NUVIA says they have an ARM processor for servers in the works that's way better than everyone else

  • Thread starter Deleted member 89018
  • Start date
D

Deleted member 89018

Guest
Basically: some former Apple and Google processor engineers decided to found their own company to do ARM processors for the server space.

Anandtech decided to post their latest ad as a news article.

The company claims that their processors will be "+50% to +100% peak performance of the other cores, either for the same power as other Arm cores or for a third of the power of x86 cores." Then they shared a single benchmark based on a simulation of actual performance, leaving out the top part of it for some reason.

https://www.anandtech.com/show/1596...0-st-performance-over-zen-2-for-only-33-power

I like seeing ARM showing up in more spaces, but this felt so much like an ad designed to drum up investment capital for the "next big thing". I know simulations are a thing but why do a simulation of your OMGWTFast processor, telling everyone how fast it will be, and then leave out the part of the simulated benchmark that's supposed to be so awesome?

TL;DR: more ARM. Discuss.
 
Well, at least we now know they're not going to try the pointless RiscV server processor route. Processor architecture was the only unresolved rumor about the new offshoot of Apple.

As I've said before, ARM already won the RISC server invasion 2.0. It's been established for 6 years now. The fact that ARM Inc felt there was enough demand to design Neoverse N1 (used by Amazon's Gravitron2) last year is a testament to how far they've already gone.

https://aws.amazon.com/ec2/graviton/

It would take a miracle of cash plus chip design prowess for RiscV to catch-up with that much momentum.
 
Last edited:
I read in Maximum PC that x86 is long overdue and ARM has a really good chance taking over X86 which was suppose to be dead like 15 years ago.
 
Comixbooks said:
I read in Maximum PC that x86 is long overdue and ARM has a really good chance taking over X86 which was suppose to be dead like 15 years ago.
Click to expand...

It pretty much already has. Most peoples personal computing isn't done on x86 anymore. Their work computing is their personal not so much. Super computers outside of the ones contracted by the US (who will prop intel and x86 up for another 20 years if need be) are moving on. The fastest machine right now is ARM powered... Cray is actively selling ARM servers, and the Chinese are doing their own thing. In a couple years Apple will have switched away from x86... and we will all be surprised how many creator types switch with them. In the server market, the market has been looking for lower powered ARM options for a few years.... but now it seems to be lining up. There are multiple ARM server plays in motion right now, the software stacks are mostly where they need to be for the switch... and the hardware is starting to look to actually perform on par not just good enough at 40% less power. (which for a lot of applications is more then enough to sell customers on already)

Yes ARM servers have been hyped for years now... but it seems like things are starting to fall into place for ARM servers to start being a real threat to Intel, and a lesser extent AMD. AMD is in a perhaps slightly better position imo, as they have already done more then just initial work on Server class ARM chips... they never brought K12 to market. I suspect however that if in 2 years from now it makes sense they could have top tier class ARM server chips shipping in a quarter... where as Intel would likely never eat the humble pie required to even start such a project.

Next couple years will be very interesting in the Chip world. Yes it was x86s time over a decade ago already... Intel did a very very good job of shutting down all competition and making it impossible for anyone to really compete. Even if you could fund a few years of development without product to sell.... who exactly where you going to sell it too anyway. Transmeta was I think the last real x86 startup to learn that lesson. They produced I would say the last real different take x86 CPUs that could have been the basis of what powered every phone and laptop in the world right now... but Intel starved them. When they had working product to ship most OEMS didn't want to piss off Intel and the few that gave Transmeta contracts did so for super budget devices that made transmeta look terrible. After that one.... most big name chip designers had no desire to try anything start up wise, and investors where a lot less likely to bankroll 2-3 year development phases. Things have changed now... ARM chips of all stripes have markets to sell into many of which Intel (or any other super player) have much less or no control at all over. Investors are starting to take a few risks again on chip startups, and engineers seem to be willing to leave their decent paying jobs with companies like Apple. Should be fun to watch.
 
Energy costs are going up, and processing capacity keeps increasing. Most hosted server VM’s aren’t running 100% all day every day they are running 12-30% if you can take that load from 12w to 3w it’s a massive win on the energy bills. Arm is becoming a better data centre solution every year. Getting in now is a good plan on their part and I wish them luck.
 
