G4 vs. G5
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12058 from 2003/5/22
    From: Germany
    >>> from very beginning i report to aeon the bed video gpu performances ...
    >>> result they change the document specs of x5000.

    >> Which specs in particular did they change after your reporting?

    > made more better the pcie 16x description ;-)

    You mean when A-Eon wrote "PCIe x16 slot (Gen2 x4)" in their November 2016 press release? Or do you mean they omitted the fact in the original Cyrus Plus technical reference manual (version 1.1.1 from July 2014 makes no secret of the fact)?

    >> Together with the TerraSoft/Fixstars/YDL PowerStation, I guess :-)

    > I was thinking this machine are more for server use :P

    It was advertised for both server and workstation use.

    >>> on my quad [...] only r7 is not running

    >> Isn't that a contradiction to what you just said yourself? :-)

    > where?

    There: "the same gpus work on G5 and on x5000" :-)
  • »10.09.17 - 20:53
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Quote:

    slot (Gen2 x4)" in their November 2016 press

    this ... i need to make a correction my x5000 come 29/9/2016 .. i was sure it come in october but i was wrong... just find my video of unboxing.

    Quote:

    There: "the same gpus work on G5 and on x5000" :-)


    ah ah you right. for have the gpus run on x5000 the gpu firmware have to be build on the kernel. On g5 quad need to be installed on the lib/firmwares on the hd. i did not have the firmware in the lib when i made the test on G5 this is why im not sure.

    Quote:


    It was advertised for both server and workstation use.


    you have better memory then me :-)

    [ Edited by tlosmx 10.09.2017 - 22:02 ]
  • »10.09.17 - 21:01
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Andreas my english is not good and for me is difficult to write .


    >Why? I don't see where the lstopo output shows more than one SoC.

    the soc if one on the topo have to be one only ... if you see the image you will see two with same id


    >So you say that the Cyrus board topology is wrong because QEMU can't recreate it? Wouldn't it rather mean that there's >something wrong with QEMU if it can't recreate the topology of an existing board?

    Nope

    >> this has nothing to do with the board design, don't you think?
    yes the cpu scheme is all ok, are the pci bridges and soc strange

    >What's wrong with the L3 cache in the lstopo output?
    Nothing strange there about L3 cache is ok in the topo . but it is used only for corenet and not as sdram
    for be sdram have to be set up in source config of uboot ... but no one setup it. and build.
    i hope now is better :P



    > my fear is the e5500 true speed is ddr 200 at true 400MTs data rate.
    by the way 1 GB/s to 4 gb/s max for 5040 are not DD3 1600 speed .. (stream and linux kernel report)
    1 GB/s to 2 gbs max for 5020

    G5 Quad is 7/8 Gb/s near 10Gb/s in Altivec

    Os4 ragemem Performances of 5020
    image.jpg
    vs x1000
    image.jpg

    Quote:

    So you say that the Cyrus board topology is wrong because QEMU can't recreate it? Wouldn't it rather mean that there's something wrong with QEMU if it can't recreate the topology of an existing board?

    nope ... the topology of the cyrrus is totally strange compared all topology in the world Informaic Tecnology.
    qemu dont manage it because strange.



    check and compare topos:


    SAM 460

    https://www.dropbox.com/s/pps1q9ropeusfh0/16196243_10208175166078464_2023629800_o.jpg?dl=0

    X1000

    https://www.dropbox.com/s/synwq2x1k081jn3/16117271_10208143633210162_1285018612_n.jpg?dl=0

    PowerMac G5 Quad

    https://www.dropbox.com/s/b9ajetsf8r1kjh0/16118530_10208141403954432_751303834_n.png?dl=0

    Tabor
    https://www.dropbox.com/s/joziw57zfoq0h3u/16237927_10208180847220489_965077643_n.jpg?dl=0

    a PC
    https://www.dropbox.com/s/fu3bwiooqg4q9as/16111653_10208132217124767_495101249_n.png?dl=0

