AMD's Ryzen CPU's (Ryzen/TR/Epyc) & Vega/Polaris GPU's

Discussion in 'Hardware Components and Aftermarket Upgrades' started by Rage Set, Dec 14, 2016.

  1. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    4,650
    Messages:
    16,175
    Likes Received:
    19,884
    Trophy Points:
    931
    ThreadRipper only :)

    ASRock Ultra Quad M.2 Card Review
    Level1Techs
    Published on Jul 4, 2018
     
    jclausius likes this.
  2. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    4,650
    Messages:
    16,175
    Likes Received:
    19,884
    Trophy Points:
    931
    Kaby Lake-G Gaming, What Can You Play? [Part 1] Fortnite, PUBG, Battlefield 1 & More
    Hardware Unboxed
    Published on Jul 5, 2018
     
  3. jclausius

    jclausius Notebook Virtuoso

    Reputations:
    983
    Messages:
    2,492
    Likes Received:
    989
    Trophy Points:
    131
    I love this add on board, but it should work in either platform. There's no 6 lane bottleneck in x299, so he's a bit confused on that.

    For example, my asrock has 24 usable PCIe lanes going right to the cpu... So there is an x16 and two x4s. https://www.tomshardware.com/reviews/asrock-x299e-itx-ac-skylake-x-motherboard,5299.html

    Another explanation-
    https://www.reddit.com/r/intel/comments/6lha8y/x399_amd_vs_x299_intel_pcie_lanes_explained/

    http://www.tomshardware.com/answers/id-3449787/understanding-skylake-pcie-lanes.html
     
    Last edited: Jul 5, 2018
    ajc9988 and hmscott like this.
  4. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    4,650
    Messages:
    16,175
    Likes Received:
    19,884
    Trophy Points:
    931
    From your link:

    "Intel is connecting the CPU via DMI to the chipset. This has PCH lanes, which are equivalent to a PCIe3 x4. This means all 24 PCIe lanes from the chipset have to share the PCH (PCIe3 x4) link to the CPU.

    Depending on what devices you are using the chipset lanes on, this will result in slightly worse performance (vs CPU lanes), to extremely bottlenecked performance."

    This what he is trying to point out, even if his words don't always match up with his thoughts, it eventually clears up in what he says further. Scripted for technical details is often better, but not fun to spend time doing, so we get the off-the cuff delivery + corrections later. :)

    With x299 the lower core count CPU's have a higher percentage of PCI lanes going through the chipset, and it's board by board how that breaks down.

    IDK if any PCIe sockets are chipset connected, but I wouldn't be surprised to find that an open socket after GPU(s) and 10Gb ethernet / etc would be the one(s) available for such a RAID M.2 card, and might be chipset connected and bottlenecked, which wouldn't be fast enough to get full performance. That has been a problem on laptops as well.

    The M.2 on board sockets for many desktop motherboards are a mix of CPU and Chipset PCIe lands, which limits M.2 RAID participation.

    IDK, maybe that card will work on an X299 board too, if you get one let us know how it works out. :)
     
    Last edited: Jul 6, 2018
  5. jclausius

    jclausius Notebook Virtuoso

    Reputations:
    983
    Messages:
    2,492
    Likes Received:
    989
    Trophy Points:
    131
    Yes, the PCH / MDI is a x4 link, but that is separate from the CPU lanes counted for the other M.2 sockets which tie directly to the CPU lanes (bypassing PCH). For example, on the x299e-itx/ac, there are one x16 and 3 M.2 slots. So the 1 x16 goes to the CPU lanes, and two of the M.2 sockets go directly to the CPU lanes (8 lanes), the other M.2 socket is tied to the PCH which is then using the DMI link through the chipset. This is why I cannot run RAID-5 on all three M.2s... "because one of the sockets is not like the other..." that is the one tied to the SATA controller / PCH does not share a database with the other two PCIe x4 set of lanes. I can do RAID 0/1 over the two M.2s which go directly to the CPU, but don't know if the VROC crap comes into play here. I haven't done any research there.

    I added a second Tom's question which explains it a bit more - the comments in there are particularly helpful.

    Regardless, I don't have enough x16 slots in the mighty mini, otherwise, it would be nice to configure RAID 1+0 with this sweet add-on card.
     
