I know it might be a little early yet, but can anyone post their pcie 2.0 x16 bandwidths for the new x58 chipsets?
I’d like to know which board offers the best bandwidth, preferably in SLI.
Cheers
No one?
SLI is not supported in CUDA, so this is a little bit a strange question. For the numbers I cannot help you (unfortunately I don’t have a 58 :()
I’ve got an X58 board that I haven’t been able to test yet–hopefully this week. I’ll let you know. I doubt it will be much better, though.
@E.D. Riedijk
Sorry i should have said dual cards, rather than SLI.
Cheers…Looking forward to it.
Tmurray any news?
Cheers
I’ve just gotten my X58 board and will be completely up and running in about a day. How can I test PCI-E bandwidth?
In the CUDA SDK, there should be a program:
./bandwidthTest --memory=pinned
OK, I have some results:
Running on…
device 0:GeForce 9800 GT
Quick Mode
Host to Device Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 __________ 5575.4
Quick Mode
Device to Host Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 __________ 4945.6
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 __________ 40823.4
&&&& Test PASSED
Press ENTER to exit…
Sorry, my Nehalem box won’t be up until next week. (You can buy 4GB DDR3 DIMMs now, so I’m waiting for those to show up before I finish the thing.)
(apparently I wrote year first, that’s not true at all. or at least I hope that’s not some self-fulfilling prophecy)
Cheers Mr_Nuke, can i enquire what motherboard you use?
That would be the ASUS Rampage II Extreme.
The biggest difference is the system memory bandwidth improvement, see my result below with X58 verse P35 based motherboards using the same GTX 280:
Asus P6T Deluxe (Intel X58)
3.6GHz OC Core i7 920, DDR3 1440 8-7-7-22 (18.1 GB/s Sandra Mem)
Running on…
device 0:GeForce GTX 280
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5101.5
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5014.2
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 84300.5
&&&& Test PASSED
Results from Intel P35 board PCIe 1.0, Q6600, 4GB DDR2 800 (sandra mem bandwidth 5.2GB/s),
with same GTX 280 as above:
Running on…
device 0:GeForce GTX 280
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1991.7
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1521.4
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 87300.5
&&&& Test PASSED
P35 is PCIe 1.0 (and that’s quite low for 1.0 16x anyway).
So what should have we expected out of the X58?
5, 6, 7 GB/s?
I would have expected closer to 6, but that might be a BIOS issue right now. I still don’t have my board running (boo shipping).
Well, 5.1 GiB/s is not too bad for pageable memory. Perhaps ncc4 can repeat the tests with pinned memory? That helps a lot on older chipsets, but might make less of a difference now that you can get 18 GiB/s bandwidth from main memory.
That’s pageable?! Holy crap, I thought it was pinned just from the numbers (didn’t actually notice the pageable part!). For reference, C1060 on Server 2008 with a Harpertown Xeon:
Running on…
device 0:Tesla C1060
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1717.1
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 1396.6
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 56451.0
&&&& Test PASSED
Press ENTER to exit…
WOW, that’s a pretty expensive machine right there…
You’re lucky if you’re just waiting on shipping once. Mine fried within the first few days. I am still waiting on the replacement (hope newegg can keep up to their reputation).