Folding@Home
- Thread starter The JoJo
- Start date
CougTek
Hairy Aussie
Handruin said:
Ohh, how I agree, It's not fair but it is the way that it is. I have 8 machines folding and my three Intel machines way out perform the five AMD's by around 3x. Now I should qualify the fact that my Intels are 3200's and my AMD's are 2500+'s but that is not enough of a justification for the difference.
With a little luck, the AMD lawsuit against Intel will help because one of the more likely compromises will be to prevent Intel from deliberately crippling their compliers and libraries on AMD machines. But at the rate such lawsuits run, it may take years. Another solution would be for F@H to benchmark their WU's on multiple machines and standardize their values but that would open up a big can of worms for F@H that so far they haven't seemed willing to do: They just want to concentrate on the "Science" which is the whole point of these donations of CPU time.
Perhaps, I shouldn't have shown Cougtek (and others here) how to optimize his machines for QMD's but it really helped his and the teams production. Was it worth the internal strife? I think so, but I understand why do you potentially wouldn't Handruin.
Life ain't fair, but its the only life we've got.
Life ain't fair, but its the only life we've got.
Deliberately cripple. Hmmm. It is Intel's compiler after all, they can do what they want with it.
The problem (as I understand it) is that no other compiler performs as well--but then Intel have invested $$$ in it, so they should reap the reward. See para 1
What compounds this, and I haven't seen stated explicitly anywhere else, is that x87 and SSE2 perform much of a muchness on AMD with any compiler other than Intel's anyway. Intel's SSE2 advantage has as much to do with raw clockspeed and data locality as anything else. AMD might get boost from SSE2, but it won't be nearly as much as Intel got percentage wise. It would redress the disparity somewhat.
x86-64 has no such limitation. Compilers cannot use the CPUID string check, so maybe Stanford should release a 64-bit binary and drive the adoption of XP-64 </removes tongue from cheek>
The problem (as I understand it) is that no other compiler performs as well--but then Intel have invested $$$ in it, so they should reap the reward. See para 1
What compounds this, and I haven't seen stated explicitly anywhere else, is that x87 and SSE2 perform much of a muchness on AMD with any compiler other than Intel's anyway. Intel's SSE2 advantage has as much to do with raw clockspeed and data locality as anything else. AMD might get boost from SSE2, but it won't be nearly as much as Intel got percentage wise. It would redress the disparity somewhat.
x86-64 has no such limitation. Compilers cannot use the CPUID string check, so maybe Stanford should release a 64-bit binary and drive the adoption of XP-64 </removes tongue from cheek>
Mercutio
Fatwah on Western Digital
Is there a non-pain in the ass unified installer somewhere?
I forgot to add, Intel's behaviour with only running x87 or SSE code if a non-Intel CPU is detected is IMHO morally suspect. If they really want to keep their work for themselves, then they should probably require that software houses building software with the INtel compiler alert users to the fact that their software will perform poorly on non-Intel CPU's. Intel's argument is or course that they are providing "convenience" for users not having to know the "mechanics" of their computer, but to contribute to over using automotive metaphors in IT, most users know whether their car requires unleaded, hi-octane unleaded or diesel. I know, I can see all sorts of problems with this analogy already, don't crucify me!
Gilbo
Storage is cool
The linux packaged versions might be nicer. The Portage package automatically sets up multiple directories for multiple clients, includes a configuration script that sets up all the clients (if you have more than one on the machine), and includes an init script to run it as a daemon. The whole thing was very nice.Mercutio said:
I wouldn't be surprised if other distributions' packages had similarly nice conveniences added on, but I haven't tried them myself.
P.S. Damn you LiamC! There were moments were I thought I had a hope...
Mercutio said:
There are several installers of which I haven't tried any of them. I just don't find installing F@H difficult enough to bother. For noobies, Like Merc., try looking at the following post F@H Tool List and find one that fits your needs.
P5-133XL said:
I'm only joking around with CougTek. I'm happy that he's folding and turning in good numbers for the team (even if he passes me). I'm only disappointed that I have a decent machine and I feel like it's not being fully utilized by F@H.
I realize they want to focus on the science, but if they can make better use of equally powerful machines, then doesn't it sound worthwhile to invest some R&D into it? I realize I should be bitching over at F@H forums and not here...I'm just making chat for the hell of it.
CougTek
Hairy Aussie
We sound like a bunch of poorly-paid and pissed-off employees and Mark is our union representative that's eating all our shit :mrgrn:
Thanks for dealing with my complaints. 
