add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube

Comments (Continued)

  • 11 months ago
  • 2 points

No problem, glad I could help

  • 11 months ago
  • 1 point

Hi again, I had one more quick question if I could trouble you with it.

You mentioned that you used 2 drives to get the fastest speed possible. Do you mean that you got 2 of the exact same drives because you are running them in RAID 0? And, if so, how well does the RAID 0 work in this setup? Do you know how it functions in terms of the chipset's PCIe x4 bus? For example, in your case (assuming you are in RAID 0), do the two drives act like one drive and thus not compete for chipset<->CPU bandwidth?

Thanks again:)

  • 11 months ago
  • 1 point

Yep, I run 2 identical drives in RAID 0. The two drives do act as one drive software wise, and while I'm not sure 2 drives not in RAID would compete for resources, these certainly do not, but certainly are bottlenecked. I get about 3525MB/s read and 3371MB/s write. A single 970 evo is rated for 3500MB/s read and 2500MB/s write. As you can tell the read speed is definitely bottlenecked but the write is improved somewhat. With no bottlenecks, RAID 0 should in theory double the speed but here it obviously does not.

In the end they still are 2 separate devices on the PCIe bus so I'm not sure if there really is any bandwidth advantage over just having them normally run.

  • 11 months ago
  • 1 point

Awesome, thanks! Ha, yes, that's super fast for sure.

I wonder if there's any additional latency with DMI 3.0 that isn't necessarily reflected in the benchmarking speeds. What do you think? If so, are there other benchmarks that might help show that latency? I'll look into it. Of course, we're not really talking about anything discernible... but it's always nice to know we're running things to their fullest!

If I'm looking at the Hero XI manual correctly, it looks like there are 2 CPU PCIe slots that can run in x8/x8 or x8/x4+x4 (I assume "x4+x4" means I could run two M.2 in that same slot each using x4 with no penalty?). I'm toying with the idea of putting 1-2 M.2 drives on one of the CPU slots and a GPU on the other that might not saturate. I'm happy with 1080p displays, so I think that might work well for my purposes. Still, I'm probably totally overthinking things and should go the easier route of using the board's M.2 slots!

  • 11 months ago
  • 1 point

Huh, I think I found a few relevant benchmarks.

Seems like the PCH will be totally fine for me :)

  • 11 months ago
  • 1 point

Glad to hear you got it figured out. Yes the board/chipset does support PCIe bifurcation and you could run the 2 M.2's over CPU PCIe.

Sort

add arrow-down arrow-left arrow-right arrow-up authorcheckmark clipboard combo comment delete discord dots drag-handle dropdown-arrow errorfacebook history inbox instagram issuelink lock markup-bbcode markup-html markup-pcpp markup-cyclingbuilder markup-plain-text markup-reddit menu pin radio-button save search settings share star-empty star-full star-half switch successtag twitch twitter user warningwattage weight youtube