would like to see, considering Nvidia's current business practices, leadership, and similar situations in other companies having similar strategies in the past.  Then again, I'm not a business analyst nor a clairvoyant, so I'll leave this at that.)

[SiliconWizard]

We already discussed that in a previous thread... but yeah, now it's happening.

I don't know what will happen in the next few years. IMHO, it's not unlikely nVidia won't keep ARM for that long anyway, but of course this is just my current thought. Things may evolve.

Given the current market for ARM cores, which is quite large, I don't think it'd be in nVidia's interest to do anything that would drive customers off, especially licensing-wise. They could NEVER compete with the hundreds of companies designing and selling ARM-based processors, so keeping ARM "all for themselves" wouldn't make any sense whatsoever - no way the potential sales (with the associated market disruption) of nVidia processors would make up for the loss of most licenses for ARM cores around the world. Thinking otherwise would be pretty deluded IMO.

So this acquisition is likely not to have any major effect in the short term, all the more that there are not that many alternatives anyway. Don't answer "RISC-V". Beyond that, I don't know.

[Nominal Animal]

Given the current market for ARM cores, which is quite large, I don't think it'd be in nVidia's interest to do anything that would drive customers off, especially licensing-wise. They could NEVER compete with the hundreds of companies designing and selling ARM-based processors, so keeping ARM "all for themselves" wouldn't make any sense whatsoever - no way the potential sales (with the associated market disruption) of nVidia processors would make up for the loss of most licenses for ARM cores around the world. Thinking otherwise would be pretty deluded IMO.

I see you are not at all familiar with Nvidia's current and past business strategies, then.

You might wish to check on how and what graphics cards manufacturers using Nvidia's IP/cores/chips feel about Nvidia, and what kind of relationship Nvidia has with such customers.

[bson]

I think Nvidia views ARM as a potential channel for their own IP, to get their graphics hardware into everything from phones, tablets and TVs to car dash systems.  With ARM they might also see a long-term opportunity to compete with Broadcom, Marvell, Qualcomm for markets where their existing footprint is close to zero.  There also has to be a significant market for small ARM based SoCs with integrated graphics peripherals and touch control that they can license to chip vendors like TI, NXP, ST, etc.  That might be more of an opportunity for ARM than Nvidia though.

[Nominal Animal]

I fear Nvidia sees controlling ARM as controlling a large swath of their potential competitors, especially if they were to fracture the various ARM cores/IP into separate walled gardens with very limited public documentation.

If the acquisition does happen, I expect the first end-developer-visible change would be to MCU peripherals and their documentation.  Register level access would be sparsely documented (individual registers and bits named, but operational requirements or order of operations not stated), and documentation would be centered around Nvidia/ARM-provided Hardware Abstraction Library (which would eventually be littered with specific hardware quirks and duplicated code, as only the code implements the operational requirements that are not documented anywhere that end-developers can access).

This is what Nvidia has always done, and I believe they cannot even imagine doing otherwise.  Disagree?  Look at what they do, not what they say.  CUDA is a good example, I believe.

[coppice]

If the acquisition does happen, I expect the first end-developer-visible change would be to MCU peripherals and their documentation.

What peripherals does ARM currently design and licence to the MCU market? For the A-series chips they make the Mali GPU cores. but those are too big and complex for most MCUs.

[brucehoult]

basilisk II is an Open Source 68k Macintosh emulator.

It allows you to run 68k MacOS software on your computer, even if you are using a different operating system. However, you still need a copy of MacOS and a Macintosh ROM image to use Basilisk II. Basilisk II is distributed under the terms of the GNU General Public License (GPL).

I think there is something similar for PPC MacOS.

I use SheepShaver, which emulates a PowerPC Mac. The PowerPC version of MacOS includes a software emulation of M68k. So my accounting program is undergoing two levels of emulation. It's still far faster than on a real 68k Mac.

I'm running Mac OS 9.1, but actually I could do this with anything up to OSX 10.4.11 "Tiger" from 2007, which contains the "Classic" emulator with Mac OS 9.1, which in turn includes 68k emulation.

[westfw]

I don't think MCUs were even a consideration.

Well, exactly.  And I don't think they will be.  And that's a problem.

[Nominal Animal]

If the acquisition does happen, I expect the first end-developer-visible change would be to MCU peripherals and their documentation.

What peripherals does ARM currently design and licence to the MCU market? For the A-series chips they make the Mali GPU cores. but those are too big and complex for most MCUs.

I'm not sure; I'm extrapolating from Nvidia, not ARM (and using Microchip aquisition of Atmel as a base pattern).  Perhaps instead of peripherals, I should have said "complex internal subsystem details"?  In any case, documentation of the deep details; possibly a tiered system of documentation.  Definitely somehow limited access to full documentation, anyway.

[tszaboo]

When I see a $1 RISC-V in LQFP that I can buy from digikey.. Sign me up, until then ARM all the way

I think the goalpost has been moved. For $1, there are already MCU's with Bluetooth, and built in DC-DC controllers and crypto. If all that RISC-V does is STM32F103 clones, I dont see it getting any traction. The way I see it, the core of the microcontroller is one block in the block diagram. And it is not even the most important one.

I don't think MCUs were even a consideration.

Well, exactly.  And I don't think they will be.  And that's a problem.

I think it was a consideration. I think they might already have a buyer for the MCU segment. I dont think they would keep the R and the M, only the A.

[Jeroen3]

If the acquisition does happen, I expect the first end-developer-visible change would be to MCU peripherals and their documentation.

What peripherals does ARM currently design and licence to the MCU market? For the A-series chips they make the Mali GPU cores. but those are too big and complex for most MCUs.

https://www.eetimes.com/arm-artisan-complete-acquisition/
I suspect you can get quite far with only ARM and Synopsys licenses.
Are there any architects on this forum?