First time poster, can you check my setup?

[hungyzerglin]

This is my first time setting up and using the SMP F@H client, previously I used the single CPU version (for like a week before reading about the SMP client). My system specs are AMD 1090T, Radeon 6850, 8GB DDR3 (included in images below). Currently I'm using the latest beta SMP client (6.34). Couple questions for you guys:

1) I have the flags -SMP 6 -advmethods -verbosity 9 set. After reading mixed reviews on -forceasm, should I include this flag as well (also is -forceSSE no longer a working flag for AMD chips)?
2) I understand -bigadv seems to work only on Intel i7 CPUs, I can only use this flag if I emulate an 8 core VM?.
3) Are -SMP 6 and SMP pretty much the same for my cpu (I guess I should run -SMP 5 if I want to run a GPU client)?
4) I haven't folded very much (probably less than 2 WU), so my PPD with HFM shows only about 600-800PPD. This will increase once I've completed like 7-8 WU right?
5) Aside from the above flags, is there anything I am missing that will help increase my folding?

Thanks for your time in reading all that.

[patonb]

You seem to still be running a unicore unit. You'll need to let this one finish and see if it will pick up an a3 core, but it looks okay.

[7im]

Hello hungyzerglin, welcome to the forum.

1. -forceasm is hard coded to ON in the SMP client. -forceSSE is not used any more.
2. -bigadv is for 8+ core systems (according to the OS).
3. For a 6 core system, -smp and -smp 6 are the same. -smp 5 depends on your system and which GPU you run. That ones harder to answer, you usually just have to see for yourself which one produces more points, SMP or SMP+GPU.
4. Yes, but only if you have the latest HFM software, and have HFM configured to show the bonus. And you must complete 10 SMP WUs for the bonus to kick in.
5. Missing? If you fold on the AMD GPU, look in to using environment variables so the GPU client doesn't waste a lot of CPU cycles, per the GPU install guide.

[HendricksSA]

Hungyzerglin, welcome to the Fold. It is a particular delight to say "Hi" to another X6 user. And it is really nice to get all the info needed to get started on questions with the very first post. To add to what 7im said:

Question 3. With an ATI GPU you will probably want to run SMP 5 as the ATIs take a lot of CPU time. The new 6.34 client maps larger smp numbers to non-prime numbers of threads. The mapping NT from X to X appears in your SMP log and you can tell the number of threads used there. 6.34 is new and I don't know where the remapping starts. This would apply to the Intel crowd as well.

This should crank right along processing SMP. Getting your GPU running with it will take some fine tuning. You will have to experiment with various smp X settings and the ATI environment variables to maximize your points production. Come back here if you have any questions. Welcome aboard and good luck!

EDIT to fix the thread number paragraph. Apparently the remapping begins above 8 threads. Thanks, fuzzy!

[hungyzerglin]

Thanks for the responses guys. Just finished my first work unit, and HFM is showing a higher PPD value for this 2nd WU (~5000PPD). Seems like I picked up an A3 Core, as the console client required FahCore_a3.exe. Hopefully once I finish 10 WU's I'll see a nice boost in PPD. Will update my PPD once I hit 10 WU's. Quick question: The client is pulling 100% of cpu draw right now, which is to be expected. Since this isn't a dedicated F@H box, what's the best way to fold while still being able to use the machine for gaming and whatnot. Should I run it -SMP 3 and only let it use 3 cores, set cpu usage requested to a lower percent (maybe 50), or do a combination of the two? I guess I can always reconfig and restart the client at max settings when I'm not going to use the system for an extended period. Thanks again and it's great to be a part of the fold!

[uncle fuzzy]

That said, I think the new 6.34 client maps odd smp numbers to evens. If that is the case you will see the mapping NT from 6 to 4 appear in your SMP log and you will actually only be using 4 threads. 6.34 is so new I've not seen many folks talking about this. This would apply to the I7 crowd as well.

My i7's show "Mapping NT from 7 to 7" with 6.34 and multilpe GPU3 clients.

[uncle fuzzy]

Quick question: The client is pulling 100% of cpu draw right now, which is to be expected. Since this isn't a dedicated F@H box, what's the best way to fold while still being able to use the machine for gaming and whatnot.

