Announcement

Collapse
No announcement yet.

Gigabyte X399 AORUS Gaming 7 Works As A Linux-Friendly Threadripper Motherboard

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • starshipeleven
    replied
    Originally posted by asyn View Post
    Can you confirm bluetooth is working well on this board?
    According to the Gigabyte website, it is using a Intel AC 8265 wifi+bluetooth (it's usually a minipcie/NGFF card for laptops), so you can try googling about linux support for that.

    Leave a comment:


  • asyn
    replied
    Can you confirm bluetooth is working well on this board?

    Leave a comment:


  • edwaleni
    replied
    Originally posted by starshipeleven View Post
    This is what I was talking about. Unless it is certified it won't get bought, regardless of the fact that it may work fine or not.

    Because certification is a magic thing that allows the company to move the responsibility to someone else if the certified feature goes wrong, and many companies care more about legal responsibility than of shit actually working at all.
    Correct. Except companies like Facebook don't really care about it. If the board fails for whatever reason, they simply swap it out.

    Companies like Backblaze have found no measurable difference between consumer hardware and certified hardware in 24x7 use cases.

    So I agree, the whole certification angle is to measure risk and deflect blame when something goes awry.

    Leave a comment:


  • starshipeleven
    replied
    Originally posted by illwieckz View Post
    I'm not sure these "uncertified consumer hardware with unvalidated ECC" will be worst than those costly and faulty-by-design hardware. The thing is : there is one kind of hardware you can't be fired for having bought them, no one can be fired for having bought HP or Dell servers you know.
    This is what I was talking about. Unless it is certified it won't get bought, regardless of the fact that it may work fine or not.

    Because certification is a magic thing that allows the company to move the responsibility to someone else if the certified feature goes wrong, and many companies care more about legal responsibility than of shit actually working at all.

    Leave a comment:


  • coder111
    replied
    Oh hell, I feel for you man. The thing I am really mad about is that major corporations & banks (I'm a software dev working for one at the moment) will keep buying same crap hardware from same crap vendors, and the managers making purchasing decisions won't give a damn about the actual problems... I have similar stories about problems and purchasing decisions with software...

    Originally posted by illwieckz View Post
    Many servers from big vendors (Dell, HP…) I've put my hands on the last 9 years had very 'validated' and 'certified' problem...

    Leave a comment:


  • illwieckz
    replied
    Originally posted by starshipeleven View Post
    ECC on consumer hardware isn't validated, so while it seems to work it's not certified to work.
    Many servers from big vendors (Dell, HP…) I've put my hands on the last 9 years had very 'validated' and 'certified' problem, like one having a NIC that hangs when you just do a `lspci -v` (it was a defect for all the servers of this kind, I've got multiple ones and all were affected). Another one didn't have any option to show single drives to the OS without having to make a raid0 with one drive for each drive (who said plug an play?, who said btrfs, who said zfs? Another one that were in-factory thermal management disabled so that leaded to unattended crashes : costly vendor support replaced the motherboard multiple time and the company paid costly warrant extension for that until I found that the thermal management mechanism was disabled for this precise model and was not enable-able at all (it was written in a very small line in the documentation that this power management disablement was a feature of this precise model, other variants had, that was a server I “inherited” from the company who had hired me at this time, and it was the one in recommendation document for this kind of job), etc. etc. Oh, I also discovered that redoing the thermal paste on that server allowed us to not reach the shutdown limit (under the same very-thermal-controlled environment than before)… Well, in fact I discovered they were not using thermal paste at all, but a kind of solid carbon sheet instead (that was probably more acting as a thermal insulator since it added one additional layer of misconduct to the original one when using nothing, I've never seen that again). That was my “certified” and “validated” experience. I still have in production some of these servers which can lose their network link just doing “lspci -v”, what a joke. And you probably know that very big problem of waiting 15min to boot a server with 32Gb or ram? or things like that? And I haven't told you any horror story about faulty boot sequences due to bios misconception yet, when a drive is bootable by hundreds of computer but the server, and one random drive (same exact model) with exact same setup is, by bootable I mean the BIOS being able to find out the drive and find out the bootloader in time… I unwrapped many of brand new drives (and did the dd dance each time to put the exact same 1:1 system on it) before finding “the good one”.

    I'm not sure these "uncertified consumer hardware with unvalidated ECC" will be worst than those costly and faulty-by-design hardware. The thing is : there is one kind of hardware you can't be fired for having bought them, no one can be fired for having bought HP or Dell servers you know.
    Last edited by illwieckz; 09-20-2017, 04:48 AM.

