Given that RISC-V hasn't finalised the vector extensions as of yet I somehow have doubts at this point in time. Then again it would be awfully expensive for a startup to get an Arm architecture license.
The license isn't actually needed though. Until you start manufacturing physical devices, you can design, create, prototype your design even without the ARM license, because you're not officially profitting off the lack of a license.But IANAL.
"That technical data" is usually the IP of the company you worked for, and most companies have legal protections for their proprietary IP. A quick search of the US PTO database shows Mr. Gulati's name on patents issued to Apple. He's the inventor - they are the grantee of the patent. I expect the specific tech his patents were granted to would request ponying up large sums of licensing money. Let's see - IIRC, Apple's lawyers went after companies trying to use the lower case letter "i" as the first letter of a product name followed by all capitals; I can't imagine this would get a free pass.
The very reason this start-up is focusing on data center is the same reason the companies with existing tech will rabidly defend it - it's where the big money is.
As others noted, it will be interesting to follow their progress.
RISC-V is nice, but probably irrelevant point here.
It is meant to compete in ARM/MIPS etc class of cores, not as qualitatively new approach. It is nice that it is open, but that in itself won't bring the advantage that is claimed here...
Yeah seriously, they even convinced people knowledgeable about tech to invest 53 million into their company. Now they are going to trick the big companies that buy server cpus with their marketing. This is some dark magic level marketing at play here. We have to stop them.
Well hopefully it works out for them. A lot of companies have started and then bailed on ARM server cpu's. I don't see any reason ARM servers can't be popular - other than it's a lot of work to design all the hardware and software.
Andrei has mentioned its existence a few times, but nothing more than that appears to exist in the public record. I'm guessing he was probably talking to an ARM person as a journalist on the record, saying something like "ARMv8 is picking up minor revisions every year. Do you ever wish you had the chance for a do-over, to fix it with everything you know now?", the ARM person said "Well of course we are working on a successor, ARMv9" then immediately realized "OMG, I've said too much" and refused to answer any more questions...
ARMv9 was a reference when ARM clears Huawei of using ARM technology and that both its ARMv8 and v9 are British Tech rather than US.
So all we know is ARM has v9 in the making and nothing more, and a notes to its investor that even future ARM design would not have any problems with US security restrictions.
What's needed for a truly revolutionary rethink of CPUs is to change the boundary between HW and SW, in a few places. And I'm fairly confident in saying that's well beyond the capability of a little startup. It needs at least one huge customer, with a full software stack and long-term vision, to partner with a chip maker of at least AMD's size and capability. But, right now, I feel like AMD is getting pulled in too many directions to take on such a project.
The other thing we can say is that China will not be buying these CPUs. So, that means the bulk of their sales would have to go to a half dozen or so cloud operators and hyperscalers. If they don't already have a couple of those big customers lined up, out of the gate, I'd say they're probably on thin ice.
Dell is not going to create markets for them. People don't buy Dell for bleeding edge - they buy Dell for tried-and-true solutions, all nicely packaged up and well-supported. So, the mere fact of having funding from Dell doesn't really do anything to solve their customer situation.
Sorry, the world doesn't agree with your priorities.
Anyway, I'd imagine a couple Chinese GPUs are in the works. So, after Intel launches their Xe GPUs, that'll probably be the next market entrant you can anticipate.
BTW, it'd be interesting if someone scaled up Imagination's GPU architecture to build PC GPUs. I wonder how ARM's off-the-self Mali cores would scale, or Qualcomm's Adreno, for that matter. With Qualcomm's chips now getting into Windows laptops, they basically already have a toe in the water.
The world? I dont see CPU manufacturers popping up here and there and challenging Intel and AMD. Only this one. As consumers, we dont need them. Period. They cant reinvent the wheel. GPUs however are years behind. And Windows is an absolute nightmare since version 10. A lot of the world is agreeing on this.
If you don't see new CPUs popping up here and there, then you don't read this site enough.
The problem with the consumer GPU market is that it has high barriers to entry and it's very mature (meaning products are very refined and feature-rich, so it's hard to find grounds on which to compete). Most importantly, the consensus (and long-term trend) is that it will shrink, while cloud is a growth market.
