• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

-smp 7 not recommended going forward...

T

Tobit

[H]ard|DCer of the Month - March 2010/May 2011
Joined
Dec 12, 2005
Messages
5,145
A new A3 SMP project recently emerged in beta testing - P7500 - where people running -smp 7 were unable to fold said unit. Dr. Kasson replied:

kasson said:
Hmm--the core shouldn't allow SMP 7 in the first place. Please don't run SMP 7--it's a bad idea for WU stability. I'll talk to the core developers about removing it.
Click to expand...

I then asked for more clarification:

Tobit said:
Dr. Kasson - Could you please make a public statement regarding this? I know of at least two dozen donors who fold with -smp 7 but this is the first I've heard of this. I know there were issues in the past with large prime numbers but -smp 7 is hardly large. I'd really like to be educated on this please. I've performed a search here on FF but have found no other posts relative to this.
Click to expand...

Dr. Kasson just answered me now on this:

kasson said:
The problem is that Gromacs does a 2D decomposition based on factoring the number of threads you give it. So if you give it a number like 7 or 13, the best it can do is 7x1 or 13x1, whereas 12 can yield 4x3 and 8 can yield 4x2. The more thinly the system gets broken up, the more likely it is to fail. Hence easily factorable numbers are better...

Most of the early A3 SMP work units we've done involve relatively large systems (particularly many of the influenza-related projects I have). So I think we may have had an early bias towards systems where one could get away with less-stable settings more of the time.
Click to expand...
 
So would smp -6 be better? It'd be 3x2 that way or does it need to be a multiple of 4? If that's the case, what about using smp on dual cores or tri-cores? Is that done for?
 
Well, this shouldn't be an issue for -bigadv units, so I don't think they should take it out. I get much better performance running -smp 7 than I do with just -smp, but I don't fold normal SMP units on this machine.
 
[H]ugh_Freak;1037230611 said:
So would smp -6 be better? It'd be 3x2 that way or does it need to be a multiple of 4? If that's the case, what about using smp on dual cores or tri-cores? Is that done for?
Click to expand...
I know Dr. Kasson had said -smp 3 and -smp 5 were fine. Let me check with him on -smp6 and I will report back.
 
Zero82z said:
Well, this shouldn't be an issue for -bigadv units, so I don't think they should take it out. I get much better performance running -smp 7 than I do with just -smp, but I don't fold normal SMP units on this machine.
Click to expand...

I don't run -bigadv, I have done a couple of the -bigadv units though, just to see what the points would be like, and I don't like the fact I lose about 4000-5000 points just because it takes an hour to upload the results. Plus, in losing the points, its only about a 2K point difference between -bigadv and normal -smp. I'll take the fewer points and not have it hog the bandwidth for an hour.
 
a3 isn't big advc, they are regular smp units. a5 are big units

Theres a few units like this, that a uneven number cause instability.
 
[H]ugh_Freak;1037233753 said:
I don't run -bigadv, I have done a couple of the -bigadv units though, just to see what the points would be like, and I don't like the fact I lose about 4000-5000 points just because it takes an hour to upload the results. Plus, in losing the points, its only about a 2K point difference between -bigadv and normal -smp. I'll take the fewer points and not have it hog the bandwidth for an hour.
Click to expand...
Well, that's your choice, but I don't see what it has to do with my comment. I'm not saying you should or shouldn't use -bigadv, I'm saying that for people who run -bigadv units, the -smp 7 option should be left in.
Patonb said:
a3 isn't big advc, they are regular smp units. a5 are big units
Click to expand...
True, but since the A5 core is a tweaked version of the A3 core, they might decide to carry this over.
 
with an dedicated i7 930 and a gtx460, I got better production without -smp 7 than with it. It may be different story if you are using two GPUs though.
 
Yes. I have run two different setups over the past year that involved multiple GPUs and a 4 core/8 thread CPU, and running smp 7 bigadv is clearly the optimal configuration.

In fact, I have been considering building basically a GPU folding box with an i7 CPU to put in the garage this summer. I will be disappointed if they take away the option to run smp 7 bigadv on the CPU.
 
Zero82z said:
Well, that's your choice, but I don't see what it has to do with my comment. I'm not saying you should or shouldn't use -bigadv, I'm saying that for people who run -bigadv units, the -smp 7 option should be left in.
Click to expand...

Sorry must have misunderstood what you where saying since this has nothing to do with bigadv. It only affects those of us running normal smp.
 
[H]ugh_Freak;1037235175 said:
Sorry must have misunderstood what you where saying since this has nothing to do with bigadv. It only affects those of us running normal smp.
Click to expand...
Unless they take the option out of the A5 core as well, in which case it will affect people who fold -bigadv units as well.
 
The bigadvc units are now not possible on anything less than smp 8, so it really is a mute thing.

Which blows as it cripples 8 core cpu and gpu folders.
 
Patonb said:
The bigadvc units are now not possible on anything less than smp 8, so it really is a mute thing.
Click to expand...
That's incorrect. The client will only download -bigadv units if it detects a CPU with 8 or more threads, but it can still be configured to use only some of them.
 
Zero82z said:
That's incorrect. The client will only download -bigadv units if it detects a CPU with 8 or more threads, but it can still be configured to use only some of them.
Click to expand...

This, I run SMP-7 on my 2600k even when not gpu folding. It makes the machine output much less variable with my daily use.
 
Zero82z said:
That's incorrect. The client will only download -bigadv units if it detects a CPU with 8 or more threads, but it can still be configured to use only some of them.
Click to expand...

Not on the V7 client - if you configure it for less than 8 cores it will not download a bigadv. On V6 -smp 7 will continue to work, but may blow up on some WUs.

H.
 
Haitch said:
Not on the V7 client - if you configure it for less than 8 cores it will not download a bigadv. On V6 -smp 7 will continue to work, but may blow up on some WUs.

H.
Click to expand...
Perhaps, but at this point, I really don't care about v7.
 
You must log in or register to reply here.
Back
Top