F@H

[Pokey]

Since this is a new thang, I thought this might be helpful.

This is the result from a Xeon 2.4Ghz with 1.0GB ram
Processing the p1910_ALA-depeptide worth 450 points
I'm running FAH502_console in second folder assigned as Machine ID 2
with flags: -local -forceasm -advmethods
FAHCore_96

Right now I am at 84% complete in approx. 57 hrs.
If my math is correct, that converts to approx. 189 PPD.

I'll let you decide if it's worth it. It seems to generally be higher PPD but not by a lot.

Thoughts??

 

[Insidious]

Thanks Pokey!

I've been curious about those new WUs. I don't know If I'll get them assigned to me because I don't have SSE2 (AXP Barton), but I'd love to get just one to add to the benchies.

Now we just need to find someone with an FX55 or some other Killer rig to add to yours.

-Sid

 

[GLeeM]

Is that computer a dual rig?

And how many instances (of what) are you running on it?

 

[Pokey]

Not a dual rig. Single processor with HT. I'm running GUI as ID 1 and CLI as ID 2

 

[GLeeM]

What is the other WU?

Be warned that two of these QMD at the same time can cause problems, one can freeze for a while. I had one not doing anything for 8 hours!

That's not too bad ppd, but seems like it should be better.

How many minutes per percent is it taking?

On my dual Xeon running a p1911 and three (!?) :roll: E@H instances the QMD is taking ~25 minutes per percent, which should give ~260 ppd (Is my math right?)

EDIT: ram type and speed can make a huge difference on these WUs.

 

[JeffCos]

Originally posted by: GLeeM


EDIT: ram type and speed can make a huge difference on these WUs.

[/b]

So what if I only have 512mb of pc2700ddr. Even though that's plenty to take any WU, would adding another 512mb make it crunch faster?

 

[Moltres]

Originally posted by: Insidious
Thanks Pokey!

I've been curious about those new WUs. I don't know If I'll get them assigned to me because I don't have SSE2 (AXP Barton), but I'd love to get just one to add to the benchies.

Now we just need to find someone with an FX55 or some other Killer rig to add to yours.

-Sid

**WARNING: The following post contains spoilers for the 60K race...hehehehe**

I have an Opteron 250 running FAH. It has 1GB of 3-3-3-8 memory per processor. It just finished this: Protein: p1130_L939_K12M_355K. Is that a new one?

I also have an FX55 with faster 2-2-2-2 memory that I will start maybe today. I don't know much about all this but if you tell me what to look for, I will tell you how long it took and how many points, etc.

Also, is there any benefit to running two instances of FAH on a HT P4? Seems I remember seeing something about it not really being beneficial and that it could even slow things down from the two fighting with each other. I have a 2.8 and 3.0 P4 that just started crunching. I only have single instances on each of them.

Finally, AMD has 3000+ designations on a lot of processors. Is there any speed difference in AthlonXP 3000+ vs. Sempron 3000+ vs. A64 3000+?

 

[GLeeM]

The QMDs are only going to Xeon and P4 with SSE2, something about not being able to distinguish between the AMD with SSE2.

Amount of ram will determine if you can get QMD - need 512 MB, IIRC.
I suppose amount or ram could get more speed if you are running two WUs and/or other things and the OS is using hd for VM.

I am at the edge of my understanding with the stuff below

Type and speed of ram:
I read at the F@H forum that dual channel was ~50% faster than single channel.
QMD uses available bandwidth - doesn't leave much for other WU instances.

!@#$% Don't try to run four on a dual Xeon %$#@!

You will not gain, in fact you will lose ppd.

I think there will be another version of the core soon - this one has problems if more than one QMD is run, it just sits there not crunching but still uses resources.
With earlier version core I ran two on HT and only got 1 or 2% gain, not worth it really.

Now to the benefit of HT: "Your mileage WILL vary"
With regular WUs and some BigWUs you can get a boost in total ppd (up to 30%, but usually lower) by running two instances of F@H. It will take longer to finish each WU, but you will finish two WUs sooner.

A new WU is made from the results you send in when you finish a WU. More WUs down the line (in the sequence) can not be made ahead of time.

The researchers at Stanford want to finish a sequence as soon as possible. It's just like at work when they give you a job and you ask when they want it done and they say "Yesterday" or "last week".

You cannot run two GUI instances at the same time. One GUI and one CLI or two CLI. Also you need to configure the second instance as a different MachineID than the first and add the -local flag to the startup shortcut of both instances.

 

[Insidious]

Great info GleeM.. Thanks

I can't bitch because I knew my procs. would remain unchanged to support the graphics updates this time around.... we have our priorities ( ... things must die colorfully ... )

-Sid

 

[Pokey]

GLeeM

I think the GUI was (is) running a P1130 protein, while the cli is currently processing a p1902_32_water_molecules. (125 points)

The cli is showing about 10 min per 1% for the p1902 which translates to about 180 ppd.............close to the previous units (p1910) ppd, which I noted in the opening post.