Ubuntu Desktop Installation Guide for F@H

musky said:
I haven't heard reports on BFS "bitching" with AMD systems other than feather (and we haven't ruled out "operator error" on that one.) What you will see is a big performance increase one way or the other. Install without BFS and run a few frames, then install BFS and run a few more frames to see which is faster.
Click to expand...

i need to do it on a vanilla install, i did tweaks (mainly GUI based) after the fact which shouldnt have been an issue, but who knows.....other than that, the machine did have bitch fits with linux before i installed with a DVD(as opposed to a CD rom)....dont ask why because i'm new to linux as well
 
Does anyone know an approximate gain ratio for each of the tweaks? For instance, I've heard you can get up to 30% over windows. How much of that gain comes just from linux? How much more from BFS? If I leave the GUI running so I can use it as an HTPC, is that going to mitigate much of the benefit that Linux offers?
 
Depends on your setup, but for me on 2p SR2s, cut and paste from the original thread:

If I remember right, BFS is not a nice to have, it is a necessity on SR2s...

As tested by Musky and myself, leaving GUI running has no measurable impact on SR2 class rigs. Musky does show instructions for running without the GUI if you want to.

One note - the Krakken wrapper is an affinity lock for 2p / 4p rigs - this makes it faster on a dedicated machine, but will run worse if you are multitasking a lot - the nature of the beast.

(Windows average p6901 - 11mins 22sec)

Ubuntu DFS kernal: p6901. Numa disabled. C-states enabled. Captured p6901 from Musky.

10:18
10:17

Benefit - 10.5% frame time = 16% better PPD

161,000 ppd up to 186,700 ppd.
Click to expand...

Later - this is on different rig - ADDING Krakken - Tears wrapper
Oh hells yeah - 23 seconds better.:eek:

SR2#3 @4.3Ghz: 6901 benchmark file:

Frame time without - NUMA Disabled:
9:54 = 197,700 ppd (windows= 11:12)
Frame time with wrapper - NUMA Disabled:
9:53
Frame time with wrapper, NUMA Enabled:
9:30 = 210,000ppd

Finally broke the 200k barrier. :D So now we are up to +46,000ppd and 28% better PPD than windows.
Click to expand...
 
Patriot said:
I had issues on mine... I would give it a shot... we need the data point either way...
Click to expand...

W.Feather said:
on which system ? your 2p system ?


I mean hell give it a go, since worst case you just boot into a different kernal...
Click to expand...

Patriot said:
yeah had issues on my g34
Click to expand...

musky said:
I haven't heard reports on BFS "bitching" with AMD systems other than feather (and we haven't ruled out "operator error" on that one.) What you will see is a big performance increase one way or the other. Install without BFS and run a few frames, then install BFS and run a few more frames to see which is faster.
Click to expand...

Sorry I have been in a haze today... headache all day things just not computing right...

Ck headers installed just fine on my amd system... those kernel mods include BFS

which "worked" perfectly.... just killed my tpf...4min....
so it works as in improves tpf on SR2s with 24 threads but hurts 2p and 4p amd 24 and 48 cores...

now that we have a 1p amd octo... I wonder what ck does....
 
I've got BFS on the dual hex core AMD, runs fine, took a couple of seconds TPF of each frame. Currently crunching on a 2684 :rolleyes:

Switching from Windows to linux on this machine was good for around a 62% PPD increase - 36.8k ppd to 59.5k ppd - BFS and The Krakken have raised it a litle more.

H.
 
Haitch said:
I've got BFS on the dual hex core AMD, runs fine, took a couple of seconds TPF of each frame. Currently crunching on a 2684 :rolleyes:

Switching from Windows to linux on this machine was good for around a 62% PPD increase - 36.8k ppd to 59.5k ppd - BFS and The Krakken have raised it a litle more.

H.
Click to expand...

so I wonder where the cutoff is....16 or 24 cores lol...

deff still working at 12 with dual hex... amd side
 
Patriot said:
so I wonder where the cutoff is....16 or 24 cores lol...

deff still working at 12 with dual hex... amd side
Click to expand...