FWIW, you've been a big help with all the FAH questions that everyone asks (not to exclude all the other folks who have helped). Hopefully this sunday I can help my friend optimize his 3.2GHz P4 for folding. He's crankning out some low numbers right now because he isn't optimized for QMD's.
FWIW, you've been a big help with all the FAH questions that everyone asks (not to exclude all the other folks who have helped). Hopefully this sunday I can help my friend optimize his 3.2GHz P4 for folding. He's crankning out some low numbers right now because he isn't optimized for QMD's.
CougTek
Hairy Aussie
Have you shut down all your computers? You haven't sent any unit for more than two days. This, combined to a lesser extend (;-) )with Pradeep's production attrition, has seriously eroded our team's capacity.
Santilli
Hairy Aussie
- Joined
- Jan 27, 2002
- Messages
- 5,078
Why, oh why, even though the names are the same, are my stats on three different names? GregorySantilli, GregorySantilli, and GregDualXeon?
I've changed the names to all be the same, Gregory_Santilli, but, they won't combine. :rant: :cursin: :bounce:
gs
I've changed the names to all be the same, Gregory_Santilli, but, they won't combine. :rant: :cursin: :bounce:
gs
The JoJo
Wannabe Storage Freak
*grumblemumble* Running with my P4 3.2, 2xA64 3.2 and my T40 1.6, and not getting more units than this produced, grumble....using forceasm and advmethods on all.
CougTek said:
Nope, they're still running.
This is what I have on my 3200+ right now:
Code:
Name: p2019_BBA5
Download time: December 3 04:16:38
Due time: December 31 04:16:38
Progress: 45% [||||______]This is on my Celerslow:
Code:
Name: p2052_abeta_4mer
Download time: December 1 02:21:35
Due time: January 17 02:21:35
Progress: 94% [|||||||||_]Santilli said:
Once the points are allocated to a name, F@H does not change them. If all the names are the same now, then all future points will be allocated to that one name and all past points will be allocated to the other names. Regardless, all the points still count towards our team (Unless, you also accidentally changed team numbers)
Fushigi
Storage Is My Life
My main home PC is down WRT Folding right now. It appears to overheat (I believe) and then spontaneously reboot when F is running. Even reducing the CPU to 70% doesn't clear it up.CougTek said:
When I get a chance I'll lay a fresh coat of Arctic between the CPU & HSF; maybe the thermal conductivity is shot. May be a few days, though. In the mean time I still have 2 laptops and 2 Athlon XPs running @ 100%.
Will Rickards
Storage Is My Life
Well I added these to my service configurations for each of the two folding instances (5.02 console client)
-svcstart -local -forceasm -service -advmethods -verbosity 9
Windows XP Home
AMD X2 3800+
Lately I've had low points per day.
I was using my work laptop for a while but had to stop.
It was a Pentium M 1.6Ghz.
Hopefully changing the flags will have an effect on my daily point production.
-svcstart -local -forceasm -service -advmethods -verbosity 9
Windows XP Home
AMD X2 3800+
Lately I've had low points per day.
I was using my work laptop for a while but had to stop.
It was a Pentium M 1.6Ghz.
Hopefully changing the flags will have an effect on my daily point production.
If you are running as a service then you need to edit the registry to get your parameters to stick. Try using one of the 3rd party tools for that purpose. Or you can do what I do -- Use an old version of FireDaemon.
Will Rickards
Storage Is My Life
yes I edited the registry per your instructiuons a couple of pages back in this thread
If all you did was change the flags, then you didn't do enough. You should actually reconfig the program to allow for large WU's. The easiest way is to simply stop the service and run the console with the -configonly parameter.
Will Rickards
Storage Is My Life
checked the config and that was already set.
CougTek
Hairy Aussie
I was looking at the users stats and there's something I don't get. Silicon_Wars_-_a_New_Hope, member of Overclockers Australia, is averaging 54706 point per day. Wondering how many processors he uses to achieve this, I went to Stanford's web site and, oh surprise, he only uses 19 processors to do his enormous production.
Tell me what kind of CPU in this world can do ~2800 points a day all by itself? I'm pretty sure even an IBM Power5 can't do this. The only thing I could see is The Cell, but AFAIK, there's no client for it.
Tell me what kind of CPU in this world can do ~2800 points a day all by itself? I'm pretty sure even an IBM Power5 can't do this. The only thing I could see is The Cell, but AFAIK, there's no client for it.
