// DBOINCP-300: added node comment count condition in order to get Preview working ?>
dmike
Joined: 11 Oct 12
Posts: 76
Credit: 31,369,048
RAC: 0
14 Nov 2012 15:42:57 UTC
Topic 196627
(moderation:
)
Curious about WU time completion on the Xeon and a Tesla card. If you have either (or both), please share with me which model and what your average WU completion time is. Thanks in advance!
I have seen some Xeon boxes running when browsing profiles, I know we have that around. Got to be at least a Tesla running somewhere. I'm thinking though that the Tesla will not be a huge benefit to E@H for what it is.
I have access to a Tesla K10 and an M2090 (I think) but as it's a shared resource I can't set up Boinc on that machine.
Is there a way to run our BRP cuda apps one at a time just to see how long it takes. This thing has 4 GPUs with 4GB each so it can probably run maybe 32 or 48 wu simultaneously.
My tests so far show it about the same as my GTX670 for a single task. I'm not sure why but the bandwidth is not as good. Still working on that.
I believe that's 3 GPUs total. The K10 has 2 and the other has one (which is Fermi).
The K10 has 4GB for each GPU, the other has 6GB.
I don't think E@H would benefit much though because Tesla really shines with the double floating point precision which the app doesn't run, so hearing that it runs about like a 670 seems about right. I imagine the M2090 would be much faster than a 670 with the Fermi, but could be wrong.
I had wrongfully imagined that the Xeon would be blasting away even the 3rd gen i7 in crunching. Of course, even that latter Xeon is much older.
If you're referring to the Xeon 5365, that's even before the first Core iX series - way older, so it's not a fair comparison to the current iX series. It was powerful at that time (circa 2007), but also power-consuming, measuring up to 150W each.
I've got my old Mac Pro Xeon 5365 replaced with the one on 5550 variants (2009) running at the same amount of cores at 2.67GHz only (cooler 90W each), but Einstein crunches twice the performance of the 5365s.
Current Xeon units should be on par or better than the 3rd gen i7s, but then again, I've never got one in hand to crunch...
Yeah, these are older Xeons; the 5150's were standard in 2007 Mac Pros, and the 5365's were an optional high-end upgrade. 5150's are essentially free on eBay now; the 5365's cost about $150 each at a minimum because of demand from people with first generation Mac Pro's who want to squeeze more life from their machines.
They are indeed power-hungry and run hot; that machine running an ATI 5770 with two simultaneous BRP tasks and 5 CPU tasks eats up 450w with the CPU temps in the 65c range, and I'm in the process of retiring it as a cruncher. It's still a decent machine for everyday work - those two Xeon 5365's give me a CPU Passmark score of 8000.
I believe that's 3 GPUs total. The K10 has 2 and the other has one (which is Fermi).
The K10 has 4GB for each GPU, the other has 6GB.
I don't think E@H would benefit much though because Tesla really shines with the double floating point precision which the app doesn't run, so hearing that it runs about like a 670 seems about right. I imagine the M2090 would be much faster than a 670 with the Fermi, but could be wrong.
Sorry Mike I wasn't clear.
The system with the K10 has 2 cards so 4 GPUs and also has 4 Xeon E5-4650 for 32 cores, hyperthreaded. Obviously not a home system, I can't even afford one of those CPUs let alone the stuff it takes to run it.
Anyone running on a Xeon or Tesla?
)
Are you a member over at the nVidia site?
If you don't get the info here you may be able to at their forum.
http://www.nvidia.com/page/home.html
I have seen some Xeon boxes
)
I have seen some Xeon boxes running when browsing profiles, I know we have that around. Got to be at least a Tesla running somewhere. I'm thinking though that the Tesla will not be a huge benefit to E@H for what it is.
I have access to a Tesla K10
)
I have access to a Tesla K10 and an M2090 (I think) but as it's a shared resource I can't set up Boinc on that machine.
Is there a way to run our BRP cuda apps one at a time just to see how long it takes. This thing has 4 GPUs with 4GB each so it can probably run maybe 32 or 48 wu simultaneously.
My tests so far show it about the same as my GTX670 for a single task. I'm not sure why but the bandwidth is not as good. Still working on that.
Joe
I believe that's 3 GPUs
)
I believe that's 3 GPUs total. The K10 has 2 and the other has one (which is Fermi).
The K10 has 4GB for each GPU, the other has 6GB.
I don't think E@H would benefit much though because Tesla really shines with the double floating point precision which the app doesn't run, so hearing that it runs about like a 670 seems about right. I imagine the M2090 would be much faster than a 670 with the Fermi, but could be wrong.
Dual Xeon 5150's (in Gen 1,1
)
Dual Xeon 5150's (in Gen 1,1 Mac Pro), 2 cores each @ 2.66GHz, about 6 hours running 4 tasks.
Dual Xeon 5365's (same machine after upgrade), 4 cores each @ 3.00 GHz, about 5 hours 45 minutes running 8 tasks.
Thanks for that. I had
)
Thanks for that.
I had wrongfully imagined that the Xeon would be blasting away even the 3rd gen i7 in crunching. Of course, even that latter Xeon is much older.
RE: Thanks for that. I
)
If you're referring to the Xeon 5365, that's even before the first Core iX series - way older, so it's not a fair comparison to the current iX series. It was powerful at that time (circa 2007), but also power-consuming, measuring up to 150W each.
I've got my old Mac Pro Xeon 5365 replaced with the one on 5550 variants (2009) running at the same amount of cores at 2.67GHz only (cooler 90W each), but Einstein crunches twice the performance of the 5365s.
Current Xeon units should be on par or better than the 3rd gen i7s, but then again, I've never got one in hand to crunch...
Yeah I was just wondering how
)
Yeah I was just wondering how they did crunching. Seems like i7 still the best CPU as far as CPU goes.
Yeah, these are older Xeons;
)
Yeah, these are older Xeons; the 5150's were standard in 2007 Mac Pros, and the 5365's were an optional high-end upgrade. 5150's are essentially free on eBay now; the 5365's cost about $150 each at a minimum because of demand from people with first generation Mac Pro's who want to squeeze more life from their machines.
They are indeed power-hungry and run hot; that machine running an ATI 5770 with two simultaneous BRP tasks and 5 CPU tasks eats up 450w with the CPU temps in the 65c range, and I'm in the process of retiring it as a cruncher. It's still a decent machine for everyday work - those two Xeon 5365's give me a CPU Passmark score of 8000.
RE: I believe that's 3 GPUs
)
Sorry Mike I wasn't clear.
The system with the K10 has 2 cards so 4 GPUs and also has 4 Xeon E5-4650 for 32 cores, hyperthreaded. Obviously not a home system, I can't even afford one of those CPUs let alone the stuff it takes to run it.
Joe