SMP FAH on VMware

Ah, glad you pointed that out. Changing it to NAT removes a dependency - without it, you need an external DHCP server.

wraieghaeien: Is there a web proxy at work? Does it work with the network adapter configured for NAT? Copying the working directory likely won't make it work any better in the long term, but if you copy the work dir from the Xeon home you may be able to get the WU uploaded. After you do that, clean out the /home/fold directory (and then copy /home/startfah back into it) and re-run getfold. That way you've got a clean copy of FAH.
 
No web proxie at work at all on either end.

Both are behind Nat routers. I will give it a whirl configured to NAT The weird thing is that it can download but not upload.

Once there, i'm just going to try grabbing the working directory from my Opty that is behaving.
 
Still having the issue of frames on a 2605 taking over an hour though. Is that normal? This is on a freshly downloaded copy of Mages linux VM distro.


Quad (2x Dual cores with HT turned off in bios) Xeon @ 3ghz per core. 2 gigs of ram. Running 2 copies of the VM distro with affinity set to cores 0,1 & 2,3 respectivly
Each with a gig of ram allocated to the Virtual Machine.


Tried rebooting. Tried running just 1 VM (same issues 1hr plus for frames)





 
hi guys I'm back again with another problem :D

after sometime of running vmware i keep getting this error.. do you know what causing it?

errorvmware.jpg


thanks again!!

Regards
 
wraieghaeien (how do you pronounce that!?): I don't know. p2605 is worth 1760 points, so at an hour per frame (there are 100) you're getting only 1.76 points per hour. Since a p4 2.8 benchmarks at 110 PPD, and you're getting about a third of that, I'd say yeah, something's wrong. I don't have any 64-bit hardware, and it's apparently a spurious problem, or other people would be getting it too. How many memory buses are in this system? Dual-channel? Quad?

ferds: No idea. The List of Known Bugs doesn't mention it. Post a bug report there and I'll try to remember to look it up.
 
most peopley just say "RAY" as for its correct pronounciation...your guess is as good as mine.
The best I have heard is "Ray-ghe-in"

but who knows. its a buncha letters that I never have to fight over when it comes time to sign up for stuff online.


Dual Channel.

2x 1 GB sticks.


I thought memory bandwidth as well, but I am only getting 1 hr frames when I only run 1 instance.


I'm gonna try nuking VMware in the morning, and staritng totally from scratch. Maby even using the working dorectory from my opty at home.


 
wraieghaeien said:
----2 gigs of ram. Running 2 copies of the VM -------
Each with a gig of ram allocated to the Virtual Machine.
Click to expand...

Try dropping the VM down to ~500 meg each.
That should be enough to run each client.
But with only 2 gigs of ram and all of it alocated to the VM's there may be nothing left to runn anything else.

Luck ............. :D
 
Ray also might want to try setting the affinity to 1 and 3 on one VM and to 0 and 3 of the other and see if that makes a difference. Because of the shared cache bus of the Dual Core Intel line there may be a issue there?
 
I tried 1 instance over night set to 512 MB ram with nothing else going on. Still like 1.5 hours per frame.


I am going to scrub VMware when I get back to the office and start anew. Frustrating since my Opty at home was just fire and forget.

 
Quick question for all you smart people out there.

I have the SMP client running in VMware just fine, works like a charm. But i get this feeling that i could be better. I want run my GPU client at the same time. And i have been with no effect on the SMP client at all but becouse of the lack of CPU cycles for the GPU client its slower than if i was running it on a wrist watch.

Is there anyway to lower the CPU usage on the smp client so the gpu client will run a bit quicker? I also understand its all about ballance and i dont want to restrict the production of either to the point where i am doing more damage than good.

This will all take place on the system listed below.

 
I think that's just crazy talk!

It takes more than 10% of a cpu to feed the GPU. I remember when I had a x1900xtx and was playing the with gpu client, having 2 consoles running and also a gpu console hosed the gpu too much, no matter what i set the cpu usage idle/low percentage at.
 
BLUEFISH44 said:
Quick question for all you smart people out there.

I have the SMP client running in VMware just fine, works like a charm. But i get this feeling that i could be better. I want run my GPU client at the same time. And i have been with no effect on the SMP client at all but becouse of the lack of CPU cycles for the GPU client its slower than if i was running it on a wrist watch.

Is there anyway to lower the CPU usage on the smp client so the gpu client will run a bit quicker? I also understand its all about ballance and i dont want to restrict the production of either to the point where i am doing more damage than good.

This will all take place on the system listed below.

Click to expand...


I've found that for my machine (C2D @ 3.4 GHz and X1950XTX), from a PPD standpoint, it's only worth running the GPU client when the SMP client is crunching one of the smaller WU (587 or 716 pointers). When it's working on a 1760 WU, it pegs the machine at 95%+ CPU usage and the GPU slows down to where it isn't worth it.

I do this manually, as it's not hard to tell from looking at the time steps when you've got a 1760 pt WU if you're in front of the machine a lot.

It'd be great if the final version of the GPU client needed much less CPU to run optimally, but I'm not holding my breath.
 
skindogg said:
I think that's just crazy talk!