    X5000 /20...
    https://www.dropbox.com/s/wdwuib52bii2bcn/16176754_10208155966798494_1032589834_n.png?dl=0

    X5000/40
    https://www.dropbox.com/s/8rgcvr6532fmzt5/16128954_10208172248685531_756994869_n.png?dl=0




    [ Edited by tlosmx 10.09.2017 - 23:12 ]
  • »10.09.17 - 21:30
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Hmm, those memory benches are very disappointing.
    "Never attribute to malice what can more readily explained by incompetence"
  • »10.09.17 - 21:58
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Quote:

    Jim wrote:
    Hmm, those memory benches are very disappointing.


    exactly for the 5020 is the same speed of P4 born in 2001 ddr 266mhz
  • »10.09.17 - 22:10
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    if you need stream result here are for compare
    https://www.cs.virginia.edu/stream/by_name/Bandwidth.html
  • »10.09.17 - 22:14
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Quote:

    tlosmx wrote:
    Quote:

    Jim wrote:
    Hmm, those memory benches are very disappointing.


    exactly for the 5020 is the same speed of P4 born in 2001 ddr 266mhz


    Not the greatest analysis, since the P4 did have pretty good memory bandwidth (for the period).
    And, for some reason the X5000/40 has slightly better bandwidth.
    The really telling thing is how poorly the X5000 compares to the X1000.

    So, just how accurate are these figures?

    And, I wish we could get a better idea of how this compares to the quad G5, but that comparison can only occur under Linux right now.

    You had me wondering why an advocate for the X5000 would suddenly be recommending the G5.

    The X4 PCI-E slot is not that big a deal for me (it should be adequate for the cards that are suited to linux and NG OS').
    But these other issues are...unsettling.
    "Never attribute to malice what can more readily explained by incompetence"
  • »10.09.17 - 22:58
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12058 from 2003/5/22
    From: Germany
    > the soc if one on the topo have to be one only ...
    > if you see the image you will see two with same id

    You mean the "PCI 1957:0451" ID? If yes, are you sure this is the P5040 SoC?

    > L3 cache [...] is used only for corenet and not as sdram
    > for be sdram have to be set up in source config of uboot ... but no one setup it. and build.

    Then this should be changed in U-Boot. Who is the developer/maintainer of U-Boot for Cyrus/X5000?

    > 2 gbs max for 5020
    > G5 Quad is 7/8 Gb/s

    Yes, I already compared my PowerMac G5 to the X5000 (both with MorphOS) last year and basically came to the same conclusion:

    AmigaMARK Memory, stream:
    http://morph.zone/modules/newbb_plus/viewtopic.php?forum=11&topic_id=11137&start=593

    RAGEMEM:
    http://morph.zone/modules/newbb_plus/viewtopic.php?forum=11&topic_id=11137&start=624

    > the topology of the cyrrus is totally strange compared all topology in the
    > world Informaic Tecnology. qemu dont manage it because strange.

    Okay, so the Cyrus board topology is strange, not wrong :-)
  • »10.09.17 - 23:46
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Quote:

    And, for some reason the X5000/40 has slightly better bandwidth.

    P5040 is different from 5020 in number of cores,mhz and ram P5020 use ddr3 1333 P5040 ddr3 1600 plus the cpus differs in the logic scheme . P5040 is "more modern"


    Quote:

    So, just how accurate are these figures?

    they reflect the same situation that i found on linux. was made by one of my friend with last release of os4.


    Quote:

    The X4 PCI-E slot is not that big a deal for me (it should be adequate for the cards that are suited to linux and NG OS').



    the pcie 4x is not the best under linux too.more faster is better will be the result ... this is why every time i switched from x5000 to G5 quad was like change from an amiga 500 (standard) to an amiga 3000 in performances.
  • »10.09.17 - 23:48
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Quote:

    Andreas_Wolf wrote:
    > the soc if one on the topo have to be one only ...
    > if you see the image you will see two with same id

    You mean the "PCI 1957:0451" ID? If yes, are you sure this is the P5040 SoC?
    yes if not the soc we will have a more big problem.

