• 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.

Announcement: SMP2 Cores

T

Tobit

[H]ard|DCer of the Month - March 2010/May 2011
Joined
Dec 12, 2005
Messages
5,145
Written by Dr. Kasson:

After a long development process, we are excited to announce the upcoming release of SMP2 (threads-based SMP) cores to public testing. The first SMP2-based core is the A3 core, and it will soon become available on advanced methods for OS/X (Intel), 64-bit Linux, and Windows. We are still doing development work to refine the A3 core, but it is at the point where we are ready for public testing.

We are excited about the SMP2 cores because the threads-based parallelization allows us to dispense with the MPI-based parallelization that added an extra layer of complexity and was particularly troublesome for Windows users. We anticipate phasing out the earlier SMP cores and work units in favor of these new ones; at this point in the changeover process, our Windows SMP client will still require MPI to be installed so that the client can handle an A1 work unit if no A3 work units are available. In the near future we will release an updated Windows client that does not require MPI.

The SMP2 cores require a client update; please upgrade your SMP-capable clients to at least version 6.29. We will gradually discontinue SMP projects for earlier clients.

Important: the SMP2 cores use the early-completion bonus system that we piloted with the bigadv work units. We have revamped the benchmarking system to work with this bonus system. The base point values for SMP2 work units will appear low; the benchmarked points values **include bonuses.** Some third-party utilities have been updated to include these bonuses in their calculations.

Please see an accompanying post regarding the bonus system.

One important part of the bonus system is that users:
1. Must use a passkey to receive bonus points
2. Must successfully return >=10 A2 or SMP2 work units with their passkey to receive bonus points
3. Must successfully return >80% of A2 or SMP2 work units to receive bonus points

We will shortly perform a limited "reset" of the bonus-qualifying work unit history. Important: users who have qualified for a bonus will remain bonus-qualified. We will also maintain the % returned for users but will reduce the overall counts to 10. As we do not have an automated "timeout" for bonus qualification history, we may perform rare periodic such resets in the future.

Thanks! We are excited to release these new cores to the public.
Click to expand...
 
EvilAlchemist and I have done a lot of beta testing with A3 cores on both Linux and Windows. Let us know if you have any questions or need help once these start rolling out. :cool:
 
I'd like to know by which time I can safely switch to the SMP2-capable Windows client without risking having to crunch WUs using the A1/2 cores. Is there any estimate for when this may be? I could switch my X2 system to the Windows client first and see when it is using the A3 core, I guess.
 
That's a good question Elledan. I would probably give it another day or two. When you are ready to swtich, make sure to set -advmethods in the config as this will be the only way to get A3's for now.
 
Tobit or EA can you ellaborate on this "Reset". I was left confused after reading it.

How will it affect us?
 
DooKey: nothing at this time, -bigadv still requires Linux or OS/X for now as it is A2 based and the A2 core still only runs on Linux or Mac OS/X clients. However, as stated in the new SMP client thread I posted recently, you still want to upgrade the client as, within the next couple weeks, -bigadv will no longer work on the older clients. Stanford will be requiring the new client for -bigadv work as there were bugs found and corrected in 6.29 which was just released.

Mr. P: Clients have never downloaded automatically, only cores. I announced the new SMP clients here: http://hardforum.com/showthread.php?t=1488423 -- hopefully notfred will update his stuff with them soon. His FredNotFred -bigadv image downloads the latest clients automatically, not sure of his other image. I've never used NotFred.

nitro: let me try and interpret that myself. ;)
 
Last edited:
nitro: I am not sure why they are doing the bonus reset. However, it does not sound like it will impact those of us who are already eligible. I'm not really sure at this time. Let me ask Dr. Kasson for more clarification.
 
404: All of the A3 WUs I've tested here on Windows used ~128MB of memory typically. I'll be posting some benchmarks shortly.
 
Last edited:
Ok, will it be better to run A3 units under windows with a windows client or A3 units in a Linux VM? I've upgraded my EVGA VM to the 1.0 jfs which has the 6.29 client. Am I still setup for max ppd?
 
DooKey said:
Ok, will it be better to run A3 units under windows with a windows client or A3 units in a Linux VM?
Click to expand...
If all you want to crunch are A3 WUs, I see no reason to run a Linux VM. In testing, many of us found the performance of A3's to be the same on either Windows or Linux native boxes. However, until A1's are completely phased out, you will run the risk of getting these WUs in the off chance there is no A3 work available. Keeping the Linux VM for now, would mean you would also be able to keep crunching A2 work if this were to happen. A1's will definitely be phased out. However, there is no time frame for this right now.
 
