6 ms·

Qualcomm enters server CPU market with 24-core ARM chip

mtgx 11y ago
I assume it will be using the new Kryo core (or whatever hey are calling it). Hopefully it will be a 14nm chip, too. Things should get interesting for ARM in the server market if that's the case.
jeswin 11y ago
Intel's real problem isn't the power efficiency of ARM CPUs, but rather the entire CPU industry becoming a high-volume, low-margin business. Currently none except Intel (and well AMD) can make chips for the Server/Datacenter business, and they get to keep the healthy margins. If ARM gets into the Datacenter, expect Chinese manufacturers to put out slightly less efficient, but perfectly capable CPUs for a fraction of the cost. The death of the x86 cartel is a good thing for customers.
- dogma1138 11y ago
  AMD is fabless they already are using other manufacturers to build their CPU's and GPU's. Intel is the only one who's currently both designs and manufactures all of their components. As for the Chinese fabs, there aren't that many Chinese manufacturers that can produce an ARM SOC at this moment, the only one of note is MediaTek (And while they are cheap, they aren't that cheap to cause any major issues), you can't just copy a CPU easily, it's not like you can buy a full blown functional reference design from ARM that will be commercially viable you actually need to develop it yourself (especially considering that you need quite a bit of stuff in the SOC that ARM doesn't cover). If you look at the current ARM SOC's even if they are based on the same ARM architecture they are virtually nothing alike. And it's not like Intel didn't had (Acer, SiS, NEC, UMC, IBM, National Semiconductors, NexGen, Transmeta, fuck the 90's to early/mid 2000's were filled with x86 knockoffs) or has competition in the x86 arena from Chinese (And non-Chinese) companies heck VIA is still making cheap X86 CPU's many of them are very low power (watt per watt until the new Core-M stuff some would actually out-perform intel in some cases) the problem is that all of them are kinda crap, and most importantly don't have a well established ecosystem (interconnect-buses, chipsets, memory controllers etc.) and outside of cheap embedded systems (or 20$ 7" "netbooks" from DealExtreme or Alibaba) you won't see them anywhere.
  - petra 11y ago
    The situation now is a bit different: china has decided to encourage switching from intel(due to security reasons), and there are already a few companies working with semi-open IP(ARM/POWER) designing processors for the chinese server market. And i won't be surprised to learn quallcom will be also aiming at that market in some form. And from that base ,competing with intel is a serious possibility.
    - dogma1138 11y ago
      Not really, yes both Russia and China seem to be going for some self-reliance in a form of home-grown CPU's but Quallcom is still an american company so as far as it comes to national interests and national security they are just as much of a liability as Intel. While Quallcom may very well be setting it's sights on competing with Intel doing so because countries do not trust Intel anymore won't be much of an opportunity for them because they'll be painted with the same brush.
  - userbinator 11y ago
    the problem is that all of them are kinda crap, and most importantly don't have a well established ecosystem I'd say performance was the more important factor, as most of the early non-Intel x86 were physically compatible (drop-in replacements) - and many were actually faster than Intel's CPUs, but Intel (and to some extent AMD) quickly caught up and passed them. The P6 (Pentium Pro/II/III) is when Intel really started pulling away from the rest. https://en.wikipedia.org/wiki/List_of_former_IA-32_compatible_processor_manufacturers https://en.wikipedia.org/wiki/List_of_former_IA-32_compatibl... Note: MediaTek is Taiwanese and also fabless. Their phone SoCs are nowhere near as performant as Qualcomm's or Samsung's, and largely viewed as the low-cost "value" option.
    - maaku 11y ago