    > L3 cache [...] is used only for corenet and not as sdram
    > for be sdram have to be set up in source config of uboot ... but no one setup it. and build.

    Then this should be changed in U-Boot. Who is the developer/maintainer of U-Boot for Cyrus/X5000?
    Yes. Developer varisys, manteiner was before i quit Perez.

    > 2 gbs max for 5020
    > G5 Quad is 7/8 Gb/s

    Yes, I already compared my PowerMac G5 to the X5000 (both with MorphOS) last year and basically came to the same conclusion:

    AmigaMARK Memory, stream:
    http://morph.zone/modules/newbb_plus/viewtopic.php?forum=11&topic_id=11137&start=593

    RAGEMEM:
    http://morph.zone/modules/newbb_plus/viewtopic.php?forum=11&topic_id=11137&start=624

    > the topology of the cyrrus is totally strange compared all topology in the
    > world Informaic Tecnology. qemu dont manage it because strange.

    Okay, so the Cyrus board topology is strange, not wrong :-)


    Working bad... issue with some boards: example cmi sound card made machine freeze, slow performances pcie sata 3 controllers with fast ssd sata3 max 400mb/s r 100Mbs w . flashdrive max 900mb r .. 150 w.... and so and so.


    this are my bechmark not with altivec : https://www.dropbox.com/s/wshgd98gfny33ma/bench.xlsx?dl=0

    distributed G5 Quad on fedora ppc64 dnet bench

    [Aug 23 16:52:41 UTC] Automatic processor type detection found
    a PowerPC 970MP (G5) processor.
    [Aug 23 16:52:41 UTC] OGR-NG: using core #1 (KOGE 3.1 Hybrid).
    [Aug 23 16:53:00 UTC] OGR-NG: Benchmark for core #1 (KOGE 3.1 Hybrid)
    0.00:00:16.29 [42,505,737 nodes/sec]
    [Aug 23 16:53:00 UTC] RC5-72: using core #4 (KKS 7450).
    [Aug 23 16:53:21 UTC] RC5-72: Benchmark for core #4 (KKS 7450)
    0.00:00:17.47 [18,852,434 keys/sec]
    [Aug 23 16:53:21 UTC] Compare and share your rates in the speeds database at
    http://www.distributed.net/speed/
    (benchmark rates are for a single processor core)

    Quote:

    You mean the "PCI 1957:0451" ID? If yes, are you sure this is the P5040 SoC?



    yes if not the soc we will have a more big problem.

    edit: e5500 not have some math instruction that are emulated by linux kernel. plus dont have the numa too.

    [ Edited by tlosmx 11.09.2017 - 01:25 ]
  • »11.09.17 - 00:08
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12058 from 2003/5/22
    From: Germany
    >> Who is the developer/maintainer of U-Boot for Cyrus/X5000?

    > Developer varisys, manteiner was before i quit Perez.

    I hope they've been made aware of the problem. It doesn't sound too difficult to fix.
  • »11.09.17 - 10:04
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    they need only enable an option in the uboot .config and disable another one... but i asked about in march 2016 when i had the 5020 ... now we are in september 2017 and:
    uboot is the 2014 version.
    nothing change about L3 options.
    Uboot integrate the varisys Cyrus config in mainstream github :
    shot

    [ Edited by tlosmx 11.09.2017 - 14:11 ]
  • »11.09.17 - 13:10
    Profile
  • Order of the Butterfly
    Order of the Butterfly
    Spectre660
    Posts: 275 from 2015/6/30
    PCIex1 CMI8738 sound card freeze issue also occurs on Sam460ex .
    http://forum.hyperion-entertainment.biz/viewtopic.php?f=17&t=3528

    PCiex1 ENVYHT24 with ASM1083 revision 1 bridge csip works so it is possible that the issue with the
    CMI PCIex1 cards may be able to be fixed in software.
    http://forum.hyperion-entertainment.biz/viewtopic.php?f=57&t=3625&start=40#p41538

