Softbank selling ARM to Nvidia, finally after year of speculation...

5150Joker

Supreme [H]ardness
Joined
Aug 1, 2005
Messages
4,357
Yeah, that's the obvious thing (which I brought up in one of these threads, lol). It's the best way to break into the market and stop the cometition. Like AMD licensing/building a GPU with Samsung for ARM. This will be a thing of the past Why fight your way in when you can buy your way in. It's not dumb by any means, and a lot of businesses do this all the time. They can license their IP for a "small" up-charge with the ARM core and get their product into possibly billions of devices over the next decade. Of course, nothing was stopping them from doing this already, like AMD is doing with Samsung, but owning the market makes it much simpler than trying to compete ;). I'm not saying this as good or bad, it's just businesses and how they work.
Yup, the reference ARM designs will be filled with NVIDIA GPU IP, licensees will likely pay a lot more than they do now but they won’t have any choice since most likely the reference NVIDIA version will be far better than a barebone ARM cpu that can have an AMD gpu bolted on to it. Just think, you get millions of Android phones with DLSS capable SoC that can run circles around anything even Apple can make. Google will probably give NVIDIA corporate fellatio and everyone else will fall in line. It will be AMD who loses out and in the long run Intel will be crying too when NVIDIA manages to get Windows for ARM + native PC games (or ported from x86) with top tier “built for GeForce” optimizations that you will not get on x86. Imagine buying a 6090 or whatever in the future and on Windows ARM you get 20-30% higher FPS because nvidia hobbles their stuff on x86. That will be the nail in the coffin for both AMD and Intel. Jensen is a very smart guy, the last of the Steve Jobs types in this industry. That’s why I always laugh when people try to measure Lisa Su against him. Sure she’s an ok leader but she’s not even in the same league as him.
 

erek

Supreme [H]ardness
Joined
Dec 19, 2005
Messages
6,505
Yup, the reference ARM designs will be filled with NVIDIA GPU IP, licensees will likely pay a lot more than they do now but they won’t have any choice since most likely the reference NVIDIA version will be far better than a barebone ARM cpu that can have an AMD gpu bolted on to it. Just think, you get millions of Android phones with DLSS capable SoC that can run circles around anything even Apple can make. Google will probably give NVIDIA corporate fellatio and everyone else will fall in line. It will be AMD who loses out and in the long run Intel will be crying too when NVIDIA manages to get Windows for ARM + native PC games (or ported from x86) with top tier “built for GeForce” optimizations that you will not get on x86. Imagine buying a 6090 or whatever in the future and on Windows ARM you get 20-30% higher FPS because nvidia hobbles their stuff on x86. That will be the nail in the coffin for both AMD and Intel. Jensen is a very smart guy, the last of the Steve Jobs types in this industry. That’s why I always laugh when people try to measure Lisa Su against him. Sure she’s an ok leader but she’s not even in the same league as him.
NVIDIA CEO Jensen Huang Says NVIDIA-Branded CPUs Could be Coming
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,655
Why not?

It takes balls the size of Jensen to actually put the SVE units you designed 3 years before the Neoverse N1 (but didn't include
) into the Neoverse N2.
ARM corporation seems to be allergic to their own concept. (even though every x86 server processor on the planet has native 256-bit vector units)

And Nvidia has the money to afford more targeted high-end ARM development, while getting rid of ARM's overhead of making MALI (and selling it for next-to-nothing with a processor).

NVIDIA's GPU development already pays for itself, so if they can leverage their GPU superiority in all ARM chips, it's going to work like butter. And hopefully, Nvidia will be able to absorb most of the MALI devs into their existing teams.

They could even reuse the MALI brand-name for their previous generation (and continue to give those away just like current ARM).
 
Last edited:

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
Here is one for the conspiracy theorists: this buyout will place arm licensing under US jurisdiction. Today, to use an understatement, is not a happy day for the Chinese tech sector.
Yeah the EU courts ruled that ARM was not included in the Huawei tech ban because it was a UK tech not a US one, that is no longer the case...
 

