Just got a Project 2684

One of my -bigadv machines downloaded a new version of the A3 core, v2.21. My standard SMP2 clients are all still using v2.19. Anyone else receive a new A3 core?

2.21 has been out for a while but mainly for OS/X users, until now it has not been a forced upgrade. Maybe this will help with some of the problems you are having - let us know how you get on
 
2.21 has been out for a while but mainly for OS/X users, until now it has not been a forced upgrade. Maybe this will help with some of the problems you are having - let us know how you get on
The particular system that downloaded the new core has proved most stable running P2684s so far. I don't think it will be a good yardstick since it completed more of these WUs than the other systems in my farm. One thing is certain, PPD hasn't changed.
 
2.21 has been out for a while but mainly for OS/X users, until now it has not been a forced upgrade. Maybe this will help with some of the problems you are having - let us know how you get on
I just checked another -bigadv system with a VM and it already had the v2.21 A3 core for days. This machine was the system that required a reboot I mentioned the other day. When I restarted the VM, the 2684 WU crashed, so it appears that the new core didn't help with that particular client.
 
I just checked another -bigadv system with a VM and it already had the v2.21 A3 core for days. This machine was the system that required a reboot I mentioned the other day. When I restarted the VM, the 2684 WU crashed, so it appears that the new core didn't help with that particular client.

Damn thats not good. Looks like the only way to deal with 2684 in a VM is to suspend the VM - not that this helps the native *nix users though
 
2.21 has been out for a while but mainly for OS/X users, until now it has not been a forced upgrade. Maybe this will help with some of the problems you are having - let us know how you get on

Not for me. 2.21 has been as unstable as 2.19, and like APOLLO, my performance is unchanged. In fact my most reliable system has had them fail 3 seconds after they started.

It seems like Stanford has been making a lot of progress on the stability issues of late :rolleyes:

I wonder how many more failures I'll get with 2684s that cause me to get close to the BigAdv/SMP2 20% non-completion "lose your PPD bonus" line?

If it weren't for the Win A3 unit server issues all of my VM BigAdv folders would be on SMP2, but I felt that wasting power waiting for WUs wasn't a great idea.

But lately that's Stanford and the F@H project: Rock meet [H]ard place.
 
Not for me. 2.21 has been as unstable as 2.19, and like APOLLO, my performance is unchanged. In fact my most reliable system has had them fail 3 seconds after they started.

It seems like Stanford has been making a lot of progress on the stability issues of late :rolleyes:

I wonder how many more failures I'll get with 2684s that cause me to get close to the BigAdv/SMP2 20% non-completion "lose your PPD bonus" line?

If it weren't for the Win A3 unit server issues all of my VM BigAdv folders would be on SMP2, but I felt that wasting power waiting for WUs wasn't a great idea.

But lately that's Stanford and the F@H project: Rock meet [H]ard place.

The servers are now ***MOSTLY*** fixed so you should be ok should you want to go back to regular A3
 
Not to rub it in anyone's faces here, but I just finished my 5th A2 bigadv in a row and my machine has just grabbed yet another 2683 A2 bigadv. That would now make it 6 in a row! Can't believe my luck :eek:
 
Not to rub it in anyone's faces here, but I just finished my 5th A2 bigadv in a row and my machine has just grabbed yet another 2683 A2 bigadv. That would now make it 6 in a row! Can't believe my luck :eek:

Luck has a nasty habit of turning round and biting you right in the ASS:p
Nice that you are still getting them - I can't run them yet and all teh current issues are delaying my big rig build, i was looking at a quad opteron bundle but until A3 Bigadv is sorted thats a no-no. Looks like i'll spend the rest of the year tinkering with my GPU rig:eek:
 
Luck has a nasty habit of turning round and biting you right in the ASS:p
Nice that you are still getting them - I can't run them yet and all teh current issues are delaying my big rig build, i was looking at a quad opteron bundle but until A3 Bigadv is sorted thats a no-no. Looks like i'll spend the rest of the year tinkering with my GPU rig:eek:

Definately keep us updated on your big rig build. I'm trying to save enough money to build a monster folding rig by the end of this year, so i'm naturally curious about what other people are doing. I am hoping to have enough funds to do a dual 12 core Magny Cours build by then, although that is tentative as lots changes in the realm of folding.
 