However, as China is moving to cut ties with foreign suppliers, I expect they'll have desktop GPUs, before too long. Initially, they won't need to be very good, since they can be subsidized. After a while, they'll probably even be competitive.
Interesting! Could be a serious contender if they get something based on ARM v9 out of the gate in the next 24 months. Given the people involved, ARM might just see them as the proof of concept partner they could use for v9 in the server space. A bit surprised that Amazon isn't involved; AWS would be a logical customer/investor.
When a buch of people with really big egos, group together to claim they are going to do something amazing, with really not that much money compared to the already established contenders, and solutions that probably nobody is going to want to learn to program for. Their main asset is their really really huge egos and that idiots will probably run to throw money at them knowing nothing about the market and realistic expectations.
These guys have a track record of delivering. I don't think they would embark on anything they didn't think had a decent chance of succeeding. In particular, Apple's chips are literally & consistently the best - you need a lot more than ego to pull that off.
Likewise, their investors probably aren't noobs. That's less certain, but if you're planning to enter an existing market, I think most VCs are probably rather good at spotting BS and making sure they have a plausible path to profitability.
I say "a billion transistors is a terrible thing to waste." Facebook complained years ago that Intel and AMD CPUs were not efficiently hosting their app. It is possible this startup is trying to re-engineer the silicon to conform more closely to the needs of online apps. The mindless proliferation of cores with the same old APIs just does not efficiently use all of these transistors that are available with our process technologies.
If it is anything really radically new that would justify that "step increase" claim, it will be 5 years before we see anything from them again and 10 years before actual product line on the shelves.
That also means that they have to find their solutions for connection matrix and CPU cores.
I'm more than a bit skeptical at this point, especially before hearing anything concrete about their plan.
Big nuthing burger . Chasing the next shiny object ... something that promises a better toaster that pops out buttered bread. If Arm is so important why didn't Apple and Google pursue it ? Arm is a distraction . IBM has ditched it's Arm and PowerPC platforms too so they can resell Intel boxes. So they are going to blow $60M on a summer project .
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
41 Comments
Back to Article
ChronoReverse - Friday, November 15, 2019 - link
Any possibility of RISC-V?Andrei Frumusanu - Friday, November 15, 2019 - link
Given that RISC-V hasn't finalised the vector extensions as of yet I somehow have doubts at this point in time. Then again it would be awfully expensive for a startup to get an Arm architecture license.lmcd - Friday, November 15, 2019 - link
Could be OpenPOWER or MIPS, particularly the former with all the existing work done on the arch.webdoctors - Friday, November 15, 2019 - link
The license isn't actually needed though. Until you start manufacturing physical devices, you can design, create, prototype your design even without the ARM license, because you're not officially profitting off the lack of a license.But IANAL.zmatt - Friday, November 15, 2019 - link
But without opening up your wallet you aren't getting access to the technical data around the arch.PhilipJ - Saturday, November 16, 2019 - link
Given the people who're behind this company they might not really need to as they've worked with that technical data for a fair few yearslevizx - Saturday, November 16, 2019 - link
So you memorise every single techinical details of all your work?catavalon21 - Sunday, November 17, 2019 - link
"That technical data" is usually the IP of the company you worked for, and most companies have legal protections for their proprietary IP. A quick search of the US PTO database shows Mr. Gulati's name on patents issued to Apple. He's the inventor - they are the grantee of the patent. I expect the specific tech his patents were granted to would request ponying up large sums of licensing money. Let's see - IIRC, Apple's lawyers went after companies trying to use the lower case letter "i" as the first letter of a product name followed by all capitals; I can't imagine this would get a free pass.The very reason this start-up is focusing on data center is the same reason the companies with existing tech will rabidly defend it - it's where the big money is.
As others noted, it will be interesting to follow their progress.
jospoortvliet - Sunday, November 17, 2019 - link
Yes, you do, ARM specifically has a free license to use during the design phase.Brane2 - Monday, November 18, 2019 - link
RISC-V is nice, but probably irrelevant point here.It is meant to compete in ARM/MIPS etc class of cores, not as qualitatively new approach.