It takes more than 10% of a cpu to feed the GPU. I remember when I had a x1900xtx and was playing the with gpu client, having 2 consoles running and also a gpu console hosed the gpu too much, no matter what i set the cpu usage idle/low percentage at.
Click to expand...

True, it takes more than 10%, but from what I've found with about 20% CPU available, the GPU speeds only drop by ~30-35% which can make it very worthwhile to run along side the SMP client in VMware.

I think the nature of the regular console client which pegs the CPU at 100% given the chance,
even as an idle process, makes running the GPU client with it a bad idea.

VMware tends to leave anywhere from 5-20% of the CPU free, so if you're running with 20% free, the GPU client works nicely. At 5%, which is what it seems to do with the new 1760 point WUs, it slows the GPU client down too much to be worthwhile.
 
skindogg said:
I think that's just crazy talk!

It takes more than 10% of a cpu to feed the GPU. I remember when I had a x1900xtx and was playing the with gpu client, having 2 consoles running and also a gpu console hosed the gpu too much, no matter what i set the cpu usage idle/low percentage at.
Click to expand...

I realize this, thats why i was wondering if there was anyway to limit the SMP client to free up cpu cycles for the GPU client.
 
mklinger said:
True, it takes more than 10%, but from what I've found with about 20% CPU available, the GPU speeds only drop by ~30-35% which can make it very worthwhile to run along side the SMP client in VMware.

I think the nature of the regular console client which pegs the CPU at 100% given the chance,
even as an idle process, makes running the GPU client with it a bad idea.

VMware tends to leave anywhere from 5-20% of the CPU free, so if you're running with 20% free, the GPU client works nicely. At 5%, which is what it seems to do with the new 1760 point WUs, it slows the GPU client down too much to be worthwhile.
Click to expand...

I had noticed this aswell, I just leave the GPU client running and see what i get from it.
 
BLUEFISH44 said:
I had noticed this aswell, I just leave the GPU client running and see what i get from it.
Click to expand...

Yeah, that's not a bad idea. I found that the total system PPD is almost the same. The only down side would be the extra power draw from a pegged GPU.

My default behavior for my primary system is to leave the GPU client running except if I happen to notice that I'm crunching a 1760 pt WU, then I'll stop it manually, but I don't worry if they're running together as it doesn't slow the SMP client down much.
 
http://arstechnica.com/journals/microsoft.ars/2007/2/19/7125

I saw this was reciently released. Will this Virtual server program allow for Virtualization of all 4 cores? Maybe someone can also give a comparison between it and the VMWare stuff. I just used a tool for converting the VMHD to the VHD for the windows client. gonna try it now to see if it works on the program. I don't have any processors with VT unfortunately or a dual core with Windows installed.

The virtual HD booted. Told me I had a 32 bit CPU and stopped. This is as far as I can go. Drop me a PM and I can send you the VMWare HD to Microsoft VD converter file in an email.
 
Well, it loaded and ran a test of FreeNAS. So it will run Linux. Still need a 64 bit guinea pig.
 
I've had UM's distro running on my T7200 for a while, and it has been great. Thanks again for all the info in this thread.

Now, how can I install EMIII or some other monitoring program within the VM? I just want to install something to report it, and I want to see and monitor it from the windows side. Is that possible?
 
This is soo fricking frustrating.

Ive tried versions 1.0.0 & 1.0.1 of VMware.

Ive redownloaded Mages image several times.

I have even grabbed a working config/image off of my Opteron at home.

I've tried 512 mb ram, 1 gig of ram for the virtual machine.

Ive tried processor affinity set to 0,1 0,2 0,3 1,2 1,3

I have run with 1 Virtual Machines and 2.

Nat for networking & Bridged.

I have installed VMware on both the C & D drives (2 seperate physical drives here)
as well as seperating VMware & the image it is working on.

And STILL i am getting 1-2 hours per Frame. on a 2605



 
NaCly said:
Now, how can I install EMIII or some other monitoring program within the VM? I just want to install something to report it, and I want to see and monitor it from the windows side. Is that possible?
Click to expand...

You can try (I forget if I turned this on or if you'll have to do it yourself) enabling Samba (look at /etc/samba/smb.conf) and use EM3 on the Windows side, as you put it. You'll enter \\vm.ip.address\fold into EM3 and it'll load the data from there.

I don't remember if EM3 does SMP monitoring (someone check me on this) but fahmon does. Google it, I gotta run.
 
I can't even get fahmon installed...I'm basically retard in linux. I managed to get the newer client when I first installed the VMware, so I used the same approach, and the wget worked for me and it downloaded. From there I'm lost
 
wraieghaeien said:
This is soo fricking frustrating.
...
Click to expand...
why not just install debian or suse or similar from scratch? Then you can use the finstall script and be done. I had a bit of trouble with mage's image, so I did that and it was no problem.

edit: this is not to say that I did not appreciate all the groundwork that mage did.
 
Because Nuking Windows is not an option.

