New bigadv project - 2685

have both set on mine. thought i dont always get a bigadv. ill remove advmethods anyway see if it helps.
 
Last edited:
You should only use one or the other. As for P2685, I'm seeing between 23k and 27k PPD depending on load. Quite an improvement over standard A3s; especially 6701 and 6702.
 
I have both advmethods and bigadv set on at least 2 of my machines. All 5 of them are running 2684's at the moment, so you definately can get bigadv units with the advmethods flag set. I was playing around with setting advmethods and not setting it when one of my I7s refused to pull a bigadv unit. The other machine with both set was more of a screw-up on my part, but it immediately picked up a bigadv when I originally set the bigadv flag, so I didn't mess with it beyond that.

27K was about what I was getting on my faster I7s with A2 bigadv. This was under native Linux, which is probably why you guys are seeing a little better numbers than you used to. This is still a great sign!
 
I am seeing performance between the P2684 and the A2 WUs, so once again my results differ from most people. On my fastest machine, a dual Harpertown @2.9GHz, the TPF is 46:46 for 19,256 PPD. The same machine running native Linux and A2 WUs was accomplishing 22,000 PPD @2.8GHz. There are no GPU clients. Can anyone explain this?
 
APOLLO said:
I am seeing performance between the P2684 and the A2 WUs, so once again my results differ from most people. On my fastest machine, a dual Harpertown @2.9GHz, the TPF is 46:46 for 19,256 PPD. The same machine running native Linux and A2 WUs was accomplishing 22,000 PPD @2.8GHz. There are no GPU clients. Can anyone explain this?
Click to expand...


probably something to do with the lga-771/5 architecture and multi-cpu along with windows.. it could be that the A3's are much more sensitive to ram clocks/bandwidth where the i7's benefit more from then the old LGA-771/5..
 
APOLLO said:
. Can anyone explain this?
Click to expand...
seems to me it's gotta be a function of the architecture. Your harpertowns are probably hamstrung a bit by memory bandwidth.....their data has to travel over the FSB through the northbridge then to the FB-DIMMS, whereas the i7 architecture has the qpi link, with integrated memory controller, and triple channel 1600mhz DDR3.
 
sirmonkey1985 said:
probably something to do with the lga-771/5 architecture and multi-cpu along with windows.. it could be that the A3's are much more sensitive to ram clocks/bandwidth where the i7's benefit more from then the old LGA-771/5..
Click to expand...

jebo_4jc said:
seems to me it's gotta be a function of the architecture. Your harpertowns are probably hamstrung a bit by memory bandwidth.....their data has to travel over the FSB through the northbridge then to the FB-DIMMS, whereas the i7 architecture has the qpi link, with integrated memory controller, and triple channel 1600mhz DDR3.
Click to expand...
Yeah, I was afraid of this possibility. These machines were doing fine with A2 WUs and they had massive memory footprints. :confused:
 
APOLLO said:
Yeah, I was afraid of this possibility. These machines were doing fine with A2 WUs and they had massive memory footprints. :confused:
Click to expand...
stabbing in the dark here, but maybe that actually makes sense:

It could be that these new WUs, with their lower memory footprint, are optimized for the i7 architecture, where less information is held in RAM but more information is passed between the CPUs directly.
 
I don't know about that. Everything I have seen indicates that these units actually see less benefit from higher memory speeds and lower timings with i7 rigs, so that leads me to believe that memory bandwidth is less important than with the older A2 units.
 
jebo_4jc said:
stabbing in the dark here, but maybe that actually makes sense:

It could be that these new WUs, with their lower memory footprint, are optimized for the i7 architecture, where less information is held in RAM but more information is passed between the CPUs directly.
Click to expand...
That makes sense in a roundabout way. Well, if Stanford optimized for new architectures with superior interconnects between processors, wouldn't the Hypertransport of older AMD dual sockets benefit also, or is this really optimized for Nehalem and Nehalem alone? If the latter, then I'm not sure if I want to own other architectures or even consider it. Sort of reminds me of the past QMD situation, albeit not precisely the same.
 