I suspect it has nothing to do with the number of cores. My dual X5550 rig was significantly better on BFS. It is all about whether or not the stock CFS scheduler can load all the cores 100%. For all of my systems, CFS would leave 1 core unloaded always. All of mine were Intel, so that seems to have some effect.

Keep the reports coming, especially with AMD rigs.
 
musky said:
I suspect it has nothing to do with the number of cores. My dual X5550 rig was significantly better on BFS. It is all about whether or not the stock CFS scheduler can load all the cores 100%. For all of my systems, CFS would leave 1 core unloaded always. All of mine were Intel, so that seems to have some effect.

Keep the reports coming, especially with AMD rigs.
Click to expand...

problems probably 2 tiered
soo it appears cfs has an issue with ht....
I have never seen load issues at least not more than 1-2% on a single core....

bfs does have a core scaling limitation... but it seems to be a core and not thread limitation...
 
It may be HT-related. A 2600K has the same issue with CFS. What we need is a CFS/BFS comparison on something like an AMD X4 or X6 and a C2D or C2Q.
 
sounds like a weekend project

I have an x48 board all hooked up... not sure if it has the E8500 or Q6600 in there either way I can swap for testing...
have an x4 and x6 and dual x12
 
musky said:
It may be HT-related. A 2600K has the same issue with CFS. What we need is a CFS/BFS comparison on something like an AMD X4 or X6 and a C2D or C2Q.
Click to expand...

What we need is a database of different systems, OS's, configs. Musky, Patriot, Jebo, and MIBW check your email for an email from Patrick.

H.
 
Haitch said:
What we need is a database of different systems, OS's, configs. Musky, Patriot, Jebo, and MIBW check your email for an email from Patrick.

H.
Click to expand...

nitrobass24 said:
dont we have one of these?
Click to expand...

its not as detailed as we want lol... its a very good start though... but is lacking windows/ linux differentiation ...

Patrick wants to make a linux fah benchmark and SQL database
musky has a windows bench that would be a good place to start for a port...
 
Nitro,

Jebo's spreadsheet is locked for editing. Patrick and I are working on something better.

H.
 
if you want to get into benchmarking, perhaps we should share a captured unit(s) like we did for Windows with Muskys benchmark.

I have been using a 6901 he captured for me -if he is ok with sharing it further we can use that for apples to apples comparisons.
 
may i be worthy said:
if you want to get into benchmarking, perhaps we should share a captured unit(s) like we did for Windows with Muskys benchmark.

I have been using a 6901 he captured for me -if he is ok with sharing it further we can use that for apples to apples comparisons.
Click to expand...

if I knew how to capture....

I have a few test subjects...:rolleyes:
 
what data are we missing?

wouldn't it be better to adapt the existing spreadsheet the way we want it?
 
Haitch said:
OS Info, CPU specific info ?

H.
Click to expand...
a lot of cpu specific info is in the "notes" column. OS info is basically "windows". Other than a few results that have been noted linux.

I mean, do whatever you want, but I would just hate to see the data we've already compiled go to waste.
 
I wouldn't be opposed to some database fanciness if those guys are up to it. They can just take that data straight out of the spreadsheet.
 
Zink said:
I wouldn't be opposed to some database fanciness if those guys are up to it. They can just take that data straight out of the spreadsheet.
Click to expand...
yeah, there's definitely nothing special about the spreadsheet itself. I'm just concerned about duplicating work already done.
 
I have a lot of results that I have not added because of this possible switch...that and my less that perfect wu name keeping....
 
Patriot,

Upload them to the spreadsheet - I can always suck the data from the spreadsheet into the database.

What I'm doing with the database app is in two parts.

- A benchmark client that can run a set of known WU, recording as much data about the system as I can get

- A log monitor app you can run that will upload the same info for every WU you fold, as soon as it's folded.

