Dual core opteron 6128 with -bigadv

The most demanding Projects are only available to a small percentage of very high-end servers.

Moderators: Site Moderators, PandeGroup

Re: Dual core opteron 6128 with -bigadv

Postby Mizzou_Engineer » Sat Dec 11, 2010 4:00 pm

iohel, your system ought to be performing about twice as fast as it does. I have a nearly-identical system to yours (two 6128s, 8x2 GB DDR3-1333, ASUS KGPE-D16) and it has frame times of 28:30 on p6900s. 10e's suggestions are a good place to start.

bruce wrote:I just searched for information on the 6128 and was not able to find the information I'm looking for. AMD has been planning on combining a single FPU with two integer cores with multiple pipelines. Does that apply to the 6128? The fundamental question is not so much how many integer cores you have but how many FPUs you have.


The Magny-Cours Opterons are essentially the same as the Phenom II and are based on the same "Stars" (aka K10) core. The 6128 has eight integer cores, each with its own FPU, for 8 FPUs per CPU. The design you're talking about is the upcoming "Bulldozer" microarchitecture. There will be one FPU per pair of integer cores, but that one FPU is twice as wide as the current Stars' FPUs and can be split to issue one 128-bit FPU command per core per clock cycle, just like the current Stars FPU. The time when it acts like one FPU is when it is doing 256-bit FMA4 instructions, then it can process one FPU instruction from one core per clock cycle.

I have a pretty good understanding of Intel's i7 which has 8 integer cores and only 4 FPUs. Although it meets the minimum of eight cores required for bigadv, it has to run above about 2.2 GHz to meet the earlier deadlines -- and I don't know how they are doing with p6900. The intel i5 has only 4 integer cores but pretty much the same floating point performance, clock for clock.


The i7s do not have 8 cores, they have 8 threads and four cores. It just looks like 8 cores to the OS, which is why an unmodified client will run on those CPUs without balking. The Opteron 6128 actually has eight cores. The i5 700 series and the quad-core i7s are literally identical in core hardware- the i5 700s just do not have HyperThreading active to schedule a second thread on the exact same core resources, so the identical FP performance should not be surprising at all.

Recent reviews for both AMD and Intel often seem to focus on applications like encoding or server apps where the integer cores and/or high memory bandwidth are important without considering applications like FahCore_a3 that needs massive amounts of floating point work.


HPC does get a lot of press for server processors, and the FP power of upcoming CPUs should be higher than existing CPUs. The addition of 256-bit FMA3/FMA4 instructions in AVX will increase the SIMD FP math capabilities of the CPUs. We just haven't seen it yet since neither Intel's Sandy Bridge nor AMD's Bulldozer are officially released, and no programs out yet support AVX. Thus the recently-leaked Sandy Bridge benchmarks didn't even see that capability of the processor, and Bulldozer is tightly under AMD's wraps at the moment.
Mizzou_Engineer
 
Posts: 30
Joined: Tue Dec 18, 2007 3:30 pm

Re: Dual core opteron 6128 with -bigadv

Postby codysluder » Sat Dec 11, 2010 5:44 pm

The issue of "integer cores" vs. "threads" is an immense oversimplification, just like whether the bulldozer has a shared FPU or has one that is wider/splittable. At a much more basic level, it depends on how many instructions of the type that FAH issues that ALU or FPU is capable of processing simultaneously.

For the Intel fan-boys, the core2 architecture provided a large increase in the FPU throughput over previous generations. For the AMD fan-boys, bulldozer will save the world. To an unbiased observer, the only thing certain is that the current generation works better than the previous generation, and the next generation will again be better that the current one. (And anything you buy will be obsolete before you know it.)
codysluder
 
Posts: 2128
Joined: Sun Dec 02, 2007 12:43 pm

Re: Dual core opteron 6128 with -bigadv

Postby iiohel » Sun Dec 12, 2010 9:08 pm

It turns out the actual issue was a short in the case which was the last thing on my check list and the one that seemed most daunting. I finally can report I'm seeing 35-36 minute frame times and dancing with my rig to the tune of so happy together.
iiohel
 
Posts: 4
Joined: Wed Nov 24, 2010 3:00 pm

Re: Dual core opteron 6128 with -bigadv

Postby markfw » Fri Dec 21, 2012 2:59 am

Where are 2 amd 3 in the bios ? I can't find them.
markfw
 
Posts: 137
Joined: Mon Feb 04, 2008 3:32 pm

Re: Dual core opteron 6128 with -bigadv

Postby markfw » Fri Dec 21, 2012 3:43 am

OK, if finished (I hope I get the credit) but what concerns me, is that it downloaded another unit and it won't start for 2 days. Won't that kil by bonus, since it won't finish for 5 days ? and 2.4 is tha max ? or does it know wnen I actually start somehow on the V7 client ?
markfw
 
Posts: 137
Joined: Mon Feb 04, 2008 3:32 pm

Re: Dual core opteron 6128 with -bigadv

Postby bruce » Fri Dec 21, 2012 4:54 am

The duration of any WU begins when the WU is downloaded and ends when the result is uploaded. If that time is shorter than the "preferred deadline" (or simply "deadline") you get a bonus which increases inversely with the time it took you to return the result. When you pass the Preferred Deadline, the WU is assumed to be lost and it is reassigned to someone else but you still get the baseline credit if it is returned before the "final deadline" --- just no bonus, . After the Final Dealine, no credit is given.

Both the deadlines and the points depend on which project you have been assigned. See http://fah-web.stanford.edu/psummary.html
bruce
 
Posts: 22616
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Previous

Return to SMP with bigadv

Who is online

Users browsing this forum: No registered users and 1 guest

cron