Guess I didn't ask the right question. For ppd are the -bigadv work units better than the A3 work units?
 
DooKey said:
Cool, how bout for science? A3 or -bigadv?
Click to expand...
I've always been of the belief that science benefits the most when one can run as many different clients/cores as possible. Each project focuses on different research goals.
 
I understand. Thanks for the updates!

I have some laptops running A2 cores, gpu's doing what they do, and 2x i7 doing -bigadv. I'd say that's a good variety.
 
Tobit said:
I've always been of the belief that science benefits the most when one can run as many different clients/cores as possible. Each project focuses on different research goals.
Click to expand...
Agreed. I have run every client except for the PS3 because I never owned one.

HighYield said:
Are the A3 actually in the wild yet? I would love to get rid of the VMs.
Click to expand...
I am much more reluctant. I have experienced more troubles with the Win SMP client than all the others combined. I realize that the new SMP2 client is intended to improve things, but I'll wait a good while before uninstalling VMware. I want to see a lot of feedback first.
 
nitrobass24 said:
Tobit or EA can you ellaborate on this "Reset". I was left confused after reading it.
Click to expand...

Reset = To get the bonus, you have to have completed 80% of the assigned A1 / A2 WU assigned to you under your passkey.

Since SMP1 has been going so long, some users who may be unable to get to that 80% statically due to running & failing so many WU over the last year.

So he have reset that counter. Fresh Start. Complete 10 WU's and you get the bonus.
 
Benchmarks for the SMP2 A3 Cores - Windows Native

All Numbers Include Bonus Points

C2Q - X3350 @ 2.66 Ghz = 6200 - 6500 PPD
C2Q - Q6600 @ 2.67 Ghz = 6000 - 6200 PPD
C2D - 3075 @ 2.67 Ghz = 2000 - 2200 PPD
Core i7 - 2x X5550 @ 2.67 Ghz = 19,000 - 21,000 PPD
Core i7 - 2x E5530 @ 2.40 Ghz = 16,000 - 18,000 PPD

******Notice*********
Core i7 HT has NO impact on folding times.
Meaning, there is not PPD differeence between
  • SMP -4 with HT off
  • SMP -8 with HT on

K Factor:
Your bouns is determined by how FAST you return a Work Unit.

So if you restart a system during a work unit, then your ppd will be lower on the unit because it took longer to complete so don't let it alarm you.

Also, make sure you 3rd party monitoring application can calculate K Factor.
FAHmon currently does not. HFM.net and FAHspy do as of last public versions.

VMs -
It makes no sense to run SMP2 in a virtural machine. It will lower you PPD.
Both the A3 Linux & A3 Windows cores are 99% the same.
Both get the same speed and PPD.
 
As promised, I had asked Dr. Kasson to elaborate on the bonus "reset". His reply is as follows:

It is being done in part because of some errors in our stats system that counted too many units as bonus-qualifying. Also, in an ideal case the >=10 qualifying threshold would be permanent but the success rate for the 80% threshold would be calculated over a moderate timescale (e.g. 2 months). That isn't feasible with our current system, so this operation has the same effect.

The operation will be along the following lines, if you take bonusExpired as the number of bonus-qualifying work units that expired without being completed and bonusFinished as the number of bonus-qualifying work units that were completed.

for each user/passkey:
fractionExpired = bonusExpired / (bonusExpired + bonusFinished)
if bonusExpired > 0, bonusExpired = round(10.0*fractionExpired)
if bonusFinished > 10, bonusFinished = 10

You will note that this preserves the success rate bonusFinished/(bonusFinished+bonusExpired)
Click to expand...
 
Here are some benchmarks of various A3 projects running on Windows 7, 64-bit, with my i7 920 @ 3.8 GHz. My machine isn't dedicated and is also my daily driver. Numbers shown are with -smp 8 running. There was also a single GTX260 folding simultaneously. Bonus points shown were calculated by HFM.net.

Code: 
 Project ID: 6012
 Core: GRO-A3
 Credit: 470
 Frames: 100

 Name: ganymede - SMP2
 Path: C:\fah\smp2\
 Number of Frames Observed: 1200

 Min. Time / Frame : 00:02:53 - 18,171.5 PPD
 Avg. Time / Frame : 00:03:00 - 17,121.8 PPD

Code: 
 Project ID: 6014
 Core: GRO-A3
 Credit: 484
 Frames: 100

 Name: ganymede - SMP2
 Path: C:\fah\smp2\
 Number of Frames Observed: 1200

 Min. Time / Frame : 00:02:57 - 18,528.5 PPD
 Avg. Time / Frame : 00:03:05 - 17,252.8 PPD
 
Noob alert:

So I need to start running fah6 with the -smp and -advmethods flags to get these units?

