OSDev.org

If it was Apple it might be more newsworthy. But it isn't.

Well, I wish them luck, but I'm not holding my breath. The CPU market is about the most expensive market to break into, and it appears they want to build server processors out of ARM chips or something. Sorry, but no-one will switch architectures just for them. So they are only going to get into new datacenters (if that), not upgrade existing ones.

Yeah, sorry, I really don't see a path to success for these guys. Not a realistic one, anyway.

i would not underestimate Apple's prowess giving how much money they got (+), and article stating how smart engineer they got with years of experience designing low power processor and leverage using their knowledge (+) toward this. But if only market responds to these.

I do remember AMD tried in the past acquiring seamicro and there are probably other efforts too and even reverse efforts (Intel training to get into Mobile and wasted billions) so I dont count on it to be successful.

ggodw000 wrote:i would not underestimate Apple's prowess giving how much money they got (+), and article stating how smart engineer they got with years of experience designing low power processor and leverage using their knowledge (+) toward this. But if only market responds to these.

I do remember AMD tried in the past acquiring seamicro and there are probably other efforts too and even reverse efforts (Intel training to get into Mobile and wasted billions) so I dont count on it to be successful.

There are two problems here: For one, this isn't Apple, this is a bunch of people who used to work there. For two, knowledge of low-power CPU design is unlikely to help with datacenter CPUs where power is rarely an issue. Just as experience designing a sports car engine will likely not help in design a semi-truck engine. You have to optimize for completely different parameters.

You want to design a mobile CPU for as little power use as possible. Latency may be a close second. A server CPU however doesn't care about power use (as much) and is more interested in throughput than latency.

ggodw000 wrote:i would not underestimate Apple's

It's not Apple.

nullplan wrote:Sorry, but no-one will switch architectures just for them.

Normally I would agree, but the market is awaiting server ARM CPUs for years now, so there's definitely a niche here. More than 90% of the servers are running some kind of Linux, and ARM support in Linux is old and mature. Changing the arch in this specific case wouldn't be a big deal.

nullplan wrote:For two, knowledge of low-power CPU design is unlikely to help with datacenter CPUs where power is rarely an issue.

I'm not so sure about that. You have no clue how horribly huge electric bills some company get every month.

nullplan wrote:You want to design a mobile CPU for as little power use as possible. Latency may be a close second. A server CPU however doesn't care about power use (as much) and is more interested in throughput than latency.

I agree, and you're right that power usage often lowered by sacrificing latency on mobiles. Small latency and large thourhgput is much more important for server CPUs, however power consumption is also. Lowering the electric bill means maximizing the profit, something that all IT companies are interested in, which creates a potential market for these guys. Imho ARM itself is perfectly capable to make similar throughput and latency as current server CPUs (maybe even better, considering the simpler instruction decoding and RISC nature). So the real question is how the SoC is manufactured.

I wish them luck, and I hope existing companies won't kill them in their early state.

Cheers,
bzt

iansjack wrote:It's not Apple.

I wonder as

Three of Apple Inc’s former top semiconductor executives in charge of iPhone chips have founded a startup to design processors for data centers, aiming to take on current industry leaders Intel Corp and Advanced Micro Devices Inc.

and Apple is just going to let these guys use their intellectual property. This might eventually turn out to be a crowd funding scam or Apple has got their fingers in the pie somewhere.

They've already got $53 million funding - some fom Dell, none from Apple. I doubt they'll be going for crowd funding. https://www.anandtech.com/show/15115/nu ... cpu-market

I sort of misread however the title can be a bit misleading. I dont expect anything from Dell to be successful.
Of course it could potentially provide some alternatives but I just dont see a need for ARM chips in datacenter.

bzt wrote:Normally I would agree, but the market is awaiting server ARM CPUs for years now, so there's definitely a niche here.

Clearly not, considering that several companies have been shipping ARM server cores and all we've seen out of it is AWS getting ARM instances based on Cortex-A72s, Huawei going "look at us we can make ARM CPUs too", and a few OEMs pumping out their own whitebox ARM servers based on Marvell's reference designs from a year and a half ago.

ggodw000 wrote:I dont expect anything from Dell to be successful.

Dell has been leading in server market share for quite some time. The only company that's punching quite at their level is HPE. Clearly they're doing something right.

Though, that being said, if they put more than a "fine here's some money go away and call us when you have returns" funding round of fifty million into this, then they're going to lose their lead to HPE and they're going to deserve it.

Kazinsal wrote:Clearly not, considering that several companies have been shipping ARM server cores and all we've seen out of it is AWS getting ARM instances based on Cortex-A72s, Huawei going "look at us we can make ARM CPUs too", and a few OEMs pumping out their own whitebox ARM servers based on Marvell's reference designs from a year and a half ago.

There might be some ARM servers already, however there's a lot to consider for a big company. OEM-trust, TCO, manageability, support, part replacement service, etc. Most of today's ARM servers are terrible at those, they are basically really nothing more than "look at us we can make ARM". I know it from personal experience that there's a need for ARM, but currently there's none that could fulfill a big IT company's expectation (not necessarily hardware-wise, although current ARM servers are not as performant as Intel servers and their IO throughput is laughable, but also concerning everything else). And frankly, I don't think that Intel is letting anyone bite out of their server market share easily, specially since they've lost the mobile segment and ARM is trying to break into the desktop segment too (and not succeed only because Win support for ARM is not really mature enough, and not because the hardware manufacturers couldn't keep up).

But if Dell backs these guys, and they really could produce a low-power ARM CPU that is in parallel with (or above the) current Intel chips, that could be a game changer. Imagine how much an ARM server could do using the same power as an Intel based server. I think in order to succeed, they'll need a flagship, for example github switching to ARM or something like that. Without a serious reference, no big IT companies would trust these new servers. We'll see. I still wish them good luck

Cheers,
bzt

Latest on this story is that Apple are now suing the CEO of this startup. https://www.theregister.co.uk/2019/12/0 ... _ceo_suit/

Apple has been in the CPU business, and then pulled their plug on PowerPC in favor of x86... so why should they even want to go back to "rolling their own"?

Reasons are rarely easy to spot from the outside in big business. Sometimes it's literaly just the current fashion which itself is based on flawed reasoning. I remember reading of a "diversification is bad" trend which did a lot of harm for no good reason. One huge company failed due to lots of poor decisions, it was blamed on diversification, and for the next decade investors everywhere refused to allow their companies to diversify.

As for power, it's been a while since I looked, but I'm pretty sure I've seen datacenters offering discounts for cohosting laptops rather than full PCs with low power consumption cited as the reason. I know there are "green datacenters" using low power consumption as a marketing point. ARM instruction decoding takes less power than x86. Mobile CPUs in general may have slower floating point support; how many servers need it?