The machine is used once or twice a month for some video production, and light 3d rendering at work.

 
wraieghaeien said:
Because Nuking Windows is not an option.

The machine is used once or twice a month for some video production, and light 3d rendering at work.

Click to expand...

I could be wrong, I often am but what I think drizzit81 might have meant was do a full blow Linux install inside a VM install.

Ultimately that’s what I ended up doing. I installed VM then popped a copy of SUSE 10.1 64bit into VM. Now I have a full version of Linux I can run and play with as well as fold at the same time.

I used the KDE interface because it basically keeps you away from the command line. Once I installed Linux I opened firefox went to Stanford’s site and down loaded the SMP client right to the desktop.

I created a folding folder in my Home directory and copied and pasted the client right into the new folding folder. Once in the folder you can right click and select unpack and it will decompress the program for you.

Then and only then did I need to open Terminal and type “cd folding”. After that you type ./fah5 –configonly and hit enter and go through the configuration. After that it’s simply ./fah5 –forceasm and boom, you are up and running.

Just a thought and as was said before none of this negates mage’s hard work.;) :)
 
BillR said:
I could be wrong, I often am but what I think drizzit81 might have meant was do a full blow Linux install inside a VM install.
Click to expand...
you are dead on. That is what I meant. Sorry if I was unclear. I used VMWare server to create a new VM. Inside of that VM, I installed {Debian Etch, Suse 10.2} and have been running happily ever after.

wraieghaeien said:
I thought you needed a paid for version of VMware to create new images/servers and the like?
Click to expand...
VMWare server allows you to create new images without a problem. In addition I have heard of third-party tools for image creation, if you want to run workstation instead.
 
drizzt81 said:
VMWare server allows you to create new images without a problem. In addition I have heard of third-party tools for image creation, if you want to run workstation instead.
Click to expand...

VMWare Server allows for creating images, and so does Workstation; the one that doesn't is VMWare Player. Server and Player are free; Workstation is still $200 or so. Its main feature is support for hardware video acceleration, which is pretty useful for game devs.
 
Rebooted my machine and reset the BIOS last night and gained 8 Min a frame on a 2604 (from 23 min to ~15 min per frame) Nice little jump for 30 seconds of work


 
I'm running just fine except according to the VMware time stamps it takes 3min per %, but in reality it is taking closer to 20min. and it realizes that and gives me this error

SMPFAH.jpg


Is that normal?
 
silverphoenix said:
I'm running just fine except according to the VMware time stamps it takes 3min per %, but in reality it is taking closer to 20min. and it realizes that and gives me this error

[snip]http://img.photobucket.com/albums/v106/silverphoenix/SMPFAH.jpg[/snip]

Is that normal?
Click to expand...

If you're working ona 2604/2605 protein, then yes.

If you're working on a 3025 or similar protein, you won't meet the deadline at that pace.
 
I had that problem myself which is why I rebooted the machine and just reset the BIOS.. I gained about 8 min per frame and the problem went away. Not sure what the problem was as that machine has been rock solid for about 3 weeks straight (when I put VMWare on it) and then last night it went flaky. I did have a hard drive crap out on me in that box yesterday so that may have been what it was but who knows.
 
My fault, mavalpha: the box I usually run BitTorrent from has been down for a couple days for reasons completely unrelated to xbox 360 modding*. And of course my guess about which drive I kept the latest version on is optimally bad; 2 of 3 drives checked and it's on the third :p It's checked and uploading now. I'm also posting it on http here. I'll edit this when it finishes uploading if I remember.

*This is a lie.
 
unhappy_mage said:
My fault, mavalpha: the box I usually run BitTorrent from has been down for a couple days for reasons completely unrelated to xbox 360 modding*. And of course my guess about which drive I kept the latest version on is optimally bad; 2 of 3 drives checked and it's on the third :p It's checked and uploading now. I'm also posting it on http here. I'll edit this when it finishes uploading if I remember.

*This is a lie.
Click to expand...
Mucho apreciado, amigo! :)

BTW, I'd still like to try out your diskless DDWRT server. I just don't have any hardware to put to use right now, still trying to get my hands on a Tualatin motherboard for my 1.4. (I'd also have to dig my DDWRT out of cold storage.)
 
Just a little tip for those (everyone?) with clock issues in the VM. If your clock is running slow, you don't have to worry about the client thinking the deadline has passed for your WU. On the other hand, if the clock is running fast on the VM, you may run into a situation where the folding client thinks the WU has expired. There is a easy way to fix that within the folding client. The next to last option in the advanced section of setup says "Ignore any deadline information (mainly useful if system clock frequently has errors) (no/yes) [yes]?". Make sure you change this to yes and you won't have to worry about the incorrect time.

Now if only there were a easy way to sync the time correctly........



 
There is a dual core patch mentioned in the long SMP thread at F@H Community Forum. I installed it on my X2 3800 and it made the times correct.
 
I want to run the VMware client. Will it run in the background of Xp Home?
E6400, Asus P5l-vm, 1gb DDR2 800, 310fsb - 2.47ghz.
 
You must log in or register to reply here.
Back
Top