    Last edited: Jul 6, 2018
    Papusan and hmscott like this.
  6. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,315
    Messages:
    8,915
    Likes Received:
    4,008
    Trophy Points:
    431
  7. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,315
    Messages:
    8,915
    Likes Received:
    4,008
    Trophy Points:
    431
  8. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    4,650
    Messages:
    16,175
    Likes Received:
    19,884
    Trophy Points:
    931
    China producing x86 chips nearly identical to AMD server processors
    Processors using licensed x86 IP could have implications for trade war, national security. A chip almost identical to AMD's EPYC processor has begun to appear in China. It's the result of an AMD joint venture with a Chinese government-owned investment group.
    SEAN GALLAGHER - 7/9/2018, 11:45 AM
    https://arstechnica.com/information...ps-nearly-identical-to-amd-server-processors/

    "Thanks to a licensing deal with AMD and a complex joint-venture arrangement, the Chinese chip producer Chengdu Haiguang IC Design Co. (Hygon) is now producing x86-based server processors that are largely indistinguishable from AMD's EPYC processors—so close in design that Linux kernel developers had to do little in the way of patching to support the new processor family, called "Dhyana." The server chips are being manufactured for domestic use only—part of an effort to break China's dependence on foreign technology companies.

    Since the Edward Snowden revelations about National Security Agency efforts to use implants in technology products to conduct foreign intelligence collection, China has been applying increasing pressure on US technology providers. The country hopes to bolster its own domestic technology industry through strict new information security regulations and investment in domestic suppliers.

    The need for a domestic producer of high-performance server processors has also been driven by US export restrictions to China—in 2015, the administration of President Barack Obama blocked a sale of Intel Xeon processors for China's Tianhe-2 supercomputer over concerns that the sale would aid China's nuclear weapons program. Export of high-performance processors to China have been restricted ever since, and the US government has also moved to prevent China from acquiring technology companies in the past over national security concerns.

    The unintended consequence of these restrictions has been that the Chinese government has spurred investment in domestic processors—and it has found ways around import restrictions with licensing arrangements and joint ventures. Suzhou PowerCore Technology Co. licensed the Power8 architecture from IBM in 2015; Zhaoxin, a state-owned technology company, has designed domestic x86 desktop processors in a joint venture with VIA. And now, the AMD licensing deal—which leverages both a licensing deal and a joint venture to allow the use of x86 intellectual property—is delivering what could be the first step toward a domestic high-performance server platform.

    Based on AMD's "Zen" core architecture and EPYC, the "Dhyana" processor appears to be focused on embedded applications for now. They're not socketed processors but are instead a system-on-chip (SoC) design, similar to the EPYC embedded computing processors being manufactured elsewhere by AMD. They're so similar, in fact, that, according to a report from Michael Larabel of Phoronix, moving the Linux kernel code for EPYC processors over to the Hygon chips required fewer than 200 new lines of code.

    The SoC design doesn't necessarily preclude the use of the Dhyana processors in high-performance cluster applications or in data center applications that would normally be filled (if not for trade restrictions) by Intel Xeon or other server processors. And given China's overall push to enhance its own information technology and manufacturing despite US trade restrictions, embedded server technology may be more in line with the current domestic demand.

    The good news for AMD is that the joint venture will net the company a steady stream of royalties in addition to the $239 million in cash the company was paid up front in 2016 by Tianjin Haiguang Advanced Technology Investment Co. (THATIC), an investment arm of the Chinese Academy of Sciences, to enter the joint venture in the first place. Just how long that silver lining lasts, given the current climate between Washington and Beijing, is uncertain. But it's certainly a benefit that comes with potential national security downsides."
    Comments
     
    Vasudev and ajc9988 like this.
  9. Papusan

    Papusan JOKEBOOKS = That sucks!! STAHP! Dont buy FILTH...

    Reputations:
    12,383
    Messages:
    18,517
    Likes Received:
    27,883
    Trophy Points:
    931
    I see AMD selling it’s soul. Oh’well
     
  10. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,315
    Messages:
    8,915
    Likes Received:
    4,008
    Trophy Points:
    431
    AMD made the deal well before present day. It made for a much needed cash infusion and set it so that it would be the dominant force within that market. Not a bad deal at all.

    Now if it were just giving it away to stop the competition, as that other entity has, that would be selling its soul!
     
    Vasudev, hmscott and ajc9988 like this.
Loading...

Share This Page