5 ms·

Because it was literally failing in the market. IBM PAID global foundries $1.5 BILLION to take the chip business from them. Now they're giving away the IP in

by tw04 7y ago

Because it was literally failing in the market. IBM PAID global foundries $1.5 BILLION to take the chip business from them. Now they're giving away the IP in hopes that someone will actually start making chips from it to drive demand for Redhat. If you think Power was a massive success you must be waiting on the edge of your seat for Itanium to finally reach global dominance too. https://www.bizjournals.com/albany/morning_call/2015/01/analysts-expect-approval-of-globalfoundries-deal.html https://www.bizjournals.com/albany/morning_call/2015/01/anal...

justinjlynn 7y ago
I didn't say it was a massive success - but it still did pretty darn well against the x86 juggernaut; and now it's doing something that juggernaut can never do, going open. This will enable so many application tailored designs that IBM itself could never do viably. We'll see how things go - I don't make predictions, but I do have hope for a more open, more hacker friendly future.
- Aloha 7y ago
  Beyond that, IBM uses it for several of their mainframe products too.
  - StillBored 7y ago
    This is false, the zXX microprocessors are custom cores that natively run the z/Architecture ISA. Even a trivial comparison will show they are different and not just a microcode layer running on a POWER processor. There is probably some level of sharing (say a high perf divide unit or whatever) but that doesn't make them the same. https://en.wikipedia.org/wiki/IBM_z14_(microprocessor) https://en.wikipedia.org/wiki/IBM_z14_(microprocessor) Edit, Adding some microarch descriptions links below: https://www.anandtech.com/show/11750/hot-chips-ibms-next-generation-z14-cpu-mainframe-live-blog-5pm-pt-12am-utc https://www.anandtech.com/show/11750/hot-chips-ibms-next-gen... https://en.wikichip.org/wiki/ibm/microarchitectures/power9 https://en.wikichip.org/wiki/ibm/microarchitectures/power9
    - justinjlynn 7y ago
      They are custom cores, but they likely do share at least some RTL - we won't know, of course, unless IBM decide to share those details and I doubt they would.
    - lboc 7y ago
      ISTR reading somewhere they're used in the I/O hardware (OSA etc)?
      - StillBored 7y ago
        Yah, there are embedded power processors scattered all over IBM's hardware lines. The z tape controller a couple years ago was just a rack mount POWER server. OTOH, so is x86, the z's HMC, PE's etc are just x86 laptops/servers. But thats like calling an x86 server an ARM because there are ARM's in the disk controllers, and on the motherboard acting as BMC's/etc.
        lboc 7y ago
        Agreed. I just read "...uses it..." to include things other than the main processors. (edit: misquote)
        rbanffy 7y ago
        And mainframes rely on them extensively. Pretty much everything that can be delegated is, leaving the z cores running as little code as possible.
    - protomyth 7y ago
      Their other mainframe product the iSeries is POWER-based. We just bought a new one and it has a POWER 9 CPU.
      - StillBored 7y ago
        iseries is not a mainframe (and you will never hear an IBMer make that mistake). That's whats traditionally called a minicomputer. Sure they make some big ones, but people don't go around calling a superdome flex a mainframe either despite it being a massive machine (something like 1k cores and 48TB of ram in its max config). Its just a fancy x86.
- StillBored 7y ago
  something that (x86) juggernaut can never do, going open Why do you think that? Sure there is some level of cross license/etc, but x86's have been designed/built by a number of 3rd party companies (IDT, rise, cyrix, transmeta, etc) in the past, and basic versions that skip over the currently IP protected bits could probably be done again. x86-64 is nearly 20 years old at this point so patents there will start to expire soon.
  - justinjlynn 7y ago
    You're right. I shouldn't say never. It might, one day, be able to do it - and I hope it does; but I'm not going to hold my breath waiting. The fact that no unified entity owns x86 or it's multitudinous extensions makes going open almost impossible - so, so many independent interests and conflicting businesses would need to align.
kev009 7y ago
And they got the highest performance 14nm FinFET process on the market for P9 and z14 while clearing the way to compete against fabless AMD and ARM vendors with P10. If anything, intel looks stupid here after falling behind in fab tech and causing immense turmoil for their external fab customers. A trivial review of IBM's systems revenue versus the datacenter breakout of any other CPU maker earnings reports will net you: yes, it is indeed quite successful.
- wmf 7y ago
  Because Power is on such a slow (and getting slower) cycle, it seems like the best case for Power10 is for it to be ahead of Epyc for one year then be behind for 4-5 years. And the worst case is that Power10 will already be behind Epyc when it is released. Revenue is a different story that doesn't necessarily reflect technical merit.
  - justinjlynn 7y ago
    > Power is on such a slow (and getting slower) cycle... so, how are we judging cycles here? I seriously doubt Intel and AMD use the same definition that IBM do. Let's just say between IBM's cycles: - new PCI Express subsystem revision (first to market with PCI Express 4.0 in P9... who knows with P10) - new memory interface architecture (Centaur and DDR4 direct attach w/ P9 and now OMI with P10) - new coherent processor external accelerator bus (OpenCAPI and CAPI...) - new internal interrupt accelerator function architecutre (XIVE w/ P9) - new privilege level for hypervisor/guest separation (Ultravisor/PEF in P9+/P10) ... and so on. Intel and AMD do NOT do that. Let's compare apples and apples here instead of using ill-defined words like 'cycles'.
    - wmf 7y ago
      Looking at the roadmap, IBM released nothing in 2011, 2013, 2015, and 2019. The number of cores has been stagnant from 2014 to 2020.
      - justinjlynn 7y ago
microcolonel 7y ago
> Now they're giving away the IP in hopes that someone will actually start making chips from it to drive demand for Redhat. And probably it'll be RISC-V at the end of the day anyway, and it'll sell Redhat just fine. ;- )

by tw04 7y ago

justinjlynn 7y ago

I didn't say it was a massive success - but it still did pretty darn well against the x86 juggernaut; and now it's doing something that juggernaut can never do, going open. This will enable so many application tailored designs that IBM itself could never do viably. We'll see how things go - I don't make predictions, but I do have hope for a more open, more hacker friendly future.

