Folding Forum

Does anyone have Benchmarks for SMP In XP32, XP64, Vista, Linux, etc. ?

I Just want to make sure I Run My AMD X2 4200 on whatever OS will give the most PPD, and don't wanty to have to reinvent the wheel, if someone has already done this comparison.

Also I am planing on Building a new PC and welcome any Guidance on PPD for AMD Phenom Black Edition Vs Intel's Quad Cores?

Any Input greatly Appreciated!

Kevin

Linux produces the most ppd on current SMP WUs.

KevinDL wrote:Does anyone have Benchmarks for SMP In XP32, XP64, Vista, Linux, etc. ?

I Just want to make sure I Run My AMD X2 4200 on whatever OS will give the most PPD, and don't wanty to have to reinvent the wheel, if someone has already done this comparison.

Also I am planing on Building a new PC and welcome any Guidance on PPD for AMD Phenom Black Edition Vs Intel's Quad Cores?

Any Input greatly Appreciated!

Kevin

Linux will do better with SMP

I wouldn't get a Phenom : wait for the TLB bug to be solved before getting one ...

toTOW wrote:

KevinDL wrote:Does anyone have Benchmarks for SMP In XP32, XP64, Vista, Linux, etc. ?

I Just want to make sure I Run My AMD X2 4200 on whatever OS will give the most PPD, and don't wanty to have to reinvent the wheel, if someone has already done this comparison.

Also I am planing on Building a new PC and welcome any Guidance on PPD for AMD Phenom Black Edition Vs Intel's Quad Cores?

Any Input greatly Appreciated!

Kevin

Linux will do better with SMP

I wouldn't get a Phenom : wait for the TLB bug to be solved before getting one ...

I have 2 Phenoms running the FAH clients and I see no TLB bug....

Well ... but Phenom is still slower than C2Q ... and they still overclock less

Linux will do better with SMP

I wouldn't get a Phenom...

More info needed here. The SMP linux client "currently" produces slightly more PPD on most WUs, but not all.

Also note the "currently" part. The project has a long history of continuing to optimize clients over time, and eventually the Windows and Linux clients end up running at the same faster speeds. I expect the same in the SMP clients as has been done with the CPU clients.

And unless you have a direct upgrade path to a Phenom (and therefore much less expensive), please note that the Intel chips "currently" lead in performance and are more overclockable. And with the 45nm Penryns just around the corner, Intel will maintain that performance lead for a while, IMO. As such, it would seem that continuing to purchase AMD over Intel then becomes a philosophical choice rather than a performance choice. And with the Intel so overclockable, I'm not sure the AMD side can claim a bang for the buck advantage either. Depends on your situation I suppose. Let's hope AMD can ramp up the Phenom improvements quickly to keep the CPU game very competative so people on both sides of the fence continue to get a win-win.

Unless you already have a motherboard capable of supporting an AM2+ CPU, then dont bother with them at present.

How much of the Linux advantage (15%), is due to it's being 64 bit, and so tweaks to the Win 32 SMP client will never overcome it. When will there be a Win-64 SMP client?

TheGlassman wrote:How much of the Linux advantage (15%), is due to it's being 64 bit, and so tweaks to the Win 32 SMP client will never overcome it. When will there be a Win-64 SMP client?

I don't think anybody knows.

The basic calculations in FahCore_a1 are floating point, so there's no significant difference between 32-bit and 64-bit. It's the MPI code that's running 64-bit. Since the primary problem mentioned in with Windows 5.91beta is a less that ideal implementation of MPI, it's not clear whether the speed differences are related to the MPI code itself or to the 32-/64-bit differences or to a combination of the two. (One of the fundamental rules of science: Only change one thing at a time -- and you're asking about a situation where there are at least two significant differences.)

TheGlassman wrote:How much of the Linux advantage (15%), is due to it's being 64 bit, and so tweaks to the Win 32 SMP client will never overcome it. When will there be a Win-64 SMP client?

The 64-bit advantage should be small, but we have no way to test it.

The fah smp client performance advantage in Linux is due to MPICH being written natively to linux, and only ported (kluged) to work in Windows (also a kluge in general). The Windows port of MPI is slower.

And if you think about what is different between x32 and x64, very little (if any) of that applies to folding, so nothing about making the client run as 64 bits applies to a performance difference.

And in some cases the x64 is slower, because the newer drivers needed are not as mature as x32 (not as well tweaked for performance). For instance, Windows XP-64 runs slower than XP-32.

Yes, it may take a while, but the Windows and Linux SMP clients will eventually run at about the same speed, IMO. There are so may Windows users, that eventually, everything ported to Windows will be improved to about the same level of performance as Linux. It's all the same hardware, so eventually the almost all code can be tweaked up to that hardware's maximum performance regardless of operating system.

The fah smp client performance advantage in Linux is due to MPICH being written natively to linux, and only ported (kluged) to work in Windows (also a kluge in general). The Windows port of MPI is slower.

Funny, in Linux, mpiexec is asleep, 0% cpu and Windows shows 0% as well.

And if you think about what is different between x32 and x64, very little (if any) of that applies to folding, so nothing about making the client run as 64 bits applies to a performance difference.

If the Linux 64-bit client doesn't use the extra internal registers, then it is a klugde job itself. So where does the extra 15% performance come from?

And in some cases the x64 is slower, because the newer drivers needed are not as mature as x32 (not as well tweaked for performance). For instance, Windows XP-64 runs slower than XP-32.

Were not talking gpu folding here, cpu drivers control power savings, not applicable when folding. The only WU I've run on both OSes is 2605, that's where I get the 15%. What other SMP WU's run on both OSes? Maybe you mean 32 bit SMP running on Win XP-64 or Vista 64. Not the same thing at all. 32 bit SMP should be consistantly slower on the 64bit Windows OSes then on 32 bit Windows OSes.

So to be clear, how can 32bit SMP (it is Windows only by the way) catch the performance of 64-bit Linux SMP?

I know in what I have run, the Linux SMP is faster. Though it still needs the 32libs for something.
The only linux I have right now is a USB stick on a 631, and its 3 MPF faster then my 631 XP machine, both running 2653.

On this machine (920@3200) I have a dual boot vista64/xphome, both running as service, working on the same
client, and the vista is only 1 MPF slower, once it settles down. Only have 2gig, so that may be part of it.

Not enough difference for me to go one way or the other.

There are quite a few benchmark numbers out of FahMon posted to compare ppd output on the thread listed below. I posted some benchmark data of same series WUs performance on same hardware at same settings (for a couple of different CPUs), & see a consistent 15-20% improvement when using the Notfred Linux folding CD versus Win XP Pro SP2 or XP MCE2005. BTW, I believe XP MCE is XP Pro in disguise.

For more info, see this thread: viewtopic.php?f=38&t=836&st=0&sk=t&sd=a