      Technically it is Intel that competes for the AMD64 market ;)
    - dogma1138 11y ago
- chubot 11y ago
  If it even happens, I don't think it's going to happen that quickly. There is plenty of server software that either only works on x86/x64, or works much better on those platforms. There's still a pretty big software gap to fill: native code from C, Go, and other languages; Java and node.js use a JIT, etc. Then there is a significant amount of software using Intel SIMD, like ffmpeg and so forth. C/C++ is probably the one toolchain that you can rely on to be equally good on ARM or x86 (?). Someone with more expertise might correct me. A big advantage of server software over client software is that you don't have to pay the expense of making it portable. Companies generally don't want to have multiple architectures in the data center. The last mile of stuff takes a long time to change.
pcunite 11y ago
Whoever gets to "1Ghz per user" in the cloud first, wins. Reason: with the future being a socket per user over which the HTML5 interface is streamed, the user experience can be speedy if that same socket has a dedicated core too. I really want 2Ghz per user, but let's not be greedy.
- dogma1138 11y ago
  1Ghz is a meaningless number I assume you base it on current IPC performance of desktop CPU's which is completely irrelevant.
  - pcunite 11y ago
    Just trying to represent time.
    - dogma1138 11y ago
      Still not sure what are you getting at. If your proposed system is basically a time-sharing system which thin-clients connect too then it's performance needs to be similar to what currently individual desktops would need. If all of the processing isn't done on your system (e.g. 3rd party could like Office 365, or in your own desktop e.g. Desktop Office) then why would you need it in the first place?
- username223 11y ago
  > the future being a socket per user over which the HTML5 interface is streamed, That is a future in which I want to open my virtual veins in a warm bathtub. If the future is mediocre and likely surveillance- and ad-supported software shoved through a standard originally designed for formatted text, we have a problem.
rasz_pl 11y ago
24 cores, now compare that to AMDs pitiful attempt at ARM with 2-4 core chips. Its like AMD is actively trying to go belly up.
- dogma1138 11y ago
  Qualcomm has 20-25 years of experience with designing ARM processors over AMD, for all of the reasons why AMD will belly up their attempts at getting into the ARM market isn't one of them.
samfisher83 11y ago
Qcom should really split into two companies. A majority of their profits come from their very high margin IP business. I don't see the point of trying to get into these low margin businesses.
- RyJones 11y ago
  After working there for a couple of years, I describe Qualcomm to outsiders as a flotilla of IP lawyers that happen to employ a handful of engineers.
- bryanlarsen 11y ago
  Those low margin businesses legitimize their IP business. Without that, they'd be a NPE (non-practicing entity), aka a patent troll.
protomyth 11y ago
So, where can one buy a 64-bit ARM motherboard with ECC memory slots at a price around an Intel / AMD offering? I keep seeing these announcements but not the corresponding competition to Intel / AMD motherboards.
- PedroBatista 11y ago
  Nowhere, and that's the problem. This kind of announcements have been going on and on for years, but I can't still get a motherboard where I can plug 2 SATA drives, let alone anything resembling a low-end Intel board.
digi_owl 11y ago
The images of the thing held reminds me of that beast of a CPU that Sun was making at one point.
bestusername111 11y ago
Lol. A single x86 core probably outguns 6 of these cores. I'd rather have a quad core x86 that is actually useful and runs popular software.
- stefantalpalaru 11y ago
  A large number of cores on a single chip is interesting to us programmers not just for the overall performance, but for testing software parallelization strategies. Low per-core performance is actually a good incentive to parallelize as much as possible.

