Folding Forum

When i switched i thought there was going to be an improvement. I was wrong before i was getting 1% every 6-7minutes with linux/vm now im getting a % done every 8-10minutes, may this be because its still in beta i hope, another complaint even with the bounces i am not getting as much ppd, how much exactly i cannot say. Now another question why is it that on my Athlon 620@ 3.1 ghz is getting the same ppd as my phenom I at 2.6ghz, that's a .5ghz difference which in my opinion is quite substantial?

hootis wrote:When i switched i thought there was going to be an improvement. I was wrong before i was getting 1% every 6-7minutes with linux/vm now im getting a % done every 8-10minutes, may this be because its still in beta i hope, another complaint even with the bounces i am not getting as much ppd, how much exactly i cannot say. Now another question why is it that on my Athlon 620@ 3.1 ghz is getting the same ppd as my phenom I at 2.6ghz, that's a .5ghz difference which in my opinion is quite substantial?

Athlon x4 have no L3 cache so that will be a factor in speed. RE PPD how are you calculating PPD - HFM.net?

Nathan_P wrote:

hootis wrote:When i switched i thought there was going to be an improvement. I was wrong before i was getting 1% every 6-7minutes with linux/vm now im getting a % done every 8-10minutes, may this be because its still in beta i hope, another complaint even with the bounces i am not getting as much ppd, how much exactly i cannot say. Now another question why is it that on my Athlon 620@ 3.1 ghz is getting the same ppd as my phenom I at 2.6ghz, that's a .5ghz difference which in my opinion is quite substantial?

Athlon x4 have no L3 cache so that will be a factor in speed. RE PPD how are you calculating PPD - HFM.net?

estimating project time points and multiplier as fahmon doesn't use the multiplier yet. whats hfm.net

A more optioned version of fahmon

http://code.google.com/p/hfm-net/

Be sure to enable the option for calculating bonuses.

The first thing I see is you are running ATI graphics and unfortunately ATI cards on an AMD setup suck the life out of the A3's. The best thing I have found to do with the AMD / ATI set-up is run the A3 -smp 3 and devote core 01&2 to A3 and core 3 to the graphics use the Variables and give the ATI graphics a normal priority setting and the a3 a low priority setting.

so if i run on only 3 cores it will be faster than 15-20% on the fourth core?
I have the level 3 cash enabled via (ACC)...
this is about how much cpu usage the card is getting, unfortunately i did not buy these computers for the purpose of folding, rather cheap gaming systems but folding works too, otherwise i would have gotten 1 295 that could do as much ppd as my to other comps
current ppd 1650 with bounces as the 3840ppd i was getting with a2/linux/vm, although this is much better than core a1 and i am thankful for that.

On my AMD / ATI rig A Phenom II 940 @ 3.4Ghz and a 3870X2 running with the variables and settings listed below I am averaging 7500 PPD

BROOK_YIELD = 2
CAL_NO_FLUSH = 1
CAL_PRE_FLUSH = 1
FLUSH_INTERVAL = 128

-smp 3
devote core 01&2 to A3
core 3 to the graphics
ATI graphics a normal priority setting
the a3 a low priority setting
You can use a program like prefinity to set the priority's http://edgemeal.110mb.com/PriFinitty/index.htm

my flush interval is set 256, its going slightly faster with only 3 cores used

hootis wrote:so if i run on only 3 cores it will be faster than 15-20% on the fourth core?
I have the level 3 cash enabled via (ACC)...
this is about how much cpu usage the card is getting, unfortunately i did not buy these computers for the purpose of folding, rather cheap gaming systems but folding works too, otherwise i would have gotten 1 295 that could do as much ppd as my to other comps
current ppd 1650 with bounces as the 3840ppd i was getting with a2/linux/vm, although this is much better than core a1 and i am thankful for that.

This isn't entirely clear yet. There are two versions of FahCore_a3 floating around and they behave slightly differntly. The older version would absolutely choke if you tried to run with 3.2 cores (assuming you're using 80% of the 4th one for your GPUs) and -smp 3 would be much faster. With the newer version, I don't yet know whether the other 15-20% would be helpful or not. A few people have suggested that stopping your GPU(s) and devoting that extra core to SMP might actually improve your PPD, but surely that depends on may factors and there is no official recommendation. It's certainly worth a try. If it's not for you, then you can always switch back.

In general, folks are finding that the A3 solution is generally equal to or better (and much simpler) than the A2/Linux/VM solution on Windows once the bonus factors are included.

Certainly the goal of the Stanford optimization is for the effective use of everything that's available. Since the A3 core is relatively new there will likely be additional changes/optimization still being developed.

i already switched both my machines back and then got a1's.....