Interesting to note: PCI-e bandwidth really does seem to matter
22 Jul 2013 2:10:42 UTC
Topic 197084
(moderation:
Just something I've now seen first-hand that may assist others: I moved my Radon HD 6870 from a PCI-e 2.0 x16 slot to another physical 2.0 x16 slot (but electrically, it acts as a 2.0 x4 slot), and my Perseus Arm Survey Work Units now take an extra 20 minutes to complete.
There have been dozens (if not hundreds) of gaming/hardware site articles that show miniscule differences in PCI-e bandwidth in games, but the difference in E@H is noticeable. My 6870 runs 2 WUs at a time; 20 minutes by itself may not be much, but over two days' worth of crunching, that's the difference of two entire Perseus Arm Survey Work Units.
So, the moral is: use the highest bandwidth PCI-e slot available for your most powerful BOINC-crunching graphics cards.
(p.s. The reason I moved my graphics card in the first place is because I got an LSI MegaRAID card that craves bandwidth -- and it may be selfish, but the LSI card is so awesome, it deserves said bandwidth).
Interesting to note: PCI-e bandwidth really does seem to matter
)
Thanks for informing us! Can you quantify the performance hit futher, e.g. in percent?
And that MegaRAID card.. can it actually use 16 lanes? 4 lanes are already quite fast for storage and I don't think I have seen any controller card using more than 8 lanes. Oh, and you'd need to connect some serious hardware to it to actually benefit from the higher interface speed.
MrS
Scanning for our furry friends since Jan 2002
Looking back at the task
)
Looking back at the task completion times, it appears it would be between 5-10% difference. Not earth-shattering, but over the course of 48 hours, it does make a difference.
The MegaRAID card doesn't actually use 16 lanes; it uses 8. And I have a few 15K Cheetah SAS 300GB drives connected ;). My motherboard on this box has a x16 slot and a x4 slot (x16 physically), and when I put the RAID card in the x4 slot it gave errors. The Radeon card is less picky. :)