It is nice that it is open, but that in itself won't bring the advantage that is claimed here...
satai - Friday, November 15, 2019 - link
What do we know about ARM v9 so far?TristanSDX - Friday, November 15, 2019 - link
buhahahatipoo - Friday, November 15, 2019 - link
Fascinated. If they IPO my eyes will be open.imaheadcase - Friday, November 15, 2019 - link
Former Apple people eh, so good at marketing a product, so can mark it up more.alphasquadron - Friday, November 15, 2019 - link
Yeah seriously, they even convinced people knowledgeable about tech to invest 53 million into their company. Now they are going to trick the big companies that buy server cpus with their marketing. This is some dark magic level marketing at play here. We have to stop them.Alistair - Friday, November 15, 2019 - link
Apple "cpu people" FTFYmode_13h - Saturday, November 16, 2019 - link
????andrewaggb - Friday, November 15, 2019 - link
Well hopefully it works out for them. A lot of companies have started and then bailed on ARM server cpu's. I don't see any reason ARM servers can't be popular - other than it's a lot of work to design all the hardware and software.RallJ - Friday, November 15, 2019 - link
> but given the engineer’s extensive experience with Arm processors I wouldn’t be surprised if it will be an ARMv9 design.It's not.
wishgranter - Friday, November 15, 2019 - link
Then im really curious how it performs vs this https://www.tachyum.com/assets/img/Tachyum%20Hotch...What should come to market next year...
Reflex - Friday, November 15, 2019 - link
It depends on how much vapor Tachyum ultimately produces...tipoo - Friday, November 15, 2019 - link
ARMv9? Is this speculation or do we know anything about a successor to v8?name99 - Saturday, November 16, 2019 - link
Andrei has mentioned its existence a few times, but nothing more than that appears to exist in the public record.I'm guessing he was probably talking to an ARM person as a journalist on the record, saying something like "ARMv8 is picking up minor revisions every year. Do you ever wish you had the chance for a do-over, to fix it with everything you know now?", the ARM person said "Well of course we are working on a successor, ARMv9" then immediately realized "OMG, I've said too much" and refused to answer any more questions...
ksec - Saturday, November 16, 2019 - link
ARMv9 was a reference when ARM clears Huawei of using ARM technology and that both its ARMv8 and v9 are British Tech rather than US.So all we know is ARM has v9 in the making and nothing more, and a notes to its investor that even future ARM design would not have any problems with US security restrictions.
Andrei Frumusanu - Saturday, November 16, 2019 - link
Arm has publicly stated that Matterhorn is a new architecture, so expect v9 silicon in your phone by the end of 2021.name99 - Sunday, November 17, 2019 - link
So A14 or A15?Maybe THAT’s what’s gating the ARM Mac? Irritating to introduce only one generation on ARMv8 then have to support it for 7+ years...
mode_13h - Monday, November 18, 2019 - link
I wonder when ARMv7 support will finally be dropped from ARMv8 chips. Maybe in ARMv9 cores?mode_13h - Saturday, November 16, 2019 - link
What's needed for a truly revolutionary rethink of CPUs is to change the boundary between HW and SW, in a few places. And I'm fairly confident in saying that's well beyond the capability of a little startup. It needs at least one huge customer, with a full software stack and long-term vision, to partner with a chip maker of at least AMD's size and capability. But, right now, I feel like AMD is getting pulled in too many directions to take on such a project.The other thing we can say is that China will not be buying these CPUs. So, that means the bulk of their sales would have to go to a half dozen or so cloud operators and hyperscalers. If they don't already have a couple of those big customers lined up, out of the gate, I'd say they're probably on thin ice.
Dell is not going to create markets for them. People don't buy Dell for bleeding edge - they buy Dell for tried-and-true solutions, all nicely packaged up and well-supported. So, the mere fact of having funding from Dell doesn't really do anything to solve their customer situation.
