The fins extend only a short way over the die. It will mean a sort of push-pull over the two CPUs with the fans over the outer part of the two parallel heat sinks.
You can also change the fan to your liking with this cooler.
You can also change the fan to your liking with this cooler.
I'm trying two of these and it may be tight - but they should fit side to side and so take the air out the top - the case has room for two case fans in the top.
Parts have arrived! However, after I plundered my dual 2660 for 32GB of its 64GB memory (thus dropping it from quad channel to dual channel), completion times on O1 tasks skyrocketed. I AM hyperthreading to 32 cores on that machine, and, contrary to what I saw in another thread, I get better overall output doing so. I'm left to wonder if running so many simultaneous work units was consuming all available bandwidth on dual channel (it wasn't the total amount of RAM that was the problem; that was < 10% being used). Unfortunately, I'm not technical enough to know for sure.
Anyway, memory is back in place to see if that fixes completion times and I've ordered 8x4GB sticks for the new build.
Speeds are back to normal now that I'm running memory in quad channel again. I would avoid hyperthreading 32 concurrent work units on these (dual E5-2660 / 2770) processors unless you're running quad channel.
Speeds are back to normal now that I'm running memory in quad channel again. I would avoid hyperthreading 32 concurrent work units on these (dual E5-2660 / 2770) processors unless you're running quad channel.
81.25% CPU usage set in options
(1) AMD 7970, doing 4 concurrent tasks
24 concurrent O1 gravitational wave searches
I could probably bump that up to 28, but I've left 4 full cores for the GPU, 24 + 2*4 = 32.
GPU is doing 4 tasks in approximately 10,000 seconds, a slight gain from 3 concurrent. 95% utilization.
CPU tasks were coming in at ~55,000 each / 64GB with quad channel enabled, and exceeding 90,000 with dual channel / 32GB. Total memory usage is low, so I'm pretty sure it was a memory controller bandwidth issue.
Z9PA-D8 available on Amazon
)
Z9PA-D8 available on Amazon USA again.
OK, as always the budget get
)
OK, as always the budget get stretched... and the Z9PA-D8 was ordered a few days ago. Thanks all.
I'm wondering about what orientation for heatsink fans see motherboard - I may not have a choice, as space is tight.
RE: OK, as always the
)
I have one on the way as well. I am trying these first:
http://www.newegg.com/Product/Product.aspx?Item=N82E16835103099&cm_re=212_evo-_-35-103-099-_-Product
The fins extend only a short way over the die. It will mean a sort of push-pull over the two CPUs with the fans over the outer part of the two parallel heat sinks.
You can also change the fan to your liking with this cooler.
RE: You can also change the
)
I'm trying two of these and it may be tight - but they should fit side to side and so take the air out the top - the case has room for two case fans in the top.
I'm also using CM 212 EVO.
)
I'm also using CM 212 EVO.
Parts have arrived! However,
)
Parts have arrived! However, after I plundered my dual 2660 for 32GB of its 64GB memory (thus dropping it from quad channel to dual channel), completion times on O1 tasks skyrocketed. I AM hyperthreading to 32 cores on that machine, and, contrary to what I saw in another thread, I get better overall output doing so. I'm left to wonder if running so many simultaneous work units was consuming all available bandwidth on dual channel (it wasn't the total amount of RAM that was the problem; that was < 10% being used). Unfortunately, I'm not technical enough to know for sure.
Anyway, memory is back in place to see if that fixes completion times and I've ordered 8x4GB sticks for the new build.
Speeds are back to normal now
)
Speeds are back to normal now that I'm running memory in quad channel again. I would avoid hyperthreading 32 concurrent work units on these (dual E5-2660 / 2770) processors unless you're running quad channel.
RE: Speeds are back to
)
Great!, could you list the final rig specs?
RE: Great!, could you list
)
This is for the system I already had running (new system with ATX motherboard getting assembled this week)
Dual E-5 2660 v1 8 cores, running hyperthreaded (32 total)
64GB DDR3-1600, non-ECC, running quad channel
ASRock EP2C602-4L/D16 SSI-EEB motherboard
81.25% CPU usage set in options
(1) AMD 7970, doing 4 concurrent tasks
24 concurrent O1 gravitational wave searches
I could probably bump that up to 28, but I've left 4 full cores for the GPU, 24 + 2*4 = 32.
GPU is doing 4 tasks in approximately 10,000 seconds, a slight gain from 3 concurrent. 95% utilization.
CPU tasks were coming in at ~55,000 each / 64GB with quad channel enabled, and exceeding 90,000 with dual channel / 32GB. Total memory usage is low, so I'm pretty sure it was a memory controller bandwidth issue.
New system being assembled
)
New system being assembled will be a nice comparison, using the next higher processor model and less overall memory, but still quad channel.
(2) E5-2670 v1 processors, hyperthreaded
32GB DDR3-1600 non-ECC RAM, quad channel
ASUS Z9PA-D8 ATX motherboard