Belated parp @ Senture
- Thread starter miniyazz
- Start date
Associate
- Joined
- 23 Aug 2007
- Posts
- 1,699
- Location
- Rothesay
Don't get too comfortable there, you may have a very short lease on the place
them sounds like fighting words
Bugger. However having clocked back my i7 slightly, I can run it on all cores without turning my room into a tornado. So my 24-hour average is reported as 20,842.1 and should be a fair bit higher than that as I'm on a rubbish P6701 at the moment, so I'm hoping victory will be mine although I may lose the next battle
Yeah them bigadv units are nice . Is P6701 as useless as P6702? The performance of P6702 on the 8 core beast was shocking. Took about 20 hours to complete and gave a lousy 6k PPD. Even the low clocked 2.66 quad core Xeon can do one in a little over a day giving just under 4k PPD so something was seriously wrong there.
. Is P6701 as useless as P6702? The performance of P6702 on the 8 core beast was shocking. Took about 20 hours to complete and gave a lousy 6k PPD. Even the low clocked 2.66 quad core Xeon can do one in a little over a day giving just under 4k PPD so something was seriously wrong there.6k PPD on an 8-core?
I don't know much about the 6702s tbh, I've not come across them before so can't compare. The 6701s can knock 5-6k+ (40%) off my PPD compared to most of the other units though, I presume the 6702s are similar.
I don't know much about the 6702s tbh, I've not come across them before so can't compare. The 6701s can knock 5-6k+ (40%) off my PPD compared to most of the other units though, I presume the 6702s are similar.
I would guess so. Just looked at the HFM benchmark, the comparison draws up some very strange goings on.
Dual X5450 (3 Ghz quad core Xeon, 1333 FSB 12MB cache 45nm) came up with this:
Number of Frames Observed: 99
Min. Time / Frame : 00:10:35 - 7,250.0 PPD
Avg. Time / Frame : 00:11:22 - 6,513.6 PPD
Yet the other machine (Dual 5150 2.66Ghz dual core Xeon 1333 FSB 4MB cache 65nm) is currently processing one:
Number of Frames Observed: 85
Min. Time / Frame : 00:15:49 - 3,968.2 PPD
Avg. Time / Frame : 00:15:52 - 3,949.5 PPD
Cur. Time / Frame : 00:15:53 - 3,942.4 PPD
R3F. Time / Frame : 00:15:51 - 3,950.7 PPD
All Time / Frame : 00:15:52 - 3,946.6 PPD
Eff. Time / Frame : 00:15:53 - 3,942.4 PPD
Something way obviously wrong with the 8 core setup, it should have had about 2.5x the processing power. Both of those are P6702
Dual X5450 (3 Ghz quad core Xeon, 1333 FSB 12MB cache 45nm) came up with this:
Number of Frames Observed: 99
Min. Time / Frame : 00:10:35 - 7,250.0 PPD
Avg. Time / Frame : 00:11:22 - 6,513.6 PPD
Yet the other machine (Dual 5150 2.66Ghz dual core Xeon 1333 FSB 4MB cache 65nm) is currently processing one:
Number of Frames Observed: 85
Min. Time / Frame : 00:15:49 - 3,968.2 PPD
Avg. Time / Frame : 00:15:52 - 3,949.5 PPD
Cur. Time / Frame : 00:15:53 - 3,942.4 PPD
R3F. Time / Frame : 00:15:51 - 3,950.7 PPD
All Time / Frame : 00:15:52 - 3,946.6 PPD
Eff. Time / Frame : 00:15:53 - 3,942.4 PPD
Something way obviously wrong with the 8 core setup, it should have had about 2.5x the processing power. Both of those are P6702
Weird 
TBH your second machine looks oddly fast though - am I right in thinking it's two dual-core CPUs, no hyperthreading? If the 6702s are similar to the 6701s, well my i7 running on 7 cores @ 4.2GHz has had as quick as 7:30ish/frame. If I were running at 2.66GHz that would give me 11:50 frame times on 7 cores, and if mine used only 4 cores it'd be about 20:40 - quite a lot longer than your frame times.
I know mine uses hyperthreading which isn't near to 100% efficient, but equally I thought the Nehalem architecture was more efficient than the Xeons, roughly cancelling out that difference - certainly I wouldn't expect a 25% difference in frame times taking both into account.
But there's a lot of assumptions in there so I may be talking out of my bum
TBH your second machine looks oddly fast though - am I right in thinking it's two dual-core CPUs, no hyperthreading? If the 6702s are similar to the 6701s, well my i7 running on 7 cores @ 4.2GHz has had as quick as 7:30ish/frame. If I were running at 2.66GHz that would give me 11:50 frame times on 7 cores, and if mine used only 4 cores it'd be about 20:40 - quite a lot longer than your frame times.