With the DB we'll be able to get comparisons of kernel X vs kernel Y, CFS vs BFS, Windows vs Linux on the the same CPU, and we can get a lot of real world data with zero effort on the part of the user. Jebo's spreadsheet is a great tool, I'm just trying to take it to another level.

H.
 
Navigation links thanks to rmdashrrootsplat - that was this morning's project...
 
Just on the below, we need this.

Jebo's sheet was the godfather but I talked to quite a few people and this has been an idea out there. We need consistent WU's and data reporting to make accurate comparisons.

That being said, let's keep this discussion off of musky's awesome thread. I think those discussing this have beta forum access.

Haitch said:
Patriot,

Upload them to the spreadsheet - I can always suck the data from the spreadsheet into the database.

What I'm doing with the database app is in two parts.

- A benchmark client that can run a set of known WU, recording as much data about the system as I can get

- A log monitor app you can run that will upload the same info for every WU you fold, as soon as it's folded.

With the DB we'll be able to get comparisons of kernel X vs kernel Y, CFS vs BFS, Windows vs Linux on the the same CPU, and we can get a lot of real world data with zero effort on the part of the user. Jebo's spreadsheet is a great tool, I'm just trying to take it to another level.

H.
Click to expand...
 
Just a post to say that this is why [H] is the best.

Threads (and the people behind them) like this are a treasure.
 
Added instructions on automatically starting Langouste on boot. If you are using Langouste, you want to do this.
 
Finally getting around to installing this and running Ubuntu for the first time: the guide has been excellent!

Right now my only question is: does it matter that I don't have Core_A5 before installing the Kracken? I entered -smp 24 -bigadv and left advmethods off, but the client downloaded only Core_A3 and grabbed a regular SMP WU.

Is there something I need to do to get the A5 core or should I just install the Kracken and allow the client to work until it decides to download it?
 
If you didn't get a bigadv then Stanford is out. With that said you have to run Kraken again to get it to wrap the A5 core when you get one.
 
Aix. said:
Right now my only question is: does it matter that I don't have Core_A5 before installing the Kracken? I entered -smp 24 -bigadv and left advmethods off, but the client downloaded only Core_A3 and grabbed a regular SMP WU.

Is there something I need to do to get the A5 core or should I just install the Kracken and allow the client to work until it decides to download it?
Click to expand...

You need to have the A5 core present (although it will wrap the A3 core also.)

Two options, one is you can copy the A5 core from another machine and chmod 777 it, then Kraken -i. Another option is you can monitor the box until it installs an A5, ctrl-c, kraken, ./fah6 again.

Another thought, I never -smp 24 or 48 anymore and just let the thing go wild.
 
I'll be installing linux tomorrow night for the first time in 10 years, looking forward to a nice ppd boost on the dodeca. Thanks for the guide Musky!
 
How important is changing to the ext3 file system from ext4? I thought I was going to be given a chance to do it during a standard install to a new HDD but when I clicked next the install process began.

I decided to stick with it rather than reinstall right away, but if it's really important I will just reformat/reinstall.
 
What vm software could I use to host a Windows vm?
Posted via Mobile Device
 
Another question.

Will Windows remote desktop be able to remote into a ubuntu box?
Posted via Mobile Device
 
Aix. said:
How important is changing to the ext3 file system from ext4? I thought I was going to be given a chance to do it during a standard install to a new HDD but when I clicked next the install process began.

I decided to stick with it rather than reinstall right away, but if it's really important I will just reformat/reinstall.
Click to expand...

yup reinstall ...
 
Cool.

So would Xen allow me to run Windows Home Server on a Ubuntu desktop host?
 
I think I read your first question too fast Jebo. AFAIK, samba will only let you access files and folders on your linux box from a windows machine. It won't let you login to the machine via windows remote desktop. For that kind of functionality you're going to want VNC I believe. But this is Linux man :) You can SSH into your box and do anything you want from the command line.

I don't know much about running Xen or any type of VM on Ubuntu so someone with more experience there would have to speak on it.
 
You must log in or register to reply here.
Back
Top