So, my VMs are still up and running. I got a 2681 this morning. Seems to be an A2 Big Adv as I am getting near 2683 PpD. I don't remember these. My other VM is running an A3 2684 and should finish in 1.75 days.
 
There is a new Linux A3 core released, v2.22. It supposedly increases performance. To get it, delete the old A3 core and the new version will automatically download. I'm curious if this will reduce the P2684 TPF closer to the A2 frame times. Do not delete the core before completing a WU. Only do this after a WU completes to avoid the possibility of a crash.
 
Thanks for the heads up, I went ahead and deleted the a3 cores from my VM's and the one that just grabbed a 2684 DL the new core.

Granted the WU just started but I'm seeing the same TPF on this core as the 2.19 core. :( That may change of course once it goes through a few frames; if it does, I'll update accordingly.

I will say that I'm not even going to play with these things anymore even with a new core possibly having more stability. I'm still going to suspend the VM (and try nothing else) because that so far has been the only way I've kept 2684's from flushing.
 
I will say that I'm not even going to play with these things anymore even with a new core possibly having more stability. I'm still going to suspend the VM (and try nothing else) because that so far has been the only way I've kept 2684's from flushing.
I believe the new core addresses performance only, but I could be wrong. I only found out about the new core shortly before I posted and have very little info on it so far.
 
:eek: What specs??

Thanks for the heads up. :cool:

Well i was looking at getting the tyan 4985 board so i could upgrade the CPU's later to istanbul's but they are stupid prices, Much more doable in the short term is quad 8356's ram and board for £600 + shipping but PPD on A3 for them isn't that hot. A more detailed research session the other night had me looking at a supermicro board that will take any 8xxx operton but is still reasonable @ £350 + shipping, i could get that and the ram from the start with cheap 835x's and upgrade them later to something with more oomph. Of course with bulldozer hopefully early next year i could wait, or there was a dual 2439SE board, cpu's and ram new for £800 on ebay yesterday. Hmmmm Decisions Decisions:confused:
 
There is a new Linux A3 core released, v2.22. It supposedly increases performance. To get it, delete the old A3 core and the new version will automatically download. I'm curious if this will reduce the P2684 TPF closer to the A2 frame times. Do not delete the core before completing a WU. Only do this after a WU completes to avoid the possibility of a crash.

OK ToTow over at F@H addict is saying that v2.22 should give a 15% speed boost over earlier versions of the core and also contains other improvements, given everyone's luck so far it has to be worth a shot

http://en.fah-addict.net/news/news-0-250+the-smp2-a3-core-gets-a-speed-boost-on-linux.php
 
Well i was looking at getting the tyan 4985 board so i could upgrade the CPU's later to istanbul's but they are stupid prices, Much more doable in the short term is quad 8356's ram and board for £600 + shipping but PPD on A3 for them isn't that hot. A more detailed research session the other night had me looking at a supermicro board that will take any 8xxx operton but is still reasonable @ £350 + shipping, i could get that and the ram from the start with cheap 835x's and upgrade them later to something with more oomph. Of course with bulldozer hopefully early next year i could wait, or there was a dual 2439SE board, cpu's and ram new for £800 on ebay yesterday. Hmmmm Decisions Decisions:confused:
Any of these options would be an awesome build. There were a couple of members here who had purchased parts for a quad AMD machine, but one obstacle after another led to cessation of the project. I'm skeptical about the current state of affairs with -bigadv and multi-socket systems. People who have access to very high-end servers demonstrated inefficient scaling regardless of architecture. These systems require multiple -bigadv clients for full performance realization, which means tons of memory, and no OCing decreases potential. There remains a significant amount of uncertainty whether such projects are cost effective.