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
Well I'm happy for NVidia and ARM, it cant be a coincidence that the best ARM chips are the custom ones, Amazon's Gravaton 2 chips are going head to head with AMD and Intel in the server space and Apple's new A14X for most tasks pairs off against an Intel i9-9880H, and you know the Huawei chips land somewhere in there too but they don't matter much at this stage as they have nobody to build them... But the stock ARM chips... not exactly a product leader. Hopefully nVidia can turn that around
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,655
Well I'm happy for NVidia and ARM, it cant be a coincidence that the best ARM chips are the custom ones, Amazon's Gravaton 2 chips are going head to head with AMD and Intel in the server space and Apple's new A14X for most tasks pairs off against an Intel i9-9880H, and you know the Huawei chips land somewhere in there too but they don't matter much at this stage as they have nobody to build them... But the stock ARM chips... not exactly a product leader. Hopefully nVidia can turn that around

Actually, Huawei does the same thing Qualcomm does - they slightly tweak a standard ARM core, and then rebrand the thing (and attach thier own GPU/modem). It's called Semi-custom!

https://www.androidauthority.com/kryo-280-and-semi-custom-arm-cpu-cores-740584/

Samsung is going to join this parade, which means the only custom ARM cores still developed will be NVIDIA's Denver (high-latency, high-throughput), and Apple's A-series (low-latency, high-throughput). Everyone else can't afford the cost of keeping-up with the ARM Corporation, so they've stopped duplicating design efforts.

Even Gravitron 2 is just reusing Neoverse N1. It's just too damn expensive to builds your own custom core anymore (unless you're Apple, and it's the majority of what you spend your R&D on!). The Neoverse will eventually kill-off all ARM custom server processors, but a company like NVIDIA will make sure they do things right.

XGene 3 has been abandoned by Ampere, for semi-custom Neoverse 80-core, and 128-core parts:

https://www.networkworld.com/article/3564514/ampere-announces-128-core-arm-server-processor.html
 
Last edited:

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
Actually, Huawei does the same thing Qualcomm does - they slightly tweak a standard ARM core, and then rebrand the thing (and attach thier own GPU/modem). It's called Semi-custom!

https://www.androidauthority.com/kryo-280-and-semi-custom-arm-cpu-cores-740584/

Samsung is going to join this parade, which means the only custom ARM cores still developed will be NVIDIA's Denver (high-latency, high-throughput), and Apple's A-series (low-latency, high-throughput). Everyone else can't afford the cost of keeping-up with the ARM Corporation, so they've stopped duplicating design efforts.

Even Gravitron 2 is just reusing Neoverse N1. It's just too damn expensive to builds your own custom core anymore (unless you're Apple, and it's the majority of what you spend your R&D on!). The Neoverse will eventually kill-off all ARM custom server processors, but a company like NVIDIA will make sure they do things right.

XGene 3 has been abandoned by Ampere, for semi-custom Neoverse 80-core, and 128-core parts:

https://www.networkworld.com/article/3564514/ampere-announces-128-core-arm-server-processor.html
Well ARM can’t keep up with ARM because they’ve posted losses for the last few years..... but I thought Qualcomm was using straight up stock and Huawei was doing more customizing... I thought the Gravaton 2’s had more tweaking then they apparently do as well so I guess ARM is doing better than I thought. Apple’s implementation is still crushing it though.
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,655
Well ARM can’t keep up with ARM because they’ve posted losses for the last few years..... but I thought Qualcomm was using straight up stock and Huawei was doing more customizing... I thought the Gravaton 2’s had more tweaking then they apparently do as well so I guess ARM is doing better than I thought. Apple’s implementation is still crushing it though.

Right the two reasons ARM was losing money was because the old royalty rates were not high enough to handle these new core developments and aalso the added overhead of develop their first server processor/interconnect.

They fixed the rates a few months ago, so after the cell market recovers, they would be in-the-black,

And hopefully within a few years they will have more than just two ARM server customers using Neoverse...THAT will give them more than enough profit.

This is a growing period for both ARM an d NVIDIA in the server market, sop they are perfect for each-other. ARM will find recession-proof revenue once they become a major player in all types of servers!

The Neoverse N1 was developed on a shoestring budget, and was aimed at cloud servers, but I'm sure Jensen can do better with their offerings going forward. They still got the one design win they were aiming at (Amazon), so they have a good starting-point.
 
Last edited:

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,669
Well I'm happy for NVidia and ARM, it cant be a coincidence that the best ARM chips are the custom ones, Amazon's Gravaton 2 chips are going head to head with AMD and Intel in the server space and Apple's new A14X for most tasks pairs off against an Intel i9-9880H, and you know the Huawei chips land somewhere in there too but they don't matter much at this stage as they have nobody to build them... But the stock ARM chips... not exactly a product leader. Hopefully nVidia can turn that around
It seems you missed the last announcements about Cortex A78 and Cortex X1. Performance not only matches the best custom cores from Apple, but virtually reaches higher clocked Zen2 and Skylake.

https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging/4
 

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,669
Samsung is going to join this parade, which means the only custom ARM cores still developed will be NVIDIA's Denver (high-latency, high-throughput), and Apple's A-series (low-latency, high-throughput). Everyone else can't afford the cost of keeping-up with the ARM Corporation, so they've stopped duplicating design efforts.
Denver was replaced by Denver 2 and this was replaced by Carmel

https://www.anandtech.com/show/13584/nvidia-xavier-agx-hands-on-carmel-and-more

Other two companies that make custom cores are Marwell and Nuvia.
 

