on the q6600?

[JinJohn1966]

I am thinking of making a purchase of a q6600. I am wondering also, on the list of cores running the SMP, the 4 cores is not twice as fast as the 2 cores that are at the same speed...i know each instance would be slower, but would the ppd be higher with two SMP's running with one on each chip?

 

[fenderltd]

What are you trying to compare exactly? I am pretty busy at work and mind is else where, but I cannot say I understand your question. Maybe this will help

http://members.cox.net/foldinginfo/smp2.htm

 

[Xilikon]

The problem here is that the SMP client cannot be run on 2 of the 4 cores as it will take all the cores, 4 in this case. The mins/frame difference is double but only using similar clockspeeds. Those who have mins/frame faster on dual-cores is usually clocked higher like fenderlt.

Overclock the Q6600 to 3.2-3.4 GHz and you will be seeing 6-7 mins / frame, exactly half of the 3.2 GHz E6600.

when doing comparaisons, always take into consideration the clockspeed, fsb speed, memory speed and memory timimgs. My measly E4300 can do 10-15 mins / frame depending on the WU, on par with the E6420/E6600 with 4 Mb of cache, mainly because my RAM is very tight and it's running at 3.3 GHz.

Keep in mind that a shorter mins / frame = higher PPD.

 

[MN Scout]

JinJohn, with a quad-core you'd only have to install 1 smp client. It'll happily use all 4 cores, and your frame times will be cut in half compared to just a dual core.

 

[JinJohn1966]

I was talking as of this:

Intel E6700 1 Core2Duo Conroe 2 3.500 4 MB Jbavousett Windows SMP P2610 1523 0:11:51 19:45:50 1849.4 528.4 19:39:59 19:51:33 0:11:34 19:51:33 20:03:42 0:12:09

Intel Q6600 1 Core2Duo Kentsfield 4 3.500 8 MB 4x2 DriveEuro Windows SMP P2651 1760 0:08:05 13:28:20 3135.3 447.9 21:43:41 21:51:44 0:08:03 22:07:47 22:15:54 0:08:07

See the bold? The e6700 is oc to the same speed as the q6600...but the q6600 is not twice as fast...it need to be just over 5 min per frame. I knows. different portiens. but look here.

Intel E6700 1 Core2Duo Conroe 2 3.400 4 MB mklinger Windows SMP P2651 1760 0:12:28 20:46:40 2032.9 597.9 0:00:00 0:00:00

This machine is 100Mhz slower, and the time is not even twice the fram of the other CPU. According too these numbers, two of mklinger's machines would make more ppd's than the single quad...wich would have a higer total MHZ's.

 

[Xilikon]

I was talking as of this:





See the bold? The e6700 is oc to the same speed as the q6600...but the q6600 is not twice as fast...it need to be just over 5 min per frame. I knows. different portiens. but look here.



This machine is 100Mhz slower, and the time is not even twice the fram of the other CPU. According too these numbers, two of mklinger's machines would make more ppd's than the single quad...wich would have a higer total MHZ's.

Be careful, never compare the times with 2 different WU. 2651 and 2610 is different

As I said, it also depends on things beside the CPU speed like RAM speed/timings, motherboard chipset and such.

 

[Killer[MoB]]

snip---- According too these numbers, two of mklinger's machines would make more ppd's than the single quad...wich would have a higer total MHZ's.

You are correct, the quad will not be twice as fast. None of the data I have seen since the two SMP clients came out show the quad being twice as fast as a C2D at the same clock speed. I'm pretty sure it has something to do with the way the current intel quads are designed and the way the two pairs of CPU cores communicate with each other and data is transfered between the 2 x 4MB L2 caches. They are actually 2 dual core cores on the same chip, not a true quad core core. A true quad core (when they come out) with 8 MB of cache should be 2x as fast or faster since hopefully advancements will have been made in core design.

 

[JinJohn1966]

So if by setting the smp cores by using affinnity, could I perhaps gain some PPD by running two SMP's on the q6600? Just wondering if it would be more ifficient.

 

[Killer[MoB]]

It is possible, I have had the same theory but nothing to test it on.
The way i see it, you set affinity for each folding client to use the CPU cores that share the same actual die & cache and in theory, it should be like having 2 dual core SMP setups in one box. I don't know if it would be CPU0 and CPU2 as a pair or CPU0 and CPU1 as a pair. I'll see what I can find out.

 

[charpman]

So if by setting the smp cores by using affinnity, could I perhaps gain some PPD by running two SMP's on the q6600? Just wondering if it would be more ifficient.

I actually just read something about this on the forums at stanford... the jist of it is, yes it's possible and you'll get more points but return WU's slower. What matters more to the project is getting the WU's back faster.

To sum up it basically said:

Best for points - do this.
Best for science - do not do this, only run 1 SMP client.

 

[Killer[MoB]]

http://www.hardocp.com/image.html?image=MTE2MjM1NzYzNHkwVzRWaTZqUUpfMV8yX2wuanBn

Core pairs are CPU 0 & 1 and CPU 2 & 3.

 

[JinJohn1966]

Well...seeing that they are to making points that are for better science worth less, i have not a problem taking a few hours longer with turning points in. they are saying how much better the PS3 sciences is...yet the points are...How do you say in english - hors proportion? I have one working now, and it is to make only less than 1000ppd. so when the points are of a value equal to the sciences, i will to worry about making them in faster.

 

[charpman]

Well...seeing that they are to making points that are for better science worth less, i have not a problem taking a few hours longer with turning points in. they are saying how much better the PS3 sciences is...yet the points are...How do you say in english - hors proportion? I have one working now, and it is to make only less than 1000ppd. so when the points are of a value equal to the sciences, i will to worry about making them in faster.

Ya it's a valid argument.

 

[Tigerbiten]

As long as you can get the work back before the first deadline then that's a valid argument.

If running two SMP clients causes you to miss the first deadline where as only running one client you could have made it then the argument has less strength.
If you miss the first deadline then that work-unit will get reasigned to someone else to fold.
As long as you make the final deadline then you get the points for folding it but a lot less science is being done.

The only reason I ran both the SMP client & the GPU client on my C2D is that I could get both back before their first deadlines by setting the affinity.

Luck .........