Which OS folds faster?
Slick
Upstate New York
Does linux fold any faster than Windows assuming all varialbes are the same (ex, the same computer).
0
Comments
As for Windows vs Windows: Which one takes less CPU cycles. Prolly 2k.
F@H gets more CPU time on the Linux I am using it in than it did in Windows unless I made Windows useless for other things by giving F@H an exclusive use of the box.
A large part of it is the kernel in use, the hardware, and the client. Linux is so individualizable that your milage and mine WILL vary (you will get different results than I) in part because hardware handling affects F@H, and in part because how you run F@H is part of the key to F@H speed. I run it from a console, not a console interface from within Gnome or KDE. As such, it is not subject to KDE or X.
The client is the same one you use. Cores too. Linux gives more resources, as a proportion of the CPU time dedicated to folding. And bigger chunks of RAM for buffer so less gets swapped. Let me pull soem stats real quick:
Right now, there is a tinker that somehow the client ended up with -- about one in six or seven ends up a Tinker even with the right switches for Gromacs chosen. It is doing a tinker frame every 13 min and is gettign an average of 98% of the CPU's time cycles. (project 638)
With Gromacs, I can go about 6-8X more effective than with Tinker on Linux. Try 18-20 hours a WU, the 33 point kind. But, if I keep manually sending and receiving the time lost while the box folds and I am not around results in lower performance than letting it fold 25% tinkers. I also am letting Pande group send whatever as they are trying to trace where some records are going and I told them they could use this box as needed if they wanted.
So, I accept a mix and my WU data overall will not totally reflect the effectiveness of Linux vs Windows overall because I do not have lots of pure Gromacs only data. But a good tuned Linux install will give more CPU time to F@H than XP or 98, that I am certain of.
It screams through project 340 series and 799 series WUs. It will take about 60 hours to fold a project 638 WU. XP took 70+ hours on same project with a WU of same point value.
As I do not babysit it, it has had and finished a wide variety of WUs since very late June 14th while being used for surfing and such things:
WU 35
Proj WU
340 2
341 1
348 1
539 4
543 2
544 19
547 2
571 1
799 3
(these are completed WUs, I rejected five partly finished Tinkers in this time and finally stopped rejecting and just let the box do them).
This box is a SLOW P4 box. I do not know what an AMD box would do and have business records that are not going on the web on that box-- so it is always offline as far as the web goes.
John Danielson
Its not the same client, its not the same core, but they are the same WUs. You have to chose a Linux client which downloads a linux core.
Gromacs are known for better production. Its way they work and given points.
-advmethods is NOT!!! for gromacs. Its a 'semi-beta' switch. Mostly gromacs have come from advmethods but I have gotten a few genomes with it as well.
I have the Linux port of the 3.24 Linux Beta client, for gcc 3.2. In essence the C++ source used for the Windows command prompt\console client was recomplied for gcc 3.2 compatibility in the subversion of the client I am using. Exact code function by function no, version yes. The older Linux core was not nearly as effective from what I can see in the folding forums, in part because the gcc used to compile was buggy in some important ways to folding at home.
Cores used in last month are Core_65, Core_78, and Core_ca. Same core names, ported from common source base.
Clients themselves are about equal in base function, Linux gives more processor time because of the way it handles processes. For all practical purposes, any minor console client differnces are so minor they yeild very small performance differences when plotted against percentage of procesor time given to client all to itself. And I was speaking practically, before,trying to say that if on your box Linux does better it is cuz linux uses your hardware well.
For some hardware sets, Linux does not run the hardware right-- then windows will be better. for toehrs, Linux DOES handle the hardware and rocks with Linux better than with XP. In my case, the Intel Application Accellerator that was supposed to be so great had to get uninstalled.
I ran the Windows console 3.24 on XP and 98 SE on this box, and ran it on Linux. It runs better on Linux on THIS box. Same base for code, ported and NOT total rewrite, and the compilers are about equal in overall code efficiency right now. Different subdetails, but no high video either way so no issues of difference there either. And Linux low video is VESA, like the command console in Windows.
Comparing against the GUI Windows is not apples and apples due to the graphics handling code in the graphicla user interface Windows client.
If drivers and process handling were identical across O\Ss then the results woudl be same. Neither are identical,and some devices have very good drivers while others have to be cobbled together to run or run in a slower compatibility mode insofar as Linux is concerned. Linux has made huge strides, relatively, over the last 6 months, but you would have to bench it yourself as there is no absolute better here for folding on anywhere near all hardware.
But the F@H code itself is not an issue that would change performance more than 3% either way if same hardware and equally effective drivers were used. The issue is how much CPU time it gets in an hour and Linux will feed more to a modern CPU with decent amounts of RAM than Windows. Linux process management (not quite Windows thread manangement but close in that it does handle program tasks) switches in and out faster, adn if a higher priority task does not intervene then the active one gets more CPU time proportionately than the Windows kernel will give.
Linux also give equal priority to console vrsus GUI, while Windows shells console inside GUI usually. In Linux I do not need extra software to run a console client for F@H, it is built-in.
The clients pull about the same RAM workspace. Only significant difference is process handling and the Linux kernel has only so many drivers for it right now-- so hardware that will run on XP will NOT all run on Linux. Which is why I HAVE to waffle, there is no universal better here for allmachines on F@H by any means.
Machines that have their hardware run well in Linux will fold better in Linux. Those sets of hardware for which not so good a set of chipset drivers exists in Linux will not run F@H as well as Windows. Which will be better is greatly hardware dependent, and what is great for Windows may massively misbehave on Linux.
I'll start recording numbers when I have time, but I've averaged 18 hours / day @ school this week :P not good for the free time.