Hi Everyone,
Been messing with another server here, and have worked out I have a lot of RAM, but maybe not in the right places....
 
I have 356Gb of RAM to play with, do not have to use it all...
40 core, 4x CPU server,
32 core, 4x CPU server, 
24 core, 4x CPU server,
16 core, 4x CPU server,
 
Each server currently has different amounts of RAM, I can move the RAM around as it all works anywhere..
 
So My question is, Should I give each CORE a certain amount of RAM? and how much??
I know some projects are very high on RAM usage.... 
 
Is 2Gb of RAM enough per CPU core??
 
Appreciate any comments or advice..
 
Cheers
Steve
 

I have 32GB on a 32 core rig.  It is just enough, except for those virtual box projects like ATLAS.
I think if I had 64GB I would be better off, aligning with your 2GB per core estimate.

Excellent replies, much appreciated...
All ECC memory here..These are HP Servers...
 
I was doing some reading, and saw that I have 4 channels of memory per CPU...
 
So now I am thinking that I need to fill each channel for the maximum benefit.
So at 4Gb per channel, 16Gb per CPU, Total of 64Gb per server.... is my maths correct?? I am so bad at maths..
 
Just over the 1.5Gb per core, looks like not really enough.....
 
I have 8Gb RAM sticks but 32Gb per CPU seems a bit of overkill.....
 
More research tomorrow...
 
Appreciate all comments and experience...

Thanks for all your replies.
Will see how 2Gb per channel works and experiment with 4Gb per channel and see if there is any difference... 
 
Robbysites started to point me in the direction of VM's, just not had time to play yet..
Think I need to spend some time reading...
Any pointers??
 
Much appreciated..

devlin85
I had 32 GB in my 16 core (2x e5-2680 cpu) setup, it wasn't enough for those high memory tasks we run into from time to time, so I doubled it. 64 GB is just right.
 
I'd say 2 GB+ per core is a good rule of thumb, 4 GB per core will give plenty of overhead. When we were running atlas@home it was allocating 2.5GB of memory per task, 16 cores, 40 GB of memory! So just keep that in mind. But yeah you could go virtual for those scenarios. Just slower..
 
Side note on virtual memory:  Not really server specific, but make sure you leave some on your main windows install if you are going to allocate some on a separate drive. You used to be able to only have it on a 2nd drive, but not windows requires it on your install drive for crash purposes.

Thanks for your experience, helps very much..
Great advice on the VM..
 
Much appreciated..

Cool GTX
OK - basic questions
 
When CPU cores is used in this thread ----> are we talking physical cores only ?
 
My 5960X CPU 8 core with 16 threads and 20 MB SmartCache ----> would need how much RAM minimum ?
 
(8 X 4GB = 32GB RAM) ---  OR --- (16 X 4GB = 64GB RAM)
 
Would it be 1 VM per Core or 1 VM per Thread ?
 
I'm server poor, if I try crunching I'd have to use PCs.
 
What is the minimum CPU that would be effective at crunching ?
choices are all i7,  950, 4790K, 5830K, 5960X

Yes and No (I use Thread when HT is On and Core when HT Off). Post #1 are all CORES and wish I had some more of them.
PG LLR needs a Single Core to run well, so we disable HT or SMT.
Most all other BIONC CPU Projects you can run with HT or SMT Enabled. (Forgot to add Multi-Thread Tasks)
Also LLR do not take that much Memory as some or most other CPU Tasks.
So on your CPU you could be running 10 Tasks (10 CPU Threads) and 2 GPU Tasks, with a rig with 2 GPU.
But sometimes we even split our GPU to run 2 or even 3 Tasks Per GPU.
1 VM per Core or 1 VM per Thread = HT On Per Thread HT Off Per Core.
My biggest VM used 1 CPU Thread and 8GB of Memory so I could only run 6 of them and 4 GPU Tasks at the same time.
Time for you to Join in our Team and Learn all this as a MOD to help out others, the only other thing you can really do is become a Miner
 
What is the minimum CPU that would be effective at crunching? 4 Core 8 Threads with 1 to 2 GPUs (But you have a Great Rig Now!)
You want to leave a Thread or Core per GPU as a rule but not the case for more BOINC GPU Tasks.
I split my GPUs to use .2 CPU for Each GPU Task. So with 4 GPU running 1 Task Each they are using less than 1 Thread or 1 Core.
 