If you do take the plunge, make sure you have parts that are fully supported by the mainboard manufacturer, and that you follow their instructions to the letter. These systems are more complicated and far more finicky than any dual socket build. I had assembled a 6-socket PPro system 9 years ago for use in G@H, and it took me ages to get all the parts that would function well together. It was trial and error all the way though and spent more than I needed to if I had selected the right parts from the onset.
 
Any of these options would be an awesome build. There were a couple of members here who had purchased parts for a quad AMD machine, but one obstacle after another led to cessation of the project. I'm skeptical about the current state of affairs with -bigadv and multi-socket systems. People who have access to very high-end servers demonstrated inefficient scaling regardless of architecture. These systems require multiple -bigadv clients for full performance realization, which means tons of memory, and no OCing decreases potential. There remains a significant amount of uncertainty whether such projects are cost effective.

If you do take the plunge, make sure you have parts that are fully supported by the mainboard manufacturer, and that you follow their instructions to the letter. These systems are more complicated and far more finicky than any dual socket build. I had assembled a 6-socket PPro system 9 years ago for use in G@H, and it took me ages to get all the parts that would function well together. It was trial and error all the way though and spent more than I needed to if I had selected the right parts from the onset.

Thats the big doubt that keeps stopping me, i've seen the posts that the guys on here were making although there was one guy who managed it without too much trouble on an old dell board. Its not something thats going to be rushed as this will have to last a long time. If the plunge is taken its got to have at least a 3year lifespan otherise the wife will not be happy.

From my research so far cranky bioses, dodgy memory and cheap PSU's seem to be the major issues.

OP sorry for the OT posts, if htis does go ahead i'll start a new thread....
 
Well i was looking at getting the tyan 4985 board so i could upgrade the CPU's later to istanbul's but they are stupid prices, Much more doable in the short term is quad 8356's ram and board for £600 + shipping but PPD on A3 for them isn't that hot. A more detailed research session the other night had me looking at a supermicro board that will take any 8xxx operton but is still reasonable @ £350 + shipping, i could get that and the ram from the start with cheap 835x's and upgrade them later to something with more oomph. Of course with bulldozer hopefully early next year i could wait, or there was a dual 2439SE board, cpu's and ram new for £800 on ebay yesterday. Hmmmm Decisions Decisions:confused:

What about doing a quad socket G34 board(LGA 1944)? There are rumors that the Opteron version of Bulldozer may use this socket. If that is the case, you could go with quad Operteron 6172/74/76's right now and have the potential to go to Bulldozer later. Of course there is no garentee that this is true, or whether the motherboard manufacturer would even offer an updated bios if that was possible.

BTW guys, I think i'm going to Ctrl-C my current A2 bigadv so I can delete my current version 2.19 FahCore_A3. In linux, what would be the command that you type to delete something at the command prompt? Is it "del FaHCore_A3.exe" just like in DOS?
 
wait till its done
it will trash your WU

Arghhh..........

Too late. I ctrl-C's to shut the Linux client down in order to delete my A3 core. I find that the command to use was rm FaHCore_A3.exe in Linux.

Upon restarting, it crashed and flushed my precious A2 work unit. However, it gave me yet another A2 work unit which means that I have now recieved 7 A2's in a row.

I have always used ctrl-C to shut the Linux client down in the VM, and this is the first time that it has crashed and flushed an A2 bigadv WU using that to shut the client down.
 
Arghhh..........

Too late. I ctrl-C's to shut the Linux client down in order to delete my A3 core. I find that the command to use was rm FaHCore_A3.exe in Linux.

Upon restarting, it crashed and flushed my precious A2 work unit. However, it gave me yet another A2 work unit which means that I have now recieved 7 A2's in a row.

