NVIDIA Reportedly Near Deal To Buy Arm For $40+ Billion Dollars

[oiaohm]

Can't buy this processor though.

Graviton 2

https://en.wikipedia.org/wiki/Annapurna_Labs

Neoverse N1 is the cores in Graviton 2.

https://www.arm.com/products/silicon-ip-cpu/neoverse/neoverse-n1

Yes Neoverse N1 is the A76 design modified for servers. A77 and newer have all those modifications, So there are handful chips out there with Neoverse N1 cores or better.. Like the Snapdragon 800 with A77 cores comes to mind.

[superstructor]

So, have you perhaps tried mainline kernel on these devices?

No, but most Arm64 SBCs do not have mainline kernel support and NVIDIA's kernel source is buildable from public NVIDIA Git repositories.

E.g. Even HardKernel's ODROID series do not run a mainline kernel or until recently the Raspberry Pi couldn't run a mainline kernel.

Just because the kernel is in a different Git repository doesn't make it closed source.

[oiaohm]

No, but most Arm64 SBCs do not have mainline kernel support and NVIDIA's kernel source is buildable from public NVIDIA Git repositories.

E.g. Even HardKernel's ODROID series do not run a mainline kernel or until recently the Raspberry Pi couldn't run a mainline kernel.

Just because the kernel is in a different Git repository doesn't make it closed source.

There are a increasing number of Arm64 SBC that work on mainline kernel. ODROID some of their Arm64 SBC do work with mainline kernel. Work for mainlining some of the chips HardKernel uses started in 2017.

Also google with Android wanting to use mainline kernel has also brought other chip vendors of arm64 socs in the last 12 months into the mainlining process.

Yes just as everything was going in a nice direction Nvidia buys ARM.

[starshipeleven]

See Panfrost.

I was referring to the fact that you just can't boot your firmware in an ARM system if the vendor decides to lock you out using the "secure mode" aka ARM TrustZone to check firmware signatures with the keys burned in the CPU before booting. https://github.com/ARM-software/arm-...are-design.rst

I mean, for most networking embedded none is giving a crap, it's a garbage firmware made by a couple overworked dudes, the most I've seen is bootloader firmware checking but no validation of the bootloader itself so you can hack all you want. After all it's just a firewall protecting a consumer's internal network, it's not important.

But for smartphones for example it's a major pain in the ass and the whole reason behind the "unlocking bootloader" tools that may or may not work depending on how much people look for bugs and exploits (or how benevolent is the vendor in providing official tools to unlock it), and unlocking the bootloader can and will disable functionality permanently (camera enhancements, DRM keys, or more depending on what special sauce they used)

So it would take very little to NVIDIA to go full-asshole mode and require in the license agreement that all the firmware running on their licensed ARM cores and using their licensed and blessed SDK full of (tm) and (c) and (r) stuff must be locked down in a similar manner to "protect their IP" or whatever, and/or require additional cost if you want to make a generic board that can boot anything or the user can use their own keys (i.e. making sure that only a handpicked few expensive boards will exist at all).

EDIT: btw, if you think this is far-fetched, check out what is happening on the server scene, AMD is allowing server vendors to write their own keys in the one-time programmable fuses of their Epyc processors, and guess what is happening? HP and Dell and friends are writing their own key so the onboard PSP (security processor) can validate the firmware on boot, and also bond permanently the CPU to HP or Dell or $vendor hardware. That Epyc processor won't work anymore in a system made by another vendor.
And this is something they actually like/want and are trumpeting around in their PR material as "security feature", so even if AMD did it first, Intel is likely going to add this in their next gen of server CPUs.

[Zgembo]

So it would take very little to NVIDIA to go full-asshole mode...

More like it would take a lot for them to get out of full-asshole mode in the first place, as they are in such mode for quite some time...

[vladpetric]

OK you have a point, some things should be tested first commercially. But i have a feeling that they will be tested in the next 3 years by Nvidia, Nuvia, Apple. Also man no man no, the room shouldn't used for specific purpose circuit that is complex and not up-gradable:

https://spectrum.ieee.org/semiconduc...-to-speed-cpus

Linley Gwennap of the Linley Group, a semiconductor consulting firm based in Mountain View, Calif., says he’s impressed with Jigsaw but cautions that it’s not quite ready for chip-fab prime time. “The problem is generally that a scheme that’s effective on one processor may not be effective on another processor with a different hardware design,” he says. “Every time the processor changes, you have to redo your software, which customers generally don’t like.”

Many processor designs failed exactly for this reason.

I'm not sure what you mean by "the room shouldn't be used for specific purpose circuit".

If you're referring to transistor budgets ... well, when cheap x86 processors have > 5 Billion transistors ...

[tildearrow]

That's moving goal posts. From a user perspective, there is no difference between a cloud server and a server that is somewhere in a server room (in a different building and/or city).

Latency and throughput. Come on, I can't stream 2Gbps of uncompressed 4K material yet.

[tildearrow]

Graviton 2

https://en.wikipedia.org/wiki/Annapurna_Labs

Neoverse N1 is the cores in Graviton 2.

https://www.arm.com/products/silicon-ip-cpu/neoverse/neoverse-n1

Yes Neoverse N1 is the A76 design modified for servers. A77 and newer have all those modifications, So there are handful chips out there with Neoverse N1 cores or better.. Like the Snapdragon 800 with A77 cores comes to mind.

Thank you for the information...

Now I see why....

...it achieved 120 FPS on x264 because it had 64 cores...

[PerformanceExpert]

Thank you for the information...

Now I see why....

...it achieved 120 FPS on x264 because it had 64 cores...

EPYC 7742 gets 156 with 128 threads, so only slightly faster despite using more than twice the power.

[PerformanceExpert]

Latency and throughput. Come on, I can't stream 2Gbps of uncompressed 4K material yet.

All that depends on where the server is located, and not at all whether you own it or not. And if you want to own a Neoverse N1 based server that is faster than Graviton 2, Ampere is happy to sell you one.