APOLLO said:
That makes sense in a roundabout way. Well, if Stanford optimized for new architectures with superior interconnects between processors, wouldn't the Hypertransport of older AMD dual sockets benefit also, or is this really optimized for Nehalem and Nehalem alone? If the latter, then I'm not sure if I want to own other architectures or even consider it. Sort of reminds me of the past QMD situation, albeit not precisely the same.
Click to expand...


mm no because the hypertransport is only used for communication between the cpu chipset and gpu's.. has no effect on anything else.. tested it with my phenom II 940 @ 1.8ghz(3.6ghz HT stock) and 1Ghz(2ghz) and there was 0 difference in performance.. the biggest thing is the NB because it runs at the same speed as the HTT but are 2 different clocks.. but again is meaningless with the phenom II's.. NB clocks only show a benefit in synthetic benchmarks.. in the real world application theres almost nothing..
 
OK, what about the OS? How much does that play into performance with SMP? I'm using 32-bit XP but I can install 64-bit Win 7 if that will make any difference. There's only 2GB of RAM installed, though, for higher OCing and generally more stability (less FB-DIMM heat). I won't bother to install Win 7 if there's no benefit.
 
Zero82z said:
HT is also used for communication between CPUs in multi-socket systems.
Click to expand...


correct in that scenerio it does.. but in single processor setups it doesnt.. but its a pretty big bottleneck on multi socket AMD systems except i think the new g34 boards they changed the way the HTT communicates between the processors..
 
I just got back from being in the middle of nowhere and came across this thread. Looks like I'm going back to bigadv when my current WU finishes in the morning.
 
has availability been stepped up yet?

If it has I'll switch over skynet thursday if I hit 4Ghz but i still dont wamma go big unless my i7 will only get these. I'll take a points reduction in order to avoid nearly 4 day long WUs, too many ways to lose units this summer and i dont want all that down time
 
I just started folding for team 33. I am using the rig in the sig at stock speeds. This is the first time I have used this client. I used to fold for the Crysis online team, then Strategy Informer.

My first workload is a 2685. According to HFM, my TPF is 00:25:36 and PPD is 47543.116. Does that sound about right for a 980X rig at stock speeds?
 
Kwisatz Haderach said:
I just started folding for team 33. I am using the rig in the sig at stock speeds. This is the first time I have used this client. I used to fold for the Crysis online team, then Strategy Informer.

My first workload is a 2685. According to HFM, my TPF is 00:25:36 and PPD is 47543.116. Does that sound about right for a 980X rig at stock speeds?
Click to expand...
Welcome to the [H]orde! :cool:

Your TPF actually is quite good, very good I would say. What frequency?
 
Welcome aboard!
Posted via [H] Mobile Device
 
Nice numbers! I wish I could get ahold of one of these. My last 3 were all 2684's.
 
this thing will run 215X19 but the h50 cant handle the heat. someone said these new bigadv dont care much about memory bandwidth. thats 100% false. All a3's like memory bandwidth and high bus speeds. the higher you get them the more points per mhz. 210X19 vs 200X20 will be 3000ppd faster doing these 2685 2000ppd doing a 2684 +- a few hundred ppd. with my i5-750 running 215X19 i was getting almost 10,000 ppd doing a 6701 8:34 fram times. all memory running 1600 getting the memory to 2000ddr is worth another 5% +- a hair.
 
Last edited:
hmmmm higher bus means higer scores? I'm real close to 210 right now, good to know. Not sure my memory will hold at 2:8 but will 2:6 be a major drawback?
 
Parja said:
980X on an RE3? Come on, crank that thing up! It's begging you! With those CPUs, 4GHz is a starting point.
Click to expand...

I'll second that as I can easily get to 4.2 ghz on air.

That reminds me that I need to crank it back up again as I've been running at stock speed to chase out some gremlins and install Win 7.
 
whats the equation for finding mem speed again?
 
Kwisatz Haderach said:
I have 2000mhz memory, but I like my temps at 60C or less while I folding.
Click to expand...

Welcome to the team! You could do what TJ did which is just get one of those bigass Noctua NH-D14 coolers. Back when he was running at 4.2ghz and folding 2 bigadv clients on 2 separate VM's at the same time, he was still staying under 70 degrees.

Those Gulftowns never cease to amaze me with how powerful they are for heavily threaded apps like folding.
 
You must log in or register to reply here.
Back
Top