A2 SMP core

[Kendrak]

What are some of the results that you are getting?

I have two headless boxen that run notfreds.

Each one run 2x instances. Each one has one A1 and one A2 core right now. I looked at my FAHMon and I'm pushing 47k. That is 4-5k higher than I have ever done, and I did not change a thing.

Yay A2 core!

 

[AgrFan]

I have two dual cores running notfreds.

They produce 3K PPD when each one has a A2 core.
One has a A2 (2669) and one has a A1 (2665) right now.

2171 PPD

 

[SmokeRngs]

At 3.6Ghz, I can get over 3k PPD per VM instance. However, that's only true if I had faster internet upstream since the work units are close to 30 meg uploads and if the client would stop freezing after uploading the work unit half the time.

 

[Kendrak]

At 3.6Ghz, I can get over 3k PPD per VM instance. However, that's only true if I had faster internet upstream since the work units are close to 30 meg uploads and if the client would stop freezing after uploading the work unit half the time.

30 megs...... the explains the seizure my connection had the other day

 

[flogge]

My laptop has yet to see anything but a 2665, which seem to be running a bit faster now but they still take 2.5-3 days to complete. I'd love to see one of these alleged A2 cores sometime. The 8800m GTS is blowing it out of the water.

Also does anyone know what they are sending to my old P4? I have hyperthreading turned on and it is running 2 instances of project 784 or 782. Something called mini chaperonin and it takes 8 days for the tired old thing to finish 2 of them at 2.88 Ghz. Should I turn off hyper-threading?

 

[Xilikon]

30 megs...... the explains the seizure my connection had the other day

Some even go as far as 50 MB

However, Peter Kasson said that he found out some of the space is just useless data so he tweaked the workunits to not record these. Hopefully, it will shrink to 30 Mb or less.

 

[HighYield]

At 3.6Ghz, I can get over 3k PPD per VM instance. However, that's only true if I had faster internet upstream since the work units are close to 30 meg uploads and if the client would stop freezing after uploading the work unit half the time.

Wow!...I got to figure out how to set up VMs on my 64 bit vista boxen. Is there any VM software that has signed drivers yet?

 

[Xilikon]

Wow!...I got to figure out how to set up VMs on my 64 bit vista boxen. Is there any VM software that has signed drivers yet?

Nope but just hit F8 at boot then you are ok. A [H]ard folder doesn't know how to shut down

 

[HighYield]

Nope but just hit F8 at boot then you are ok. A [H]ard folder doesn't know how to shut down

Ok, maybe I'm a little soft ..is there an up to date guide for setting up VMs on windowz (vista)? Seems like this is the kind of info that might do well in the [H]orde Guides sticky.

 

[rmdashrrootsplat]

Some even go as far as 50 MB

However, Peter Kasson said that he found out some of the space is just useless data so he tweaked the workunits to not record these. Hopefully, it will shrink to 30 Mb or less.

LOL! Just as I read this one of my x64 CentOS VM's started uploading a 50MB monster. Need a fatter pipe.

 

[Kendrak]

LOL! Just as I read this one of my x64 CentOS VM's started uploading a 50MB monster. Need a fatter pipe.

I'm on DSL with a 3mb up

Cable is here but I don't want to switch. Their HD TV is crappy and costly. (I have dish) If I get it though I could kill our house line.

hummm...... to offset the extra cost of just getting cable internet.

Edit: just moved up to 768kb/6mb for 5 bucks more a month, that should help My upload just went up by 50% and my download by 2x.

 

[Xilikon]

Ok, maybe I'm a little soft ..is there an up to date guide for setting up VMs on windowz (vista)? Seems like this is the kind of info that might do well in the [H]orde Guides sticky.

Nope but I'm planning to do a good guide to setup VMWare and install a OS for folding.

 

[SmokeRngs]

Wow!...I got to figure out how to set up VMs on my 64 bit vista boxen. Is there any VM software that has signed drivers yet?

As stated, nothing that is known as of yet. However, Xilikon is correct about hitting F8 during bootup to kill the signed driver checking.

Nope but just hit F8 at boot then you are ok. A [H]ard folder doesn't know how to shut down

I only shut down my machines due the the annoying Windows updates. After about 2-3 days of the fucking bubble popping up and telling me I need to reboot I finally do it. Otherwise, the only time my systems reboot is due to a power loss.

 

[jws2346]

Ok, here's the bottom line on my A2 clients ppd All the following readings are by FAHmon 2.3.2b. Both boxen are running "native" Ubuntu 64 v8.04 with slight OC's and the LinSMP client under Ubuntu 64 v8.04 folding 24/7. I sometimes check my email or "surf the Net" with both boxen. I've never got a A2 WU on my WinSMP boxen.