    Leave a comment:


  • starshipeleven
    replied
    Originally posted by illwieckz View Post
    If you can't buy a Epyc-based server because Intel holds manufacturers by balls, you can still buy Naples-like ThreadRipper hardware. Don't mind the leds, there is the LAN you need, 16cores/32threads CPU, server graded components and ECC ram support up to 128GB.
    ECC on consumer hardware isn't validated, so while it seems to work it's not certified to work.

    Leave a comment:


  • Adarion
    replied
    Originally posted by flubba86 View Post
    The SoundblasterX 720 thing that the product page mentions, is a software layer.
    Ah, thanks. Yes, I was wondering why there would be a bigger RT Chip, but still some additional SB part. Okay, if this is just some optional W32-software it'll be okay.

    Leave a comment:


  • Akiko
    replied
    Originally posted by illwieckz View Post
    I'm myself more interested in the Asrock X399 Taichi, I prefer the M.2 placement. On the Gigabyte x399 Aorus you have to remove the GPU to get access to the first M.2 slot, that's not handy. I don't know why the Taichi one looks hard to find.
    I got exactly that one plus a 1950X, 4x16GiB DDR4-3600 Corsair LPX Vengeance (only 3333 MHz are stable), a Artic Liquid Freezer 240, a Corsair 850i power supply, a RX 460 (for the Linux desktop) and a GTX 1080 for PCI-passthrough all put together in a Cooler Master Cosmos II (only a real tower is able to hold a SSI-MEB mainboard ).

    The mainboard is a nice one, no doubt, I like the arrangement of the PCIe slots. The x1 slot is above a 16x slot, so you can put in a sound card without blocking the airflow of a graphics card. The board also have a nice Intel based wifi able to do a/b/g/n, but I don't use it. I also tried 4x16GiB DDR4-2400 registered ECC memory. And yes, that worked.

    So much about the nice things. And here is the downside. In the UEFI (last version 1.50) you can turn off wifi, but this has no effect. The board (or the Threadripper microcode? hard to say) has a real PCIe setup fuckup. First, there are very few options in the UEFI, you basically can only switch the main PCIe slots (the big ones) to older, slower modes and it's the same for the southbridge connection, that's all. If the 4x link to the southbridge runs in gen3 mode, Linux spams the kernel logs with PCIe TLP and DTLP messages and the performance goes down. And that seems to be a configuration issue of the whole PCIe subsystem. PCIe power management (PME) seems to be badly broken, for now the bus is unable to properly set power managment modes on PCIe cards, which breaks PCI-passthrough (KVM and Xen) completly. Right now that seems to be a real Threadripper issue affecting all boards out there.

    Leave a comment:


  • illwieckz
    replied
    I'm myself more interested in the Asrock X399 Taichi, I prefer the M.2 placement. On the Gigabyte x399 Aorus you have to remove the GPU to get access to the first M.2 slot, that's not handy. I don't know why the Taichi one looks hard to find.

    By the way, with this Aorus one clamiming “server-level power chokes”, the one from Asus (Zenith Extreme) shipped with 10Gbps interface, the Taichi one shipped with dual Gbps port, and all these motherboards providing ECC memory support, they clearly not only targeting the gaming market : AMD try to re-enter the high-end workstation market via the rear door. And I'm sure some will try to use some of them as custom servers. Since all big manufacturers are stuck with Intel, those “motherboards with led” are in fact targetting more than gaming enthusiasts.

    Do you remember between 2003& 2005 when big manufacturers like Dell where in big trouble because customers wanted superior AMD Opteron but they were not be able to supply them without losing the 1 billion per year from Intel received to ensure they not sell computer with CPU from AMD?

    So, this time AMD has a fallback : if big manufacturers yet again think Intel's money is still better than “hearts, minds, and wallets of their customers”, there is a motherboard market for Naples-like 16core/32thread CPU, 128GB ECC ram and 10Gbps or dual LAN network interface.

    Have you seen the footnote on the dual LAN description in X399 Taichi page? They make a strong advertisement about their motherboard providing dual LAN interface for bonding, but what the footnote says? It says that feature “is not supported with Windows® 10”. So, you got the message : they put on their motherboard some hardware Windows can't fully handle. They thought it would not be a big problem for their customers if Windows can't fully handle that.

    If you can't buy a Epyc-based server because Intel holds manufacturers by balls, you can still buy Naples-like ThreadRipper hardware. Don't mind the leds, there is the LAN you need, 16cores/32threads CPU, server graded components and ECC ram support up to 128GB.

    Last edited by illwieckz; 09-19-2017, 07:32 PM.

    Leave a comment:

Working...
X