    Quote:

    Working bad... issue with some boards: example cmi sound card made machine freeze,


    [ Edited by Spectre660 11.09.2017 - 22:12 ]
  • »12.09.17 - 01:07
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Thanks, initially I may just use an Envy 24HT PCI card.
    That will still goes through a bridge (PCI to PCI-E onboard), but maybe that will work OK.
    "Never attribute to malice what can more readily explained by incompetence"
  • »12.09.17 - 09:34
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    Quote:

    Spectre660 wrote:
    PCIex1 CMI8738 sound card freeze issue also occurs on Sam460ex .
    http://forum.hyperion-entertainment.biz/viewtopic.php?f=17&t=3528

    PCiex1 ENVYHT24 with ASM1083 revision 1 bridge csip works so it is possible that the issue with the
    CMI PCIex1 cards may be able to be fixed in software.
    http://forum.hyperion-entertainment.biz/viewtopic.php?f=57&t=3625&start=40#p41538

    Quote:

    Working bad... issue with some boards: example cmi sound card made machine freeze,



    why you dont write about a core a-eon linux manteiner who wrote "why my cmi card work on my x1000 and freeze on x5000?"
  • »12.09.17 - 10:32
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12058 from 2003/5/22
    From: Germany
    > an Envy 24HT PCI card [...] will still goes through a bridge (PCI to PCI-E onboard)

    It's even two bridges, as the PCIe-PCI bridge resides behind a PCIe-PCIe bridge.
  • »12.09.17 - 14:28
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Quote:

    Andreas_Wolf wrote:
    > an Envy 24HT PCI card [...] will still goes through a bridge (PCI to PCI-E onboard)

    It's even two bridges, as the PCIe-PCI bridge resides behind a PCIe-PCIe bridge.


    OMG, our poor developers...
    "Never attribute to malice what can more readily explained by incompetence"
  • »12.09.17 - 16:48
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    topo explained

    [ Edited by tlosmx 12.09.2017 - 22:12 ]
  • »12.09.17 - 21:12
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Quote:

    Andreas_Wolf wrote:
    > topo explained

    Link doesn't work for me.


    Me either.
    "Never attribute to malice what can more readily explained by incompetence"
  • »12.09.17 - 22:01
    Profile
  • Butterfly
    Butterfly
    Posts: 80 from 2017/9/10
    pls chek it now


    topo
  • »13.09.17 - 00:02
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    amigadave
    Posts: 2793 from 2006/3/21
    From: Northern Calif...
    Quote:

    Quote:

    Jim wrote:
    Hmm, those memory benches are very disappointing.


    .........The really telling thing is how poorly the X5000 compares to the X1000.

    And, I wish we could get a better idea of how this compares to the quad G5, but that comparison can only occur under Linux right now.

    You had me wondering why an advocate for the X5000 would suddenly be recommending the G5.

    The X4 PCI-E slot is not that big a deal for me (it should be adequate for the cards that are suited to linux and NG OS').
    But these other issues are...unsettling.


    My X1000 is looking better and better each day, for running OS4 anyway. I like that it has Altivec and is using the e6500 core (I think), instead of the "weaker?" e5500, e600, or any other PPC core design.

    If I remember correctly (which is highly unlikely), the memory read/write speeds and bandwidth for the X1000 was one of it's very few strong points, when the X1000 was compared to G4 and even G5 systems. That superior memory speed and maybe bandwidth, partially may have made up for the many other areas where the PA6T, or X1000 was/is slower than used PPC Mac hardware.

    It is a shame, and I feel bad for Trevor and Matthew, that the X5000 is experiencing these apparent problems, or short falls in performance, since so much has been invested in creating what "is" the current (and possibly last) ultimate PPC AmigaOS4 compatible system. I kind of doubt that Hyperion will ever port AmigaOS4 to any Power9 (or even Power8 or lower) work station class systems, and no talk has been leaked that anyone is even thinking of porting AmigaOS4 to the x64 architecture.