I like to read about 5x different newspapers online (ie," New York Times, LA Times, Chicago Tribune, Houston Chronicle, Seattle Post Intelligencer, etc), so that's pretty much my use of either boxen. I never touch my WinSMP boxen because it's running CPU2 under WinSMP w/SP3 and IMO, the Linux GPU2 port is just too complicated for a non tech person like me

The first boxen I'm posting about is a E6600 mildly OC'ed to 3.0 GHz (3x333).. According to FAHmon 2.3.2b the E6600 (dual core) with a 2669 (R2-C70-G1) WU is getting 2703 ppd with AVG= 11mns 9s, CUR= 10mn 10s and MIN= 9mn 55s Time/Frame. Again according to FAHmon 2.3.2b with a 2669 WU the other boxen I have folding 24/7 is a Q6600 (quad core), mildly OC'ed to 3.0 GHz (9x333) and running "native Ubuntu 64 v8.04 and a 2669 (R6-C58-G2) Wu and according to FAHmon 2.3.2b I'm getting AVG= 5mn 14s, CUR= 5mn 14s and MIN=5mn 14s (pretty close to 5mn and 14 seconds all the way around )

I hope this helps or at least informs a [H]orde team 33 folder

Edit: Oops, for us math challenged that's 528x something ppd for the Q6600. Below is a translation of what I was trying to say above,

Ubuntu#3 = Q6600
Ubuntu#5 = E6600
WinGPU#1 = EGA 8800gs GPU2
WinCPU#2 = Q6600 WinSMP



Edit again: Yeah, I can count, at least on my fingers, and I'm holding the #4 folding spot open for when I finally get my PNY 8800gt w/512 MB in the freaking "slow boat to China" delivery system. (it's funny, they sure ain't slow about charging your credit card )


FOLD ON!

 

[HighYield]

What is the current flavor of VMware server that everyone recommends for running with Vista. 2.0 rc2 or the 1.0x version so that we can get some of that A2 core goodness?

Is Tigerbiten's thread still the best source on how to set one of these beasts up?
http://hardforum.com/showthread.php?t=1265008
Seems like I have been leaving too many points on the table running the Windoz SMP clients (but need vista for all the gpu clients).

 

[Xilikon]

I like VMWare 1.0.6 myself, no fuss and well proven software

 

[SmokeRngs]

Stay very far away from VMWare Server 2. I tried it a while back and it was nothing but trouble. Installation was a bitch, getting just about anything to run was almost impossible and the interface required using a web browser which is something I hate with a passion.

I'm currently running VMWare 1.06 on my Vista machine and 1.05 on another machine (I think) as well as 1.04 on another machine if I remember correctly. They have not given me any real problems.

 

[Tigerbiten]

You don't need Vista for the GPU clients.
I'm running multiple GPU clients on XP without a hitch.
On my Q6600 + dual GPU boxen under XP, I set the enviromantal variable NV_FAH_CPU_AFFINITY to 8 so both GPU clients run on CPU core 3.
Then I set affinity so that both VM's can run only on CPU cores 0-2.
Both GPU clients run at full speed, the VM's fold around 20% slower.
You may loose a couple of hunderd PpD versis running on Vista but there is no hassle about signed drivers.

Luck .............

 

[SmokeRngs]

You don't need Vista for the GPU clients.


Luck .............

Unfortunately, if I want to be able to do small things like watch a movie or use certain pieces of software I do. It's on my main machine and using XP is impossible. All video will stutter and run at about 0.5 frames per second unless I turn the GPU client off. Same with the GUI on certain programs. Otherwise, I would not be running Vista on my main system.

 

[SmokeRngs]

Just an FYI for some. There seems to be a new work unit out using the A2 core. There is now the 2669. It's worth the same points as the 2662 and seems to be doing the same PPD as well. Maybe I'll get lucky and these work units won't cause the client to freeze and won't need 30 meg to be uploaded. I'm not holding my breath, though.

 

[drgnfang]

Yeah, I just got the 2669. Hitting 2k PPD on my stock Q6600 with one of these. As a comparison the same box is crunching a 2605 WU on the other VM for 1.6k PPD.

 

[SmokeRngs]

They seemed to have released a number of new units for the A2 core. After sending back a 2669 I received a 2668 for 1920 points to crunch on. Too bad the damn client got stuck again. At least I was watching this one so I didn't lose much time.

 

[Xilikon]

2668 have some problems so they repackaged to 2669 to fix the hanging issue at end. I thought it was pulled off.

 

[SmokeRngs]

I've had problems with all of them sticking. There will be at least one and up to three orphaned A2 core processes left over which causes the client to freeze and not download another work unit.

I had two frozen clients this morning and decided to try something different. I left the client open, killed the orphaned A2 core processes and the client immediately grabbed another work unit to crunch on. I could have sworn there was a similar problem early on with the SMP client and the A1 core. However, it seems nothing I do alleviates the problem. I did have one client which uploaded one of the A2 core work units and immediately grabbed another work unit last night but that only happens about 1 in every 4 work units.

 

[APOLLO]

Is there any indications from Stanford that they at least know about the hanging problem?