I have always used ctrl-C to shut the Linux client down in the VM, and this is the first time that it has crashed and flushed an A2 bigadv WU using that to shut the client down.

Unless I am mistaken, it didn't flush because you used ctrl-c, it flushed because you deleted the core before the WU finished. A better method would be to ctrl-c to stop it, then re-run with the -oneunit flag. After it finished that unit, then you delete the old core and then restart the client (w/o the -oneunit flag).
 
Unless I am mistaken, it didn't flush because you used ctrl-c, it flushed because you deleted the core before the WU finished. A better method would be to ctrl-c to stop it, then re-run with the -oneunit flag. After it finished that unit, then you delete the old core and then restart the client (w/o the -oneunit flag).

It was working on a 2683 A2 work unit using the FaHCore_A2.exe core. When I stopped the client using ctrl-C, I deleted the FaHCore_A3.exe file which is the A3 core. When I started the client back up, it flushed and downloaded another A2 bigadv.......
 
It was working on a 2683 A2 work unit using the FaHCore_A2.exe core. When I stopped the client using ctrl-C, I deleted the FaHCore_A3.exe file which is the A3 core. When I started the client back up, it flushed and downloaded another A2 bigadv.......

Why stop it at all (I appologize if this was already covered in an earlier post.) I deleted all of my A3 cores for machines running A2's yesterday without issue.
 
Why stop it at all (I appologize if this was already covered in an earlier post.) I deleted all of my A3 cores for machines running A2's yesterday without issue.

I didn't realize that you could delete the A3 cores without stopping the client. For future reference, how would you be able to do something like that? This sucks as it flushed one of those precious A2's..........
 
amdgamer,

The A2 -bigadvs have proved for me to be fairly stable, not as much as standard A1, A2 and A3 WUs but far more stable than the P2468. I have used ctrl-C to end -bigadv WUs many times but you need to wait until the client releases all threads, otherwise you run a 50/50 or higher chance of trashing the WU. There is very little likelihood of a crash if you wait for the client.

Today I did the same as you, I stopped a P2684 to download the new A3 core and the WU crashed. I knew there was a high probability of a crash. However, it had just commenced and was only a few minutes into the work. Therefore, not a significant loss. Right after, the client downloaded a P2683. I'll see how well this new A3 core performs next time I get another 2684.
 
I didn't realize that you could delete the A3 cores without stopping the client. For future reference, how would you be able to do something like that? This sucks as it flushed one of those precious A2's..........
Yes, you can delete the cores while a WU is being processed because they reside in memory, but I wouldn't recommend it. Better to just wait between WUs or at 0% like I did earlier.
 
Why stop it at all (I appologize if this was already covered in an earlier post.) I deleted all of my A3 cores for machines running A2's yesterday without issue.
To clarify my last post above, YES, this is actually OK. As long as the type of WU the client is working on is not using a core you delete, then there is no risk. So, deleting an A3 core while processing an A2 WU is fine since the client is working with the A2 core, not the A3 core. Vice versa should be OK as well.
 
Yeah, this is actually the first time an A2 bigadv work unit has crashed and flushed with me using ctrl-c to shut the client down inside the VM. I have used ctrl-c all the time without a single problem, which is why I was shocked. I thought I waited for the entire client to shut down properly though, as I made sure to wait until it displayed all of that text after you give the kill command.

Either way, I guess stopping a bigadv is a risky proposition no matter what as I can see why some choose to just suspend their VM's instead.

If Stanford would bring bigadv to Windows soon, I would be so happy........

It sucked because the A2 it was crunching was at 59%, which equals 1.5 days worth of crunching since my rig does 40% every 24 hours with A2's.

amdgamer,

The A2 -bigadvs have proved for me to be fairly stable, not as much as standard A1, A2 and A3 WUs but far more stable than the P2468. I have used ctrl-C to end -bigadv WUs many times but you need to wait until the client releases all threads, otherwise you run a 50/50 or higher chance of trashing the WU. There is very little likelihood of a crash if you wait for the client.