I know mine uses hyperthreading which isn't near to 100% efficient, but equally I thought the Nehalem architecture was more efficient than the Xeons, roughly cancelling out that difference - certainly I wouldn't expect a 25% difference in frame times taking both into account.
But there's a lot of assumptions in there so I may be talking out of my bum
Well you know what they say about assumptions "it makes an ass out of u and me"
None of the machines here have hyperthreading and indeed it is 2 dual cores. Looking at it from your frame times, it seems the 8 core beast was performing well below par but the 2x dual seems to be over performing. Biffa's TPF on bigadv was a little slower than mine (48 mins i7 @ 4ghz -smp 7 vs 43:30 2x X5450 @ 3.2ghz -smp 8) so I'm not sure if the true 8 core is faster or not than a well clocked i7.
Switched the GPU from GPU3 over to GPU2, PPD has done from 6k on the GPU3 client to 8.7k GPU2 (353 point unit)
None of the machines here have hyperthreading and indeed it is 2 dual cores. Looking at it from your frame times, it seems the 8 core beast was performing well below par but the 2x dual seems to be over performing
. Biffa's TPF on bigadv was a little slower than mine (48 mins i7 @ 4ghz -smp 7 vs 43:30 2x X5450 @ 3.2ghz -smp 8) so I'm not sure if the true 8 core is faster or not than a well clocked i7.Switched the GPU from GPU3 over to GPU2, PPD has done from 6k on the GPU3 client to 8.7k GPU2 (353 point unit)
Nice, quite an improvement
Looks like I'm on GPU2 (Core reported as GROGPU2) so no such PPD boosts for me
Hmm, thinking about it I have a feeling someone somewhere said hyperthreading on the i7s gives about 70% more performance than not having it, so the frame times could work out.. either way, I'd expect the 8 physical cores to beat out 8 virtual cores clock for clock. It's hard to compare speeds when one uses 7 cores and the other uses 8 cores though, as you can't subtract the speed linearly because other running programs and processes eat into the eighth core. So much so that I switched to -smp 7 because I was actually getting longer frame times with -smp 8 - presumably the inefficiencies introduced by only having part of a core folding affect things somehow.
Looks like I'm on GPU2 (Core reported as GROGPU2) so no such PPD boosts for me
Hmm, thinking about it I have a feeling someone somewhere said hyperthreading on the i7s gives about 70% more performance than not having it, so the frame times could work out.. either way, I'd expect the 8 physical cores to beat out 8 virtual cores clock for clock. It's hard to compare speeds when one uses 7 cores and the other uses 8 cores though, as you can't subtract the speed linearly because other running programs and processes eat into the eighth core. So much so that I switched to -smp 7 because I was actually getting longer frame times with -smp 8 - presumably the inefficiencies introduced by only having part of a core folding affect things somehow.
It does depend what sort of cpu cores they are.
There is also some sort of primary numbers bug in the new 2.22 A3 core which can screw with output if you use SMP 3/5/7 with lockups and all sorts of mess going on.
I don't get it with my rigs, weebeastie is running CPU only at the moment and gets 16K PPD on a regular A3 WU, and nearly 18K on a bigadv and thats at just 3.6Ghz.
My regular home machine can't even run an A3 unit without choking down to sub 1K PPD and it doesn't matter if its at stock or at any speed up to 4.2Ghz but the GTX470 is fine in it without the CPU running. With the CPU running the gfx card starts playing up as well
I'm thinking of flashing the latest bios, setting it at bios defaults and see what it does.
Then if its ok with CPU only at 2.66Ghz just slowly bumping it up and see what happens, if it starts to freak out again maybe an OS reinstall or a motherboard change.
There is also some sort of primary numbers bug in the new 2.22 A3 core which can screw with output if you use SMP 3/5/7 with lockups and all sorts of mess going on.
I don't get it with my rigs, weebeastie is running CPU only at the moment and gets 16K PPD on a regular A3 WU, and nearly 18K on a bigadv and thats at just 3.6Ghz.
My regular home machine can't even run an A3 unit without choking down to sub 1K PPD and it doesn't matter if its at stock or at any speed up to 4.2Ghz
but the GTX470 is fine in it without the CPU running. With the CPU running the gfx card starts playing up as well I'm thinking of flashing the latest bios, setting it at bios defaults and see what it does.
Then if its ok with CPU only at 2.66Ghz just slowly bumping it up and see what happens, if it starts to freak out again maybe an OS reinstall or a motherboard change.