 
OvrrDrive said:
So I need to start running fah6 with the -smp and -advmethods flags to get these units?
Click to expand...
They haven't quite been released yet to public testing.. give it a few more days just in case. You need to be running the latest SMP client, v6.29, and use the -advmethods flag. It should autodetect the number of processors you have but it would be safe to force it with the -smp x flag where x is the number of cores.
 
Gotcha. I already upgraded to the 6.29 client so I'll just add the -advmethods flag when my current wu completes so I won't forget. Thanks.
 
So wait, since there is no difference in -4 and -8 performance, does that mean that we can run two -4's on a single i7?
 
PuppeteerNessus said:
So wait, since there is no difference in -4 and -8 performance, does that mean that we can run two -4's on a single i7?
Click to expand...

Well, technicall you can. Here is what will happen. You PPD will be cut in 1/2 minus the additional the lost bonus points.

So instead of getting 9,000 PPD running -smp 8 you would get about 3000 PPD x 2 smp -4.

Hyper-Threading offer no PPD advantage.

Meaning you get the same poinst per day (same frame times) if you run with smp -8 (4 pysical + 4 HT) or SMP -4 (4 physical).

The Bonus factor is what makes it better to run a single client on each machine.
 
So what about if you run smp -4 on a hyperthreaded i7? Do you get 4 'free' cores for something else? Or is there a performance cut as well?
 
Mr. Pedantic said:
So what about if you run smp -4 on a hyperthreaded i7? Do you get 4 'free' cores for something else?
Click to expand...
Sure. You can run as many cores as you want. Just don't expect the same amount of PpD going from -smp 8 to -smp 4. :cool:

BTW, I should mention, that the system remains very responsive when running -smp 8. I don't have to use any third party affinity/priority manager here.
 
Mr. Pedantic said:
So what about if you run smp -4 on a hyperthreaded i7? Do you get 4 'free' cores for something else? Or is there a performance cut as well?
Click to expand...

If you are running a Core i7 with Hyper-Threading Enabled - here are your options.

A) -smp 8
B) -smp 4 (and locking affinity to the physical cores)

You will get the same PPD in each case.

If you were to run a 2nd client, on just the Hyper-Thead Cores, you first client PPD would drop down by 2/3 and the second client would do about the same.

It is best to even forget running two clients is even an option cause it will murder you PPD.
 
Oh. Pity.

So, next question, logically, would be: which requires more power? Which is hotter?
 
Mr. Pedantic said:
So what about if you run smp -4 on a hyperthreaded i7? Do you get 4 'free' cores for something else? Or is there a performance cut as well?
Click to expand...
Remember that because of the way that HT works, 4 threads can still use 100% of the CPU. So the actual load that F@H places on the CPU will be the same with HT enabled or disabled if you use -smp 4.
Mr. Pedantic said:
So, next question, logically, would be: which requires more power? Which is hotter?
Click to expand...
If they perform the same, then they should use about the same amount of power as well. The reason that i7s with HT enabled tend to consume more power than with it disabled is because when stress-testing, the CPU can be loaded more with 8 threads than it would be with 4. However, if the load is the same, then the power consumption should be as well.
 
I found that on my dedicated F@H Dual CPU i7 Xeons , i am saving about 40-50 watts by turning HT off. Also dropped the temps by about 10c.
(System 2 x E5530 @ 2.4Ghz = HT on (smp -16) 283 watts - HT off (smp -8) to 234 watts)

As we know, a dual cpu system has a few more compents then a single cpu system, so my numbers will not scale down perfectly to a single cpu system.

On a single Core i7 system, you would probley only save 10-20 watts turning HT off.
(I don't have one of these CPUs so maybe Tobit can answer this part better)
(Maybe Kyle did some test on the HardOCP benchmarks - check out the article to see)

If it is your main desktop system, leaving HT on and locking F@H to the physical cores may make the system seem more responsive.
(That is just my opinion, i have no data to suggest this would acutally be better on the system)
 
Last edited:
My secondary PC is running a P6012 right now. I will post performance numbers once it's folded a few frames.
 
Does this error have anything to do with the upgrade?

"Downloading new core: FahCore_a3.exe
Error: HTTP GET returned error code 404
Error: Could not download core
Core download errir <#8>, waiting for retry..."

After I finished the last wu I restarted the client witht he -smp and -advmethods flags and got this error that keepe retrying but never succeeds. I tried starting with just the -smp flag and with no flags at all and still get the same error.

Anyone have an idea to get this box folding again before I go to bed?
 
You must log in or register to reply here.
Back
Top