Today I did the same as you, I stopped a P2684 to download the new A3 core and the WU crashed. I knew there was a high probability of a crash. However, it had just commenced and was only a few minutes into the work. Therefore, not a significant loss. Right after, the client downloaded a P2683. I'll see how well this new A3 core performs next time I get another 2684.
 
I haven't seen a 2681 or 2683 in a couple of days and I lost a 2684 last night due to a bad GPU2 WU. The GPU2 WU caused the rig to crash and the 2684 (in Linux VM) was lost after the restart :(

They are currently chugging along (again), but at the next crash of either i7 rig, it is back to SMP2 for me.

On a good/bad note, my wife's AMD rig (former GPU2 folding rig) is having problems and I may need to replace in the next month or two when I get paid. I'm thinking an i7 930 might be the appropriate course of action here. Won't be able to run Big Adv as it will not be running when she is using it so it'll be running SMP2 only.
 
I haven't seen a 2681 or 2683 in a couple of days and I lost a 2684 last night due to a bad GPU2 WU. The GPU2 WU caused the rig to crash and the 2684 (in Linux VM) was lost after the restart :(

They are currently chugging along (again), but at the next crash of either i7 rig, it is back to SMP2 for me.

On a good/bad note, my wife's AMD rig (former GPU2 folding rig) is having problems and I may need to replace in the next month or two when I get paid. I'm thinking an i7 930 might be the appropriate course of action here. Won't be able to run Big Adv as it will not be running when she is using it so it'll be running SMP2 only.

If that is the case, why not get her a Thuban? They overclock very well and generate points very similar to a Bloomfield. Since you won't be running bigadv on it, you'll be fine and you will even have 6 real cores.

On a side note, looks like my lucky streak is over after getting 7 A2 bigadv work units in a row as it just drew a 2684 A3 bigadv this morning........
 
I gotta say, I really hate the MSI K9A2 for PII chips. There is STILL no way to change voltage on this MoBo. It'll probably re-purpose the PII 940BE.

I just want an excuse to get my SR-2 sooner.
 
Excuse my anger, but good-frickin-bye to bigadv as i'm going back to Windows SMP. I had a string of 7 straight A2's and was happy. It pulled an A3 2684 this morning and guess what, it crashed pretty early in the work unit. I got a corestate=0(0) and it terminated.

The points were nice while A2 lasted, but A3 isn't worth the baby sitting as I am a set and forget type of person. Since my rig is running at 3.46 ghz, trying to downclock it even more is not going to work as I was already running almost up on the 4 day deadline. Maybe i'll try it again if bigadv ever comes to native Windows. I had to downclock my rig to make it bigadv stable, so now i'm considering taking my original overclocks back up to 3.67ghz which is what it was stable at for 10 straight A3 SMP2's.
 
The points were nice while A2 lasted, but A3 isn't worth the baby sitting as I am a set and forget type of person. Since my rig is running at 3.46 ghz, trying to downclock it even more is not going to work as I was already running almost up on the 4 day deadline. Maybe i'll try it again if bigadv ever comes to native Windows. I had to downclock my rig to make it bigadv stable, so now i'm considering taking my original overclocks back up to 3.67ghz which is what it was stable at for 10 straight A3 SMP2's.
I totally understand and agree with your decision. These WUs should still be in beta, don't know how they got past the beta testing phase. Just to let you know, frequency is not a problem. I've had 2684 crashes at stock frequencies as well, so something else is up besides a little extra instability compared to A2.
 
I totally understand and agree with your decision. These WUs should still be in beta, don't know how they got past the beta testing phase. Just to let you know, frequency is not a problem. I've had 2684 crashes at stock frequencies as well, so something else is up besides a little extra instability compared to A2.