Most of us have PC, the Servers for the Most part came over from the BigAdv Days over on Folding @ Home.
You would need to ask Steve and Bill* about their Servers. *Bill1024 is our Team CPU Expert.
You do not need a Server to Run BOINC, in fact the More GPUs the better.
But you also need see what it is some of us are after in the end other than Finding Cures and other Biology or Chemistry Finding.....
Our Team is NOT ABOUT Mining for Coin. But we are looking for that next Badge for our Banners, this includes Team Badges as well.
 
I could have gone with 10 CPUs below Multi-Thread Task but I like to run my rigs a 90%-95% when Crunching.
 CPU Usage 74%-85%
6 Core 12 Thread CPU Client set to 70% of CPU usage to give me 8 CPU Tasks 90% gives me 10 CPU Tasks.
CPU Usage 90%-100%
 
Sorry Steve, I got a little off Topic.

At one time we were looking into a shared area on the Net, not sure where it is at I guess I should get on it.
I have the means but not sure how to go about giving Team Members access and also would need to be a hidden process form non team members and then we have team members that run on more than one Team so this to makes it hard. 
For the most part I manage all my rigs from BOINCstats-BAM! or just BAM! This is where I Join BOINC Projects and turn them on and off on each rig.
robbysites can give you the ins and outs of something's to look out for and again we have team members that can help setup things for you.
I use the same user name or ID as my Folding Name but the email address I use is only used for BOINC.
Most Projects sites we log on with our email address, the nice thing about BAM! and this is the First Place to setup your Account Info because here is where you select the Projects you want to run and it goes out to the Project Website and Creates your account for you.
This way your CPUID is set once and you do not need to mess with it.
PM me a User Name and email you want to use and the password and I will set up BAM! for you.
This is the same username and email address that will be used across all BOINC Projects or at least 95% of them.
If you have a Hotmail or Outlook email account you can create one under your Primary Account or even create a new Account.
 

HK-Steve
Thanks for all your replies.
Will see how 2Gb per channel works and experiment with 4Gb per channel and see if there is any difference... 
 
Robbysites started to point me in the direction of VM's, just not had time to play yet..
Think I need to spend some time reading...
Any pointers??
 
Much appreciated..

If you run a bunch, try to split them up amongst multiple HDD's. SSD's can handle more but the VM's will eat up a lot of drive space. Spinner drives are cheap. Depending on what work you are doing will determine how many  you could spin up. GCC work doesn't pull a lot of CPU until you run a lot of clients. Then you will find that you will need to dedicate some cores to just feeding them. I have had varied results with different OS's and drive types so far. When I have some pretty well locked down the project makes a change or something just suddenly stops being happy. Don't over commit a VM because they do not come close to comparing with bare metal setups from a general sense. However on a more positive spin, most VINA applications under a Linux VM still typically perform better than Windows on bare metal... Just some food for thought. 

yatz10
Maybe I'm doing it wrong:) But I don't think there is any strong expectation or need to be super diligent about keeping up with formula-boinc etc. It certainly adds some competition and fun, but I tend to run the projects that interest me (einstein, universe, wcg) and switch it up when I figure out that there's a need. Every little bit helps out (not just our crunching team, but advancing the research).

Not at all, Team Members run whatever Projects they like to run at anytime.
Don't let Team Events and Challenges put any pressure on what you want run. 

bcavnaugh

yatz10
Maybe I'm doing it wrong:) But I don't think there is any strong expectation or need to be super diligent about keeping up with formula-boinc etc. It certainly adds some competition and fun, but I tend to run the projects that interest me (einstein, universe, wcg) and switch it up when I figure out that there's a need. Every little bit helps out (not just our crunching team, but advancing the research).

Not at all, Team Members run whatever Projects they like to run at anytime.
Don't let Team Events and Challenges put any pressure on what you want run. 

I agree totally with bcav here yatz. Those projects are in the marathon races anyway.
Now me, I like to peruse the FB site. It's easy on my eyes for one and is laid out intelligently and easy to get information on what the competition is doing, and I don't have a life so....
Crunch what makes you happy and keep on trucking.

HK-Steve
380G6 which is 2x CPU's, 12 cores, with 48Gb RAM, I use this to Crunch and is my Plex software server but only 10 Cores used, 2 left for Plex.

 
Yeah! I use Plex Media Server too. Running it on a I5-2500k, still run crunching on all 4 cores, plex can fight for the resources it needs. Seems to do fine when streaming, as long as it doesn't have to actually process/transcode anything