What do you mean going to? They already are in those areas and have been for years. That’s why NVidia is dropping $40 billion on them.
Here’s a solid writeup on AnandTech about it. Reasons why Nvidia would do this and potential issues for them to reassure customers and government regulators this is a good deal. The companies expect this to take up to 18 months to fully close so its going to be a long process.
https://www.anandtech.com/show/16080/nvidia-to-acquire-arm-for-40-billion
It’s a bit more complicated than that. You can either license the ARM architecture or license specific chip designs.
Apple licenses the ARM architecture and completely designs its own compatible chips, just like AMD has the x86 license and makes intel-compatible CPUs.
Qualcomm and the rest do not license the architecture, they license each individual chip. The Qualcomm Snapdragon 865 has ARM Cortex A77 and A55 chips inside the SoC.
Now the details of how ARM does these licenses is secret. They certainly offer a perpetual architecture license. Do they offer perpetual chip licenses too? So they’re obligated to give Qualcomm the Cortex A79 or whatever design? Beats me, but my guess is Qualcomm, Samsung, Rockchip, etc, are not perpetually licensed.
That said, my guess is the biggest change out of this is that cheap Android phones will have better GPUs based on GeForce tech rather than ARM Mali, which was never very good. If you don’t buy cheap phones it won’t impact you at all because Qualcomm designs its own GPUs, the Adreno line.
Nvidia is also making a pretty large data center push in their business story, and ARM has several efforts to bring ARM servers into the data centers (AWS recently brought a bunch of rentable units online). So I think a large part of the $40B rationale is to solidify their data center position, as well as offering good combination of power efficient ARM servers used for ML with their DC GPUs.
Perhaps that is a justification, but ARM will never succeed in the datacenter because everybody still codes on intel. CPU power utilization in ML is negligible compared to GPU.
AWS seems to disagree.
Oh, golly, yes, they think their solution is fantastic.
Have to see how many companies actually use those machines over standard Intel ones.
If you’re doing pure compute they can make sense in a cost/performance scenario because they sip power. They also make a lot of sense for tasks with limited CPU requirements, like serving static content or GPU-offloaded machine learning. But still, they are deeply unpopular.
The fact of the matter is everybody codes on Intel Linux, it’s a pain in the ass to cross-compile everything, and ARM servers aren’t sufficiently cheaper to run to really be worth the trouble. ARM has been trying to make their stuff work in the DC for years now and it failed to happen for that very simple reason-- even though the dollars and cents make a case, nobody wants to use them.
We’ve looked at it internally. We can cross compile both Go and .net core csharp to ARM. Then we get to find all the ARM only bugs plus the regular bugs! And our developers can’t run their code on the same configuration without going into the cloud.
I wonder if ARM Macs will change the level of interest at all!
Probably not, unless people start running Linux on those Macs.
My company (I think we’re approaching 15K employees) is mostly running MacBook Pros for enterprise Java development. It’s not uncommon in California, but I recognize that we’re in our own bubble in this state.
I’ll be curious how the Apple silicon ones fair for that use case, but I don’t imagine we’ll be seeing the Pro systems running those chips until later next year.
ARM was owned by a Japanese company before, and Nvidia is HQ in the USA. It’s unclear what national security considerations exist, if any.
There’s certainly potential to harm the open marketplace, if Nvidia (for example) decided to stop licensing ARM chips or instructions. But national security-- no.
The first national security review is in and the UK didn’t like what they’re seeing. They’ve ordered a much deeper national security review.
Looks like the CMA is carrying water for Amazon (Graviton), intel and AMD here.
I mean, does anyone seriously think anyone other than NVidia or Apple could play in the datacenter space with those incumbents? (And Apple doesn’t appear to want to).
I’d be a lot more concerned about NICs and mobile and so forth, but I don’t think NVidia really wants to play there - and that could be fixed by undertakings. Which is why the CMA is focussing on datacenter. But what are NVidiarm going to do, try and screw Amazon on the license? That would be a … bold play.
FTC says “Not so fast!”
The amount of bullshit being talked about this merger is ridiculous. Not that there aren’t legitimate concerns, but anything that means more competition for Apple, Amazon, intel and AMD has a lot to recommend it.
The truth is ARMs designs have fallen behind at the high end. NVidia want to have a crack at arm64 in the datacenter and get the engineering teams, they might want to try and compete with AMD in the gaming SoC space. And that’s all good.
I can’t imagine them caring too much about mobile CPUs (might be wrong here), but they almost certainly have plans for the edge / IoT space, and those plans are probably evil and anticompetitive, this being NVidia.
The media articles are all so worthless though.
EDIT: Hmm, apparently (and much to my surprise) the Graviton chips are pretty close to reference ARM core designs. I am very confused about the quality of ARMs designs to be honest.
Nvidia is looking to get out
Maybe more accurately, NVIDIA giving up hope on getting in?
Yeah, they’re accepting reality.