Just my luck........I put my folding rig back on SMP for the first time since when I first got it, and it pulls a 6701 which i've heard people talking about on this forum. So far, TPF's at 3.46ghz on the 6701 A3 are 7:55.

The other reason why I have gone back to SMP is that i've only folded a total of 76 work units. With all of these failures, I believe I am knocking right at the 80% requirement. I had a coupled of failed A2's when I first started bigadv, and had a set of failed 2684 A3's in a row. The last thing I want is to be folding a work unit and not get bonus compensation for it.

It feels kind of good to be back on SMP stuff as I know that I can shut the client down at anytime without problems or risk of a flushing. If I hear that 2684's have gotten more stable, i'll consider coming back but my SMP client is running with the -bigadv flag just in case they stealth release windows bigadv.
 
The other reason why I have gone back to SMP is that i've only folded a total of 76 work units. With all of these failures, I believe I am knocking right at the 80% requirement. I had a coupled of failed A2's when I first started bigadv, and had a set of failed 2684 A3's in a row. The last thing I want is to be folding a work unit and not get bonus compensation for it..
Yeah, although for me the number of 2684 WUs seems to have lessened in the past week. Also, another thing I noticed is that I don't get the EUE strings anymore, where several WUs would crash in a row because Stanford was reassigning the same unstable WU over and over. I think Stanford stopped doing this, likely because of what you're concerned about. Now, if a WU crashes, the client might attempt another download and if that crashes as well, I get something completely different like an A2. I hope what I'm seeing is not just the luck of the draw.
 
Last edited:
Excuse my anger, but good-frickin-bye to bigadv as i'm going back to Windows SMP.

The points were nice while A2 lasted, but A3 isn't worth the baby sitting as I am a set and forget type of person.

QFTMFTx2.

My 2nd VM grabbed it's 50th A3 in a row (exaggerating but it was a lot) yesterday, I shut it down and went SMP2.

My primary VM went through probably 4 A2's then grabbed an A3, that has now also been shut down and SMP2 has been switched on.

I hate to sound like an a-hole but PG can shove these raw A3's, they are not fit for folding consumption. I'd rather have a total PPD drop going from A2 > SMP2 than fart around with these stupid damn things and have flushes every few days and not only lose the base pts but also the electricity and the time spent crunching them before they flushed. :mad: I know I'm not a long timer or lifer but PG's handling of this situation has really bothered me. I know there's been better and worse in the past or whatever, but this here is just some garbage.
 
QFTMFTx2.

My 2nd VM grabbed it's 50th A3 in a row (exaggerating but it was a lot) yesterday, I shut it down and went SMP2.

My primary VM went through probably 4 A2's then grabbed an A3, that has now also been shut down and SMP2 has been switched on.

I hate to sound like an a-hole but PG can shove these raw A3's, they are not fit for folding consumption. I'd rather have a total PPD drop going from A2 > SMP2 than fart around with these stupid damn things and have flushes every few days and not only lose the base pts but also the electricity and the time spent crunching them before they flushed. :mad: I know I'm not a long timer or lifer but PG's handling of this situation has really bothered me. I know there's been better and worse in the past or whatever, but this here is just some garbage.

Having crunched a few SMP A3 stuff since yesterday, I am actually not as upset over the loss of points. I am on my 4th WU since going back to Windows SMP2, and I am finding that my system at 3.46ghz is averaging 14.5k ppd so long as I don't get any more 6701's. when doing a 6701, my ppd average was only around 11k. By comparison, I plugged what TPF's into Linuxforge's calculator and found that doing nothing but Linux A3 2684 bigadv was yielding an average of 16k ppd as I was running right up on the 4 day deadline.
 
Are yall losing these on Linux VMs, or Linux Native?

Ive only grabbed one on Linux and it was in a VM and it was fine.
 
Last edited:
Back
Top