    Still, everything I have heard from the X5000 beta testers I have talked to, indicates that the X5000 is very noticeably faster at running AmigaOS4.1FE, when compared to the X1000 (or was that comparisons while running Linux?). This thread is the first I have seen where anyone has complained so extensively about the performance of the X5000. I'll have to get more info for Jim when I attend AmiWest 2017 next month, and take notes while playing with one of the many X5000's that will be displayed at the show. Maybe I will be able to set up my X1000 right next to someone's X5000, so a "side-by-side" comparison can be done, running the exact same game, or other software application.

    That reminds me, I need to upgrade my X1000's video card and drivers, before the show dates. I haven't decided which video card I am going to put into it yet. I have a 6xxx, or 7xxx Radeon card in my i5 PC, which is a double width card with double fans attached to it, but I'm not sure I want to put such a noisy video card into my X1000, and don't think AmigaOS4.1FE has any software that needs such a powerful video card. I'll probably try to get recommendations from other X1000 and X5000 owners, regarding which video card to look for, then try to have it delivered within the next couple weeks, so I can install it and play around with AmigaOS4.1FE a bit before I pack it up to take to the show.

    Let me know if there is anything specific you want tested, or asked of the X5000 owners Jim, so it can help you with your decision to buy an X5000-20, or X5000-40.
    MorphOS - The best Next Gen Amiga choice.
  • »13.09.17 - 04:08
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    takemehomegrandma
    Posts: 2720 from 2003/2/24
    The X5000 is more of a sideways move, not a forward move, from the X1000's performance level, that fact was established years ago.
    MorphOS is Amiga done right! :-)
    MorphOS NG will be AROS done right! :-)
  • »13.09.17 - 05:31
    Profile
  • Jim
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Jim
    Posts: 4977 from 2009/1/28
    From: Delaware, USA
    Quote:

    takemehomegrandma wrote:
    The X5000 is more of a sideways move, not a forward move, from the X1000's performance level, that fact was established years ago.


    I actually preferred the X1000's expansion slot setup as the P50xx doesn't offer as many PCI-E lanes, and the fact that the PA6T has AltiVec instructions is a plus.
    I don't think the PA6T core is quite as nice as the e6500 core, but it had some advantages over the e5500.

    And the Gen2 PCI-E lanes, SATA2 capability, and DDR3 off the e5500 SOCs are already becoming dated, when compared to current CPUs.

    I'd still favor the X5000 over the X1000, but the memory benches are disheartening. I wonder if these can be improved with better firmware.

    If the choices are hoping for an X64 shift that appears to be stagnant, waiting for support for X5000 and MorphOS 3.10, sticking with what we have, or hoping for PCI-E G5 support (or some other RISC solution)...

    I'm not surprised we're losing users.

    David - Have a good time @ AmiWest. My guess is the focus may be the A1222.
    The teething issues with the X5000 are actually to be expected, and some of these things will improve.
    From what I have been able to tell, the X5000/40 should be ready for release soon.
    Since that will offer twice the core count of the X1000, that's the model I really think makes more sense.

    And, of course, I still like the goals of the T2080 laptop project.

    Hopefully some solutions to utilizing the additional cores available on newer systems will eventually become available.

    [ Edited by Jim 13.09.2017 - 03:00 ]
    "Never attribute to malice what can more readily explained by incompetence"
  • »13.09.17 - 06:49
    Profile
  • Yokemate of Keyboards
    Yokemate of Keyboards
    Andreas_Wolf
    Posts: 12058 from 2003/5/22
    From: Germany
    > topo

    One of the first two squares of the 2nd PCIe controller is surely the PCIe-PCIe bridge, which creates 8 output lanes from 4 input lanes (see block diagram in the Cyrus Plus technical reference manual as well as the first link I gave you in comment #64).
  • »13.09.17 - 10:44
    Profile