No mention of when these will be available and the benches are simulated, so, 1-2 years till they hit? In any case I hope the come out swinging and deliver.
 
Oh look, another company that has an amazing product that'll totally beat up all the other guys... but you can't see it yet, just take their word for it that it'll be out Real Soon Now(tm) and will be as good as they claim!
 
Verge said:
bad time to start a server company
Click to expand...


Why? Work-from-home has meant BOTH notebooks (for the user) and servers (for the backend) have seen unprecedented demand. It's going to remain for at least next 2 years (we're 18 months from an effective vaccine hitting enough people in this country to cure us).

The only reason we have a downturn at all is because entertainment/hospitality and education/childcare have been severely impacted. And that's a good 1/3 of this country's employment.
 
x86 isn't going anywhere anytime soon. Eventually? Yes, but not within the next 10 years.
 
I agree with most in here. Paper launches are a waste of time.
Show me your product. If its got the goods: proove it. Put up or shut up.
 
Meeho said:
PC is dead, X86 is dead, it's the year of Linux desktop, next gen consoles are faster than PCs, Star Citizen is coming this year...
Click to expand...

And Duke Nukem Forever will come out......wait.........that actually finally happened.
 
deruberhanyok said:
Basically: some former Apple and Google processor engineers decided to found their own company to do ARM processors for the server space.
Click to expand...

Some Apple and Google engineers is a loose description. There are very smart people in that company, including the chief engineer behind Cyclone, Typhoon, Twister, Hurricane, Monsoon, Vortex and Lightning architectures.

Original source for NUVIA performance claims

https://nuviainc.com/blog/performancedeliveredanewway
 
Last edited:
ChadD said:
Yes ARM servers have been hyped for years now... but it seems like things are starting to fall into place for ARM servers to start being a real threat to Intel, and a lesser extent AMD. AMD is in a perhaps slightly better position imo, as they have already done more then just initial work on Server class ARM chips... they never brought K12 to market. I suspect however that if in 2 years from now it makes sense they could have top tier class ARM server chips shipping in a quarter... where as Intel would likely never eat the humble pie required to even start such a project.
Click to expand...

How so if the lead engineer left without finishing the project and the remaining engineers are all x86 guys?
 
whateverer said:
Why? Work-from-home has meant BOTH notebooks (for the user) and servers (for the backend) have seen unprecedented demand. It's going to remain for at least next 2 years (we're 18 months from an effective vaccine hitting enough people in this country to cure us).

The only reason we have a downturn at all is because entertainment/hospitality and education/childcare have been severely impacted. And that's a good 1/3 of this country's employment.
Click to expand...

The only way you have server demand numbers is if you work for microsoft/amazon/google.
 
rgMekanic said:
HadMeInTheFirstHalf.jpg haha

Headline:
Nuvia Says "PLEASE BUY OUR STOCKS FOR THE LOVE OF GOD!!!"
Click to expand...

They don't have stock. They are still in startup mode. Though they may be fishing for another private funding round.

Given the talent they have, it seems likely that they will have one of the best ARM server chips out there, if not the best.

IIRC, they have the Apple chip lead (that created the best Mobile ARM chips that exist), and his team wanted to do a kick ass server chip for Apple, and it was rejected ... So he left.
 
juanrga said:
How so if the lead engineer left without finishing the project and the remaining engineers are all x86 guys?
Click to expand...

ISA isn't specific to an engineer. There is no such thing as x86 or ARM guys. K12 was also completed.... and I can assure you it was not completed by one employee alone. They where called sister chips for a reason. they shared all logical design components.
Also don't forget AMD has already shipped on ARM server chip the opteron a1100s. No they where not sexy being A57 cores... but they had everything required for low power ARM servers. They are almost 5 years old now and where a bit ahead of the software was the main reason for their not being commercial hits imo. Anyway my point is for AMD to resurect a large scale project that had shipping server product 4-5 years later where 80%+ of the engineers responsible are still with the company... is far easier then having to start from scratch.