Most likely, he has a large farm and is using a few machines to distribute WU's to the others. It's not supported, but there have been others that have claimed to do this. Another way to do this is using timeless WU's because with those, caching upto 10 WU's at a time is builtin which makes it easier to distribute them to multiple machines.
The behavior was first noticed with AMD Develoment Center. If you look at their user list you will see computer clusters as names. When the clusters were being tested, they used F@H and were producing tremoundous scores (50,000-300,000 Pts per day) and had only a single CPU listed. Since F@H didn't support it, they obviously had written an application that would collect and distribute WU's amoung the individual machines within a cluster.
The behavior was first noticed with AMD Develoment Center. If you look at their user list you will see computer clusters as names. When the clusters were being tested, they used F@H and were producing tremoundous scores (50,000-300,000 Pts per day) and had only a single CPU listed. Since F@H didn't support it, they obviously had written an application that would collect and distribute WU's amoung the individual machines within a cluster.
CougTek
Hairy Aussie
Thanks for the explanation Mark.
Doug, I was destroying you with a single P4 2.66GHz up to maybe saturday. I know that the first time I got in front of you, it was solely with my P4. Then I added a Duron 1.6GHz that has sent two units since yesterday. I talked about it in the Resurrecting the deads thread.
Without the Duron, I would currently have 245 points less, so I would still be 8 points in front of you at the time of this message. Bottom line : you've really been whipped by a single P4
Doug, I was destroying you with a single P4 2.66GHz up to maybe saturday. I know that the first time I got in front of you, it was solely with my P4. Then I added a Duron 1.6GHz that has sent two units since yesterday. I talked about it in the Resurrecting the deads thread.
Without the Duron, I would currently have 245 points less, so I would still be 8 points in front of you at the time of this message. Bottom line : you've really been whipped by a single P4
The JoJo
Wannabe Storage Freak
Should I run one or two instances of FAH on a P4 HT doing QMD?
The JoJo
Wannabe Storage Freak
Stanford would wan't you to run 1 instance, so they get the results faster. But if you want more points, run 2 instances on a P4 HT.
QMD just requires lot's of memory, don't see it changing the above statement.
QMD just requires lot's of memory, don't see it changing the above statement.
With QMD's they get the best points per day by running a single instance. QMD's are RAM BW limited (Not CPU) and thereby running two simply causes both to slow down as they both compete for RAM bandwidth. So, Run only one QMD instance and you will get optimum productiuon.
Bozo
Storage? I am Storage!
"QMD's are RAM BW limited (Not CPU) and thereby running two simply causes both to slow down as they both compete for RAM bandwidth. So, Run only one QMD instance and you will get optimum productiuon."
Does this still hold true with 2GB of RAM available?
Bozo :mrgrn:
Does this still hold true with 2GB of RAM available?
Bozo :mrgrn:
Fushigi
Storage Is My Life
It's not so much the amount of RAM; it's the speed of the RAM's interface to the CPU.
Bozo
Storage? I am Storage!
You would think AMD CPUs would have an adtvantage then.
Bozo :mrgrn:
Bozo :mrgrn:
CougTek
Hairy Aussie
Well in fact, they have...but only in case of the dual core. See Will Rickards ; AFAIK, he only has a dual core Athlon X2 3800+. Back when I had my single core P4 2.66GHz, he out-produced me. I did around 360 points per day while he did around 430 points. Maybe if my P4 used an 800MHz FSB, and HT, it would have slightly best his X2 3800+, but it would still have been close.
Mark wrote earlier that dual core Pentium 4 don't do any better than the single core ones since they are bandwidth-choked. That's less the case with the dual core Athlon 64 judging from the results of Will Rickards.
I'm pretty sure the fastest CPU for F@h would be a dual core Athlon 64 with an overclocked memory bus above 300MHz, which is not that hard to do. I don't think even an overcloked Pentium 4 could match this, even with the QMD advantage, but maybe I'm wrong.
Mark wrote earlier that dual core Pentium 4 don't do any better than the single core ones since they are bandwidth-choked. That's less the case with the dual core Athlon 64 judging from the results of Will Rickards.
I'm pretty sure the fastest CPU for F@h would be a dual core Athlon 64 with an overclocked memory bus above 300MHz, which is not that hard to do. I don't think even an overcloked Pentium 4 could match this, even with the QMD advantage, but maybe I'm wrong.