GPU crunchers beware! this board is not ideal for DC projects whose applications are highly dependent on PCIe bandwidth, as it does not have enough PCIe lanes for a true x16x0x16x0 or x8x8x8x8 configuration. rather there is only one PCIe x16 slot, the top PCIe x16 slot, that operates at true x16 bandwidth. moving downward, the next two PCIe x16 slots operate at x4 and x8 bandwidth respectively, and both share bandwidth with the top PCIe x16 slot. the top PCIe x16 slot only operates at x16 when the two PCIe x16 slots below it, the x4 and x8 slots, are unpopulated (a x16x0x0x4 configuration). when a 2nd card is placed in the x8 slot, the x16 slot drops down to x8 operation (a x8x0x8x4 configuration). and finally, when a 3rd card is placed in the upper x4 slot (between the x16 and the x8 slots), the x8 slot drops down to x4 (a x8x4x4x4 configuration). the lower x4 slot (below the x8 slot), always has is maximum designated bandwidth of x4 b/c it does not share bandwidth with any of the other three PCIe x16 slots. also, keep in mind that the top three PCIe x16 slots (the x16, x4, and x8 slots) conform to the PCIe 3.0 standard, while the bottom PCIe x16 slot (the 2nd x4 slot) only conforms to the PCIe 2.0 standard.
all that said, Milkyway@Home might be a good project for quad-GPU setup utilizing this board b/c MW@H GPU tasks have a fairly low GPU-to-CPU runtime ratio (around 4% depending on the GPU and OS). that is to say, the great deal of the calculations are done on the GPU, and only 4% of the overall task runtime is devoted to receiving data from and sending data to the CPU over the PCIe bus. compare this to the GPU-to-CPU runtime ratio of Einstein@Home GPU tasks, which is closer to 35% on my 1090T Win7 machine, and only marginally better at 31% on my 3770K Win7 machine. they require much more data to be transferred between the GPU and CPU over the PCIe bus for the duration of the task...and they take anywhere from 100 to 600 times as long to crunch as MW@H GPU tasks! so this mobo's PCIe bandwidth will take a much greater toll on quad-GPU Einstein@Home crunchers than it would on quad-GPU MW@H crunchers.
now i don't know the Z87 range of products that well, but i do know that the
MSI Z87 XPower, on the cover of AnandTech.com right now, is another quad-GPU-ready board, and it allows for x16x0x16x0 and x8x8x8x8 configurations. granted its more than twice as expensive as the one Larry mentioned...but unless i knew for a fact that i would only ever run projects that are only mildly affected by PCIe bandwidth, i would spend the extra $230 and be done with it.
just some food for thought...
Facebook
Twitter