Aloha 7y ago
Beyond that, IBM uses it for several of their mainframe products too.
- StillBored 7y ago
  This is false, the zXX microprocessors are custom cores that natively run the z/Architecture ISA. Even a trivial comparison will show they are different and not just a microcode layer running on a POWER processor. There is probably some level of sharing (say a high perf divide unit or whatever) but that doesn't make them the same. https://en.wikipedia.org/wiki/IBM_z14_(microprocessor) https://en.wikipedia.org/wiki/IBM_z14_(microprocessor) Edit, Adding some microarch descriptions links below: https://www.anandtech.com/show/11750/hot-chips-ibms-next-generation-z14-cpu-mainframe-live-blog-5pm-pt-12am-utc https://www.anandtech.com/show/11750/hot-chips-ibms-next-gen... https://en.wikichip.org/wiki/ibm/microarchitectures/power9 https://en.wikichip.org/wiki/ibm/microarchitectures/power9
  - justinjlynn 7y ago
    They are custom cores, but they likely do share at least some RTL - we won't know, of course, unless IBM decide to share those details and I doubt they would.
  - lboc 7y ago
    ISTR reading somewhere they're used in the I/O hardware (OSA etc)?
    - StillBored 7y ago
      Yah, there are embedded power processors scattered all over IBM's hardware lines. The z tape controller a couple years ago was just a rack mount POWER server. OTOH, so is x86, the z's HMC, PE's etc are just x86 laptops/servers. But thats like calling an x86 server an ARM because there are ARM's in the disk controllers, and on the motherboard acting as BMC's/etc.
      - lboc 7y ago
        Agreed. I just read "...uses it..." to include things other than the main processors. (edit: misquote)
      - rbanffy 7y ago
        And mainframes rely on them extensively. Pretty much everything that can be delegated is, leaving the z cores running as little code as possible.
  - protomyth 7y ago
    Their other mainframe product the iSeries is POWER-based. We just bought a new one and it has a POWER 9 CPU.
    - StillBored 7y ago
      iseries is not a mainframe (and you will never hear an IBMer make that mistake). That's whats traditionally called a minicomputer. Sure they make some big ones, but people don't go around calling a superdome flex a mainframe either despite it being a massive machine (something like 1k cores and 48TB of ram in its max config). Its just a fancy x86.
StillBored 7y ago
something that (x86) juggernaut can never do, going open Why do you think that? Sure there is some level of cross license/etc, but x86's have been designed/built by a number of 3rd party companies (IDT, rise, cyrix, transmeta, etc) in the past, and basic versions that skip over the currently IP protected bits could probably be done again. x86-64 is nearly 20 years old at this point so patents there will start to expire soon.
- justinjlynn 7y ago
  You're right. I shouldn't say never. It might, one day, be able to do it - and I hope it does; but I'm not going to hold my breath waiting. The fact that no unified entity owns x86 or it's multitudinous extensions makes going open almost impossible - so, so many independent interests and conflicting businesses would need to align.

kev009 7y ago

And they got the highest performance 14nm FinFET process on the market for P9 and z14 while clearing the way to compete against fabless AMD and ARM vendors with P10. If anything, intel looks stupid here after falling behind in fab tech and causing immense turmoil for their external fab customers. A trivial review of IBM's systems revenue versus the datacenter breakout of any other CPU maker earnings reports will net you: yes, it is indeed quite successful.

wmf 7y ago
Because Power is on such a slow (and getting slower) cycle, it seems like the best case for Power10 is for it to be ahead of Epyc for one year then be behind for 4-5 years. And the worst case is that Power10 will already be behind Epyc when it is released. Revenue is a different story that doesn't necessarily reflect technical merit.
- justinjlynn 7y ago
  > Power is on such a slow (and getting slower) cycle... so, how are we judging cycles here? I seriously doubt Intel and AMD use the same definition that IBM do. Let's just say between IBM's cycles: - new PCI Express subsystem revision (first to market with PCI Express 4.0 in P9... who knows with P10) - new memory interface architecture (Centaur and DDR4 direct attach w/ P9 and now OMI with P10) - new coherent processor external accelerator bus (OpenCAPI and CAPI...) - new internal interrupt accelerator function architecutre (XIVE w/ P9) - new privilege level for hypervisor/guest separation (Ultravisor/PEF in P9+/P10) ... and so on. Intel and AMD do NOT do that. Let's compare apples and apples here instead of using ill-defined words like 'cycles'.
  - wmf 7y ago
    Looking at the roadmap, IBM released nothing in 2011, 2013, 2015, and 2019. The number of cores has been stagnant from 2014 to 2020.
    - justinjlynn 7y ago

microcolonel 7y ago

> Now they're giving away the IP in hopes that someone will actually start making chips from it to drive demand for Redhat. And probably it'll be RISC-V at the end of the day anyway, and it'll sell Redhat just fine. ;- )