mtgx 11y ago

I assume it will be using the new Kryo core (or whatever hey are calling it). Hopefully it will be a 14nm chip, too. Things should get interesting for ARM in the server market if that's the case.

jeswin 11y ago

Intel's real problem isn't the power efficiency of ARM CPUs, but rather the entire CPU industry becoming a high-volume, low-margin business. Currently none except Intel (and well AMD) can make chips for the Server/Datacenter business, and they get to keep the healthy margins. If ARM gets into the Datacenter, expect Chinese manufacturers to put out slightly less efficient, but perfectly capable CPUs for a fraction of the cost. The death of the x86 cartel is a good thing for customers.

dogma1138 11y ago
AMD is fabless they already are using other manufacturers to build their CPU's and GPU's. Intel is the only one who's currently both designs and manufactures all of their components. As for the Chinese fabs, there aren't that many Chinese manufacturers that can produce an ARM SOC at this moment, the only one of note is MediaTek (And while they are cheap, they aren't that cheap to cause any major issues), you can't just copy a CPU easily, it's not like you can buy a full blown functional reference design from ARM that will be commercially viable you actually need to develop it yourself (especially considering that you need quite a bit of stuff in the SOC that ARM doesn't cover). If you look at the current ARM SOC's even if they are based on the same ARM architecture they are virtually nothing alike. And it's not like Intel didn't had (Acer, SiS, NEC, UMC, IBM, National Semiconductors, NexGen, Transmeta, fuck the 90's to early/mid 2000's were filled with x86 knockoffs) or has competition in the x86 arena from Chinese (And non-Chinese) companies heck VIA is still making cheap X86 CPU's many of them are very low power (watt per watt until the new Core-M stuff some would actually out-perform intel in some cases) the problem is that all of them are kinda crap, and most importantly don't have a well established ecosystem (interconnect-buses, chipsets, memory controllers etc.) and outside of cheap embedded systems (or 20$ 7" "netbooks" from DealExtreme or Alibaba) you won't see them anywhere.
- petra 11y ago
  The situation now is a bit different: china has decided to encourage switching from intel(due to security reasons), and there are already a few companies working with semi-open IP(ARM/POWER) designing processors for the chinese server market. And i won't be surprised to learn quallcom will be also aiming at that market in some form. And from that base ,competing with intel is a serious possibility.
  - dogma1138 11y ago
    Not really, yes both Russia and China seem to be going for some self-reliance in a form of home-grown CPU's but Quallcom is still an american company so as far as it comes to national interests and national security they are just as much of a liability as Intel. While Quallcom may very well be setting it's sights on competing with Intel doing so because countries do not trust Intel anymore won't be much of an opportunity for them because they'll be painted with the same brush.
- userbinator 11y ago
  the problem is that all of them are kinda crap, and most importantly don't have a well established ecosystem I'd say performance was the more important factor, as most of the early non-Intel x86 were physically compatible (drop-in replacements) - and many were actually faster than Intel's CPUs, but Intel (and to some extent AMD) quickly caught up and passed them. The P6 (Pentium Pro/II/III) is when Intel really started pulling away from the rest. https://en.wikipedia.org/wiki/List_of_former_IA-32_compatible_processor_manufacturers https://en.wikipedia.org/wiki/List_of_former_IA-32_compatibl... Note: MediaTek is Taiwanese and also fabless. Their phone SoCs are nowhere near as performant as Qualcomm's or Samsung's, and largely viewed as the low-cost "value" option.
  - maaku 11y ago
    Technically it is Intel that competes for the AMD64 market ;)
  - dogma1138 11y ago
chubot 11y ago
If it even happens, I don't think it's going to happen that quickly. There is plenty of server software that either only works on x86/x64, or works much better on those platforms. There's still a pretty big software gap to fill: native code from C, Go, and other languages; Java and node.js use a JIT, etc. Then there is a significant amount of software using Intel SIMD, like ffmpeg and so forth. C/C++ is probably the one toolchain that you can rely on to be equally good on ARM or x86 (?). Someone with more expertise might correct me. A big advantage of server software over client software is that you don't have to pay the expense of making it portable. Companies generally don't want to have multiple architectures in the data center. The last mile of stuff takes a long time to change.

pcunite 11y ago

Whoever gets to "1Ghz per user" in the cloud first, wins. Reason: with the future being a socket per user over which the HTML5 interface is streamed, the user experience can be speedy if that same socket has a dedicated core too. I really want 2Ghz per user, but let's not be greedy.

dogma1138 11y ago
1Ghz is a meaningless number I assume you base it on current IPC performance of desktop CPU's which is completely irrelevant.
- pcunite 11y ago
  Just trying to represent time.
  - dogma1138 11y ago
    Still not sure what are you getting at. If your proposed system is basically a time-sharing system which thin-clients connect too then it's performance needs to be similar to what currently individual desktops would need. If all of the processing isn't done on your system (e.g. 3rd party could like Office 365, or in your own desktop e.g. Desktop Office) then why would you need it in the first place?
username223 11y ago
> the future being a socket per user over which the HTML5 interface is streamed, That is a future in which I want to open my virtual veins in a warm bathtub. If the future is mediocre and likely surveillance- and ad-supported software shoved through a standard originally designed for formatted text, we have a problem.

rasz_pl 11y ago

24 cores, now compare that to AMDs pitiful attempt at ARM with 2-4 core chips. Its like AMD is actively trying to go belly up.

dogma1138 11y ago
Qualcomm has 20-25 years of experience with designing ARM processors over AMD, for all of the reasons why AMD will belly up their attempts at getting into the ARM market isn't one of them.

samfisher83 11y ago

Qcom should really split into two companies. A majority of their profits come from their very high margin IP business. I don't see the point of trying to get into these low margin businesses.

RyJones 11y ago
After working there for a couple of years, I describe Qualcomm to outsiders as a flotilla of IP lawyers that happen to employ a handful of engineers.
bryanlarsen 11y ago
Those low margin businesses legitimize their IP business. Without that, they'd be a NPE (non-practicing entity), aka a patent troll.

protomyth 11y ago

So, where can one buy a 64-bit ARM motherboard with ECC memory slots at a price around an Intel / AMD offering? I keep seeing these announcements but not the corresponding competition to Intel / AMD motherboards.

PedroBatista 11y ago
Nowhere, and that's the problem. This kind of announcements have been going on and on for years, but I can't still get a motherboard where I can plug 2 SATA drives, let alone anything resembling a low-end Intel board.

digi_owl 11y ago

The images of the thing held reminds me of that beast of a CPU that Sun was making at one point.

bestusername111 11y ago

Lol. A single x86 core probably outguns 6 of these cores. I'd rather have a quad core x86 that is actually useful and runs popular software.

stefantalpalaru 11y ago
A large number of cores on a single chip is interesting to us programmers not just for the overall performance, but for testing software parallelization strategies. Low per-core performance is actually a good incentive to parallelize as much as possible.