• Yokemate of Keyboards
    Yokemate of Keyboards
    Posts: 12098 from 2003/5/22
    From: Germany
    >> what about the X5000's x4 slot (the one behind the PCIe-PCIe bridge)? Wouldn't
    >> a SATA3 card work full-speed in it, even if it was somehow limited to x2 speed
    >> by the bridge bottleneck?

    > Would it, as its multiplexed?

    I think so, as I doubt that multiplexing reduces the nominal bandwidth by more than 50%. There are 4 input lanes to the bridge, so using x2 bandwidth for a SATA3 card in the x4 slot leaves another x2 bandwidth for the remaining two x1 slots and the PCI bridge with two PCI slots. This means that using further cards in the two x1 slots (or in one x1 slot and one or two PCI slots) would be within the bandwidth of the PCI controller and the SATA3 card wouldn't be slowed down. Just when the two x1 slots *and* one or two PCI slots are occupied by cards, the 4 input lanes have to be multiplexed to more than 4 output lanes, so bandwidth would likely not be enough to sustain full SATA3 speed anymore.

    > Where'd you find the X1 video cards? The only X1 cards I've seen were
    > AMD FirePro cards and similar designs primarily intended to increase the
    > number of displays

    Yes, FirePro and FireMV. In EU online stores, I find currently available new:
    - FireMV 2250 (RV516) for 56 EUR
    - FireMV 2260 (RV620) for 89 EUR
    - FirePro 2450 (RV620) for 271 EUR
    - FirePro 2270 (RV810) for 113 EUR

    > I just did a quick comparison between the P1022 and the T1042's SerDes I/O lines.

    Regarding P1022:
    The P1022 on the Tabor/A1222 riddles me. A-Eon claims one PCIe x16 slot with x4 bandwidth for Tabor. If we look at the possible SerDes configurations for the P1022, we can see that the most sane configurations for PCIe that allow both SATA and Ethernet are x1 x1 x1 (0x06) and x2 (0x0e). This means the x16 slot could have x2 bandwidth at best, not x4. The only option that provides x4 bandwidth for one PCIe controller would be x4 x2 (0x1c), but this lacks both SATA and Ethernet.
    Even if we assume that assignment of the P1022's 6 SerDes lanes was completely arbitrary and not restricted by predefined configurations, PCIe x4 + 2x SATA + GbE would obviously require 7 SerDes lanes.
    Something seems not right here. Any ideas?

    Edit: mystery solved

    Regarding T1042:
    If we look at the possible SerDes configurations (or there, page 37) for the T1042, we see that the most sane configuration is 0x08 with x4 x1 x1 for PCIe, also providing 2x SATA as well as Ethernet (via "Parallel Port").

    > 20 lanes at 5 GHz of the P5040

    ...of which max. 8 can be assigned to PCIe.

    > the only thing that comes close to the 11,2 is Freescale's T4240RDB

    The T4240RDB has one x8 slot and one x4 slot (both at 5 GHz / PCIe v2), so only 50% of slots, 37.5% of PCIe lanes and 75% of PCIe bandwidth. The T4240QDS, which has up to four x4 slots and thus 100% of slots, has only 50% of PCIe lanes but 125% of PCIe bandwidth.

    > the T4240 has 32 SerDes lanes at 10 GHz.

    ...of which max. 16 can be assigned to PCIe (twelve at 5 GHz / PCIe v2, four at 10 GHz / PCIe v3).
    (Beside, the T4241 is to be preferred.)

    > since they're almost ready, how about the X5000 port and the 11,2 port

    I think the X5000 port *is* ready, or isn't it? :-)

    [ Edited by Andreas_Wolf 15.06.2018 - 13:14 ]
  • »26.09.17 - 02:00