On the AMD front we may see some interesting ARM news out of them next year no matter. Samsung will be bringing RDNA 2 to phones (perhaps not just phones) in 2021... and rumors are the early 5nm sample silicon from Samsung is for some reason named Ryzen C7. Which seems really odd if its just rocking 4 RDNA2 cores as reported. Reports are also that this chip has 2 of the new ARM Cortex-X1 out of order high end cores running at 3ghz with a couple a78 and the standard 4 low power a55 cores. Reading those rumors I can't help but think... that is perfect hardware to put in a PC wanting to compete with an Apple silicon macbook.
 
ChadD said:
ISA isn't specific to an engineer. There is no such thing as x86 or ARM guys. K12 was also completed.... and I can assure you it was not completed by one employee alone. They where called sister chips for a reason. they shared all logical design components.
Also don't forget AMD has already shipped on ARM server chip the opteron a1100s. No they where not sexy being A57 cores... but they had everything required for low power ARM servers. They are almost 5 years old now and where a bit ahead of the software was the main reason for their not being commercial hits imo. Anyway my point is for AMD to resurect a large scale project that had shipping server product 4-5 years later where 80%+ of the engineers responsible are still with the company... is far easier then having to start from scratch.
Click to expand...

Keller joined AMD because no one at AMD was familiar with ARM. He leaded a small group of x86 engineers when working in K12 and Skybridge. He even told us an anecdote about how the ARM ISA left a "little daunted" to his team.

K12 and Zen were called sister because they were designed to be socket-compatible as part of the Skybridge project, but the cores were different (there is a lot of difference between ARM and x86 ISAs). Moreover, the rumor states that K12 had 3 AGUs.

Keller left AMD when K12 was cancelled by Lisa Su, who formely cancelled Skybridge.

The Opteron A-series used microarchitecture from ARM engineers. There are a lot of difference between licensing standard cores from ARM and designing your own custom cores. Moreover, the whole Opreton A-series approach was a fiasco: the AMD chip was about two years late and missing the original target.
 
juanrga said:
Keller joined AMD because no one at AMD was familiar with ARM. He leaded a small group of x86 engineers when working in K12 and Skybridge. He even told us an anecdote about how the ARM ISA left a "little daunted" to his team.

K12 and Zen were called sister because they were designed to be socket-compatible as part of the Skybridge project, but the cores were different (there is a lot of difference between ARM and x86 ISAs). Moreover, the rumor states that K12 had 3 AGUs.

Keller left AMD when K12 was cancelled by Lisa Su, who formely cancelled Skybridge.

The Opteron A-series used microarchitecture from ARM engineers. There are a lot of difference between licensing standard cores from ARM and designing your own custom cores. Moreover, the whole Opreton A-series approach was a fiasco: the AMD chip was about two years late and missing the original target.
Click to expand...


Right, they're just reusing the same pinouts. - it's still up to the ARM engineers to write firmware to interact with that south-bridge.

This isn't the first pin-compatible microprocessor to launch - that honor goes to the MOS 6501. It was pin-compatible with the Motorola 6800 (which means you could,with a little elbow-grease, pop it into existing motherboards.)

Of course Motorola sued them over such an atrocity, and forced the market to adopt the incompatible-pinout iof the 6502.

Just because you have identical pinouts doesn't mean your chips suddenly run the same Instruction Set Architecture, but it freaked-out Motorola enough to make it disappear.

Skybridge was actually completely pointless - it was just taking that already-old A57 baseline, and adding an AMD GPU on-top.The fact that it reused the same socket didn't suddenly make this a smart investment of AMD;'s time.

The only company to find success doing this was NVIDIA (but the Switch design win only came years after wasting money developing the Tegra gaming platform.) The Tegra was only a success because of car design-wins.
 
Last edited:
[Spectre] said:
Our servers are getting hammered and I don't work for any of those. We have been scrambling the last few months to deploy more assets to cope with the load.
Click to expand...

They getting hammered with arm workloads?

Maybe the server glut is over, but when all these mid-medium sized corporations migrated to the cloud with their underutilized workloads, you couldn't sell a server to anyone. Or maybe they are coming back home. We had some GPU workloads in azure we moved back to on prem because it's an absolute ripoff.
 
You must log in or register to reply here.
Back
Top