Beaver M. - Saturday, November 16, 2019 - link
We dont need new CPU manufacturers. We need new GPU manufacturers and a real alternative to Windows.mode_13h - Saturday, November 16, 2019 - link
Sorry, the world doesn't agree with your priorities.Anyway, I'd imagine a couple Chinese GPUs are in the works. So, after Intel launches their Xe GPUs, that'll probably be the next market entrant you can anticipate.
BTW, it'd be interesting if someone scaled up Imagination's GPU architecture to build PC GPUs. I wonder how ARM's off-the-self Mali cores would scale, or Qualcomm's Adreno, for that matter. With Qualcomm's chips now getting into Windows laptops, they basically already have a toe in the water.
Beaver M. - Saturday, November 16, 2019 - link
The world? I dont see CPU manufacturers popping up here and there and challenging Intel and AMD. Only this one. As consumers, we dont need them. Period. They cant reinvent the wheel.GPUs however are years behind. And Windows is an absolute nightmare since version 10. A lot of the world is agreeing on this.
mode_13h - Monday, November 18, 2019 - link
If you don't see new CPUs popping up here and there, then you don't read this site enough.The problem with the consumer GPU market is that it has high barriers to entry and it's very mature (meaning products are very refined and feature-rich, so it's hard to find grounds on which to compete). Most importantly, the consensus (and long-term trend) is that it will shrink, while cloud is a growth market.
However, as China is moving to cut ties with foreign suppliers, I expect they'll have desktop GPUs, before too long. Initially, they won't need to be very good, since they can be subsidized. After a while, they'll probably even be competitive.
eastcoast_pete - Saturday, November 16, 2019 - link
Interesting! Could be a serious contender if they get something based on ARM v9 out of the gate in the next 24 months. Given the people involved, ARM might just see them as the proof of concept partner they could use for v9 in the server space. A bit surprised that Amazon isn't involved; AWS would be a logical customer/investor.mode_13h - Monday, November 18, 2019 - link
ARM has their own server cores, which these guys presumably would be competing against.Amazon already has their own ARM-based server chips, designed in-house (except using off-the-shelf cores, allegedly).
Opencg - Sunday, November 17, 2019 - link
When a buch of people with really big egos, group together to claim they are going to do something amazing, with really not that much money compared to the already established contenders, and solutions that probably nobody is going to want to learn to program for. Their main asset is their really really huge egos and that idiots will probably run to throw money at them knowing nothing about the market and realistic expectations.mode_13h - Monday, November 18, 2019 - link
These guys have a track record of delivering. I don't think they would embark on anything they didn't think had a decent chance of succeeding. In particular, Apple's chips are literally & consistently the best - you need a lot more than ego to pull that off.Likewise, their investors probably aren't noobs. That's less certain, but if you're planning to enter an existing market, I think most VCs are probably rather good at spotting BS and making sure they have a plausible path to profitability.
Qthulu - Sunday, November 17, 2019 - link
when you make a startup with the intent to sell it and retire. The question of competition is usually moot.mode_13h - Monday, November 18, 2019 - link
I'm sure the founders could already retire, if that's really what they wanted to do.Arch Davis - Sunday, November 17, 2019 - link
I say "a billion transistors is a terrible thing to waste."Facebook complained years ago that Intel and AMD CPUs were not efficiently hosting their app.
It is possible this startup is trying to re-engineer the silicon to conform more closely to the needs of online apps. The mindless proliferation of cores with the same old APIs just does not efficiently use all of these transistors that are available with our process technologies.
Brane2 - Monday, November 18, 2019 - link
If it is anything really radically new that would justify that "step increase" claim, it will be 5 years before we see anything from them again and 10 years before actual product line on the shelves.That also means that they have to find their solutions for connection matrix and CPU cores.
I'm more than a bit skeptical at this point, especially before hearing anything concrete about their plan.
fredss - Thursday, May 14, 2020 - link
Big nuthing burger . Chasing the next shiny object ... something that promises a better toaster that pops out buttered bread. If Arm is so important why didn't Apple and Google pursue it ? Arm is a distraction . IBM has ditched it's Arm and PowerPC platforms too so they can resell Intel boxes. So they are going to blow $60M on a summer project .