Jagger100

Supreme [H]ardness
Joined
Oct 31, 2004
Messages
7,615
ARM is a licensing designer. China will just fork from today or fork each new iteration stripping out proprietary content.

Regardless this deal should not be allowed.
 

juanrga

Pro-Intel / Anti-AMD Just FYI
Joined
Feb 22, 2017
Messages
2,669
And how does that compare to the A14?
https://www.anandtech.com/show/1608...-a14-soc-meagre-upgrades-or-less-power-hungry

Apple here claims a 40% performance boost on the part of the CPUs, although the company doesn’t specify exactly what this metric refers to – is it single-threaded performance? Is it multi-threaded performance? Is it for the large or the small cores?

What we do know though is that it’s in reference to the A12 chipset, and the A13 already had claimed a 20% boost over that generation. Simple arithmetic thus dictates that the A14 would be roughly 16% faster than the A13 if Apple’s performance metric measurements are consistent between generations.
 

Red Falcon

[H]F Junkie
Joined
May 7, 2007
Messages
10,627
APPL isn't going to license stuff from nvidia. They'll buy VIA with cash and make their own x86.
I don't think the x86 license would simply transfer like that, and Apple is all but done with x86-64 CPUs.
Their entire platform is geared around ARM64, and moving forward with any variant of x86-64 would be a massive step backwards.
 

Ready4Dis

2[H]4U
Joined
Nov 4, 2015
Messages
2,059
This would throw a wrench in Apples plans for sure. They highly dislike Nvidia (hence only using AMD cards for such a long time). I don't imagine they'll want to pay nvidia and put their GPU's into everything they sell. Maybe this will be the push to risc-v or some other alternative, lol.
 

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
Looks like the deal is gonna fall through, China needs to sign off on regulatory approval and they are very unlikely to do so. If this deal falls through its gonna be a bad time for ARM and Softbank is probably going to just chop it up and sell it off in parts. So let the patent troll purchasing commence.
 

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
This would throw a wrench in Apples plans for sure. They highly dislike Nvidia (hence only using AMD cards for such a long time). I don't imagine they'll want to pay nvidia and put their GPU's into everything they sell. Maybe this will be the push to risc-v or some other alternative, lol.
Not likely they have not made any mention of stopping their licensing deals just selling a first party product.
 

1_rick

[H]ard|Gawd
Joined
Feb 7, 2017
Messages
1,339
Not likely they have not made any mention of stopping their licensing deals just selling a first party product.
I thought Apple had something like a perpetual license, such that even if Arm stop licensing tomorrow, Apple would be set.
 

Ready4Dis

2[H]4U
Joined
Nov 4, 2015
Messages
2,059
I thought Apple had something like a perpetual license, such that even if Arm stop licensing tomorrow, Apple would be set.
They do, but if there are any major changes or updates, etc they would need a new license. I'm also not sure how much Apple is allowed to change around with their exact license as I'm sure most of us aren't preview to their agreement.
 

Lakados

2[H]4U
Joined
Feb 3, 2014
Messages
2,612
They do, but if there are any major changes or updates, etc they would need a new license. I'm also not sure how much Apple is allowed to change around with their exact license as I'm sure most of us aren't preview to their agreement.
They have a full Architecture license, they are allowed to fully alter the ISA and extend it however they see fit, they are also allowed to alter cores or design their own from scratch.
 

Ready4Dis

2[H]4U
Joined
Nov 4, 2015
Messages
2,059
They have a full Architecture license, they are allowed to fully alter the ISA and extend it however they see fit, they are also allowed to alter cores or design their own from scratch.
I thought that may be the case (seeing as how much they modify already), but didn't want to spread false info if I wasn't positive.
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,655
This huge news deserved a separate thread... and you know it!

exactly. arm adding sve for the first time means N2 is a lot more custom than N1 was. they wouldn't hve risked even more money if the N1 ws not a success,

so NVIDIA is nicely setup to replace all their Denver-compatible architectures soon. They can switch everything over to N2 (and they can make everyone else help fund their invasion of servers, while haviing one early competitir in Amazon)
 
Last edited:

erek

Supreme [H]ardness
Joined
Dec 19, 2005
Messages
6,505
This huge news deserved a separate thread... and you know it!
exactly. arm adding sve for the first time means N2 is a lot more custom than N1 was. they wouldn't hve risked even more money if the N ws not a success,

so NVIDIA is nicely setup to replace all their Denver-compatible architectures soon. They can switch everything over to N2
i don't want to get in trouble, so playing it safe with consolidated threads
 
Top