Stop the SMP client (Ctrl+C) and reconfigure. Make sure that core priority is idle. I'd also try using -smp 5. Even if it only ends up using 4 cores, that leaves more free for you to use. A quad with a decent clock will still run pretty good PPD (Q6600@3.4 gets 5-7K).

[hungyzerglin]

That said, I think the new 6.34 client maps odd smp numbers to evens. If that is the case you will see the mapping NT from 6 to 4 appear in your SMP log and you will actually only be using 4 threads. 6.34 is so new I've not seen many folks talking about this. This would apply to the I7 crowd as well.

My i7's show "Mapping NT from 7 to 7" with 6.34 and multilpe GPU3 clients.

This current WU is showing NT from 6 to 6, I'm assuming that means it is running on 6 threads? Also thanks Uncle Fuzzy, I will run it on SMP 4 and see if the throttling is manageable.

[HendricksSA]

Fuzzy, maybe I'm getting old. I thought I read that somewhere. My bag. Wait ... it's a prime thing. See: viewtopic.php?f=55&t=17695&start=15#p176116

EDIT - rewrote above entry to reflect latest, fuzzy input.

[uncle fuzzy]

It could be true with AMD, I don't know. All I can speak to is my own i7s.

To clarify a bit more on my quad. I have a Q6600 running -smp (4 cores) and also feeding 2 GTS450s, and it still does 5-7K PPD, depending on the project.

I would try -smp 5. If it works, good for you. If it steps back to only using 4 cores, you'll still be getting pretty good points (6-8K+?), and have the headroom to play games or feed a GPU client with the other 2 cores.

edit- There have been problems with various higher prime values. I've been running -smp 7 since I started with i7s and the client has accepted it. I've also used -smp 3 with a C2Q and not had issues.

[gwildperson]

I don't think that we know if FahCore_a3 and FahCore_a5 map the threads in identical ways. The topic HendricksSA referenced says that whichever core used by Project: 6901 is mapping 11 to 10. My most powerful computer is a quad, so it's not an issue for me.

[uncle fuzzy]

I'm doing a 6901 with a5 using -smp 7. PantherX also mentions 11 and 23. Maybe you need to get to the higher primes before the remapping kicks in.

[orion]

#2, only if you're willing to make it a dedicated 24/7 -bigadv folder. I have one doing it now.

Also your NB Frequency is low, you can go higher. What MB do you have?

[bruce]

I'm doing a 6901 with a5 using -smp 7. PantherX also mentions 11 and 23. Maybe you need to get to the higher primes before the remapping kicks in.

Some time back, Kasson said there was a problem with "large primes" (and with my background in Math, "large" primes have upwards from a million decimal digits) but in the context of the number of SMP threads, I suppose 11 and 13 can be considered large primes.

It makes sense for the FahCore to know its own limitations, and if that means it has to adjust some of the numbers we throw at it, it's all in the interest of good science. As Donors, we need to figure out what's best for our own machines within whatever FAH lets us adjust.

This is my first time setting up and using the SMP F@H client ....

Welcome to foldingforum.org, "hungyzerglin"

This is supposed to be a forum for "HELP: I don't know where to start" type of questions and answers. Notice how quickly your "simple" questions departed from simple answers into areas where everybody is still discovering new things about FAH. It doesn't sound like you're the type to be easily intimidated, but in any case, don't let us intimidate you. We're a bit of a mixed crowd and most anything can happen.

[hungyzerglin]

#2, only if you're willing to make it a dedicated 24/7 -bigadv folder. I have one doing it now.

Also your NB Frequency is low, you can go higher. What MB do you have?

Yea the vmware/virtualbox setup just for -bigadv doesn't seem very practical for my uses. I'm using an ASUS M4A87TD Evo board. I don't think the chipset is the fastest out there, but it does support mid range overclocking. I've seen some people push their NB's to 3k with 1.37V, but that might be too much for me. You think 2.6k-2.8k would work? I've heard keeping HyperTransport at 2000 is the best value, so I've kept it there.