 

[Xilikon]

Is there any indications from Stanford that they at least know about the hanging problem?

Hanging at end ? Yes, they do know and they are working on it as we speak.

 

[SmokeRngs]

I just discovered something about the hanging problem. I always thought the problem was that the client wasn't shutting down one of the A2 core processes. I was mistaken about that. As I was watching the process list on a client which was just finishing up I saw that all the A2 core processes were stopped and no longer present in the process list.

I figured that the client was going to grab a new work unit and start running again. It seems I was mistaken. It did not download a new work unit and was stuck like usual. Looking at the process list there was a rogue A2 core process sitting there. As usual, after killing the core process it downloaded a new work unit and started up again.

Just thought I would pass that tidbit of information along.

 

[Xilikon]

I just discovered something about the hanging problem. I always thought the problem was that the client wasn't shutting down one of the A2 core processes. I was mistaken about that. As I was watching the process list on a client which was just finishing up I saw that all the A2 core processes were stopped and no longer present in the process list.

I figured that the client was going to grab a new work unit and start running again. It seems I was mistaken. It did not download a new work unit and was stuck like usual. Looking at the process list there was a rogue A2 core process sitting there. As usual, after killing the core process it downloaded a new work unit and started up again.

Just thought I would pass that tidbit of information along.

Yes, the problem is either because there is a rogue A2 core sitting or all cores is shut down but the client didn't hear the message because it's deaf so it sit waiting for a signal which never come.

 

[SmokeRngs]

Yes, the problem is either because there is a rogue A2 core sitting or all cores is shut down but the client didn't hear the message because it's deaf so it sit waiting for a signal which never come.

That's just it, while the work unit was uploading there were no A2 core processes in the process list. Every one of the four processes were shut down when the unit completed and before being uploaded. However, a new A2 core process showed up in the process list as soon as the work unit finished uploading and when it should have grabbed a new work unit. There was no reason at the time for any new A2 core processes to be loaded as a new work unit hadn't even been downloaded yet.

The weird thing is that this does not happen when the client ends up grabbing a work unit that does not use the A2 core. I've had several clients immediately grab a 2605 or 2653 work unit after finishing the upload of the A2 core work unit. It confuses the hell out of me. I could understand if an A2 core process was left over from processing a work unit but that is not what I saw happen.

 

[cereal_killer24]

Installed 2 vmware notfred clients yesterday and i got 2 a2core projects, 2668 and 2669. For some reason fahmon went bat shit crazy and said one of the was 127xxx% complete and after a while just said both clients were hung. Fah spy worked for a while then one client said it was giving 6xxx ppd by itself. If faspy was correct i was getting around 2000-2100 ppd per client on my stock Q6600.

 

[Tigerbiten]

Installed 2 vmware notfred clients yesterday and i got 2 a2core projects, 2668 and 2669. For some reason fahmon went bat shit crazy and said one of the was 127xxx% complete and after a while just said both clients were hung. Fah spy worked for a while then one client said it was giving 6xxx ppd by itself. If faspy was correct i was getting around 2000-2100 ppd per client on my stock Q6600.

Check your clock time inside the VM.
Its a know fault that the clock inside a VM, can-will run either fast or slow if VMWare Tools is not installed.

My A2's run around 2,650 PpD vs 2,050 PpD for A1's

Luck ............

 

[SmokeRngs]

Installed 2 vmware notfred clients yesterday and i got 2 a2core projects, 2668 and 2669. For some reason fahmon went bat shit crazy and said one of the was 127xxx% complete and after a while just said both clients were hung. Fah spy worked for a while then one client said it was giving 6xxx ppd by itself. If faspy was correct i was getting around 2000-2100 ppd per client on my stock Q6600.

I've seen some weird happenings with regards to monitoring A2 core work units. I think I've seen something like the 127% deal before but it eventually recovered. Also, there have been some clients which were showing in Fahmon only about half the PPD the clients were actually running. One of my clients used to do this all the time. It would show the client only producing around 1400PPD when in reality it was producing 2800PPD. I checked the frame times and they matched up with the frame times of similar work units. I know it had nothing to do with the clocks in the VMs, although that is a known "problem". It also hasn't happened in about a week or so. I don't know what, if anything, was fixed; but it's not happening currently.

 

[Xilikon]

I've seen some weird happenings with regards to monitoring A2 core work units. I think I've seen something like the 127% deal before but it eventually recovered. Also, there have been some clients which were showing in Fahmon only about half the PPD the clients were actually running. One of my clients used to do this all the time. It would show the client only producing around 1400PPD when in reality it was producing 2800PPD. I checked the frame times and they matched up with the frame times of similar work units. I know it had nothing to do with the clocks in the VMs, although that is a known "problem". It also hasn't happened in about a week or so. I don't know what, if anything, was fixed; but it's not happening currently.

Does it happen by chance that the number of frames is something like 250001 ?