Looking for Clarification..

[screen317]

Hello everyone,

I've been folding for a few years (never really learned much about the details/intricacies/jargon; I've just been folding with default parameters), and I would like to ensure that I am running F@H in the best way possible.

I have four machines running F@H; I'm only really asking about #4:

1) PS3 (on 24/7, no issues here).
2) Win7 x86 Laptop (2.16GHz duo core CPU running SMP on V7 (only SMP-- the 8400M GS GFX card wasn't meeting deadlines, so I stopped running a GPU core); no issues otherwise)
3) Win7 x64 Laptop (2.40GHz duo core CPU running SMP on V7, plus GPU core running on a 9600M)
4) Win7 x64 Desktop; specs as follows:

Operating System: Windows 7 Ultimate 64-bit (6.1, Build 7601) Service Pack 1 (7601.win7sp1_gdr.120305-1505)
Processor: Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz (8 CPUs) (Overclocked to 4.4GHz)
RAM: 16GB
DirectX 11
Graphics cards: 2 x nVIDIA GTX 590 (quad SLI)


I currently have two GPU cores running, one for each of the 590s, and an SMP client for the CPU. When all are run in conjunction with the slider turned up to 100% CPU, there is a rather large increase in lag, usually causing me to turn off the FAHControl all together until I'm done with whatever it is I'm doing. Generally this computer is on 24/7.

Main questions:

1) Is this the best way for computer #4 to run F@H?
2) Is there a way to alleviate the lag?
3) I have a passkey set up on computer #4; does it have to also be set up on computer #2 and #3? Do they have to be the same passkey?
4) Any way to capture all of the heat released and convert it into usable electricity again? ( =P )


Thank you very much for your time,

Chris Fistonich (screen317)

[Jesse_V]

Welcome to the forum Chris!

I have a similar system, though not as powerful. My desktop has a 2600K at 4.3 GHz (was at 4.4 for several months until some recent BSODs) and a 560TI GPU on 8GB of RAM. Here are my thoughts on the matter:

1) Yes, I believe so.
2) The lag is from your GPUs. Due to a variety of factors including drivers, the OS, and FahCores, GPU folding isn't the best at priority processing. I've read that F@h sends work to the GPU in bursts, and so your other applications get processing time in between those tiny burst of work. I've heard that development is working on this, but AFAIK there's no definite ETA. I've heard that having Aero turned on helps, but since you have a dual GPU setup maybe you can somehow get your OS on one and folding on the other. I don't have any experience with dual GPUs though.
3) You should be using the same passkey for all your systems. I do, and I think most everyone else does.
4) I've heard unobtainium from the movie The Core helps with that. Its creation involved something with supercooling, so you may want to look into that. But since it converted heat into electricity I'm not sure how they were welding it during construction of the ship...

[screen317]

Hi Jesse,

Thanks for the welcome and for the reply.

Okay. I'm going to switch PhysX/OS operations to the first GPU and have folding only occur on the second GPU to see if that helps.

Oddly enough I saw The Core for the first time on TV yesterday-- it was sort of funny haha..

I'll probably post back tomorrow with an update after a few WUs have finished.

Thanks again!

Chris


Edit: Something odd I forgot to mention before. In Task Manager, CPU usage is at 100% when F@H is on (no surprise), but when I turned the slider for maximum CPU usage down to 10%, the CPU usage hovered at around 85-90%. Is that supposed to happen? The SMP core was certainly affected (time remaining went from 25 minutes to a little under 4 days), but I'm curious about the high CPU utilization reported by Task Manager, unless the GPUs were really hogging all of it?


Edit 2: With only one GPU running, there is still considerable lag. With no GPU cores running (only SMP), there is no lag at all.


Edit 3: SUCCESS! Apparently previously, people with AMD cards were supposed to set their SMP to use one less CPU core for each GPU they wanted to run. Even though I have an nVIDIA, I reduced my SMP to 6 cores from the previous 8, re-enabled all GPU cores, and the lag has substantially diminished! This may be pertinent to the many other individuals having this issue, from what I've read.

[bruce]

Welcome to foldingforum, screen317.

If you're up for a little experimentation, some things you might try:

Avoid folding on the GPU which runs your monitor. Does that get rid of the lag?
Define the other GPUs as PhysX. Does that get rid of the lag?
It's possible that data to/from the GPUs is saturating the PCIe bus, restricting prompt access of the screen from the OS. Does your hardware allow any alternative way to connect your monitor to your system so that contention on the bus won't be a factor?

We're still working on identifying ways to avoid the problem of screen lag, so any data you can gather will be appreciated.

[screen317]

Hello and thank you bruce!

I tried all sorts of combinations of which GPU was connected to the monitor, which was given PhysX operations, and which GPU was folding. No combination remedied the lag.
I am connected via a DVI --> HDMI connector and don't have any alternate options on me.

Edit 3 above shows the only success I've been able to achieve through reducing the number of CPU cores for the SMP core to run on. Reducing by one for each GPU core running seems to be working for me (now there is minimal lag).

I'm glad to help in any way I can. If there's anything I can test, please let me know.

Chris

[bruce]

When you have hyperthreading, reducing SMP from 8 virtual cores to 6 doesn't make all that much difference to SMP and it does let the GPU cores and the OS do their integer-based work while FAH has the FPUs saturated. I hadn't really tried that, myself.

Reducing the %CPU setting really only works on one core so reducing it to 10% means it was at {100 + 100 + 100 + 100 + 100 + 100 + 100 + 100} / 8 to {10 + 100 + 100 + 100 + 100 + 100 + 100 + 100} / 8 whereas using -smp 6 reduces it to {100 + 100 + 100 + 100 + 100 + 100} / 8 in accordance with the way Windows calculates the number. Note that this is NOT an accurate measurement of FAH throughput because of HT.

[screen317]

When you have hyperthreading, reducing SMP from 8 virtual cores to 6 doesn't make all that much difference to SMP and it does let the GPU cores and the OS do their integer-based work while FAH has the FPUs saturated. I hadn't really tried that, myself.

Do give it a try if you can. I imagine it would be great if the behavior I'm seeing isn't an isolated case. Perhaps it could be posted in a more appropriate area where others in similar situations could try it and report back?

[bruce]

I take it that you don't believe me.

I'm not sure what kind of report you want others to make. My report is a summary of dozens of comments on the forum as well as my own experiences.

[screen317]

I'm not sure what exactly you are referring to. I don't think I expressed any disbelief..

[7im]

You asked for others to confirm. If not disbelief of a well established behavior, then what?

[screen317]

Okay I think there is some confusion on here, probably on my part.

I first posted the bit about reducing SMP to 6 virtual cores. bruce replied with

When you have hyperthreading, reducing SMP from 8 virtual cores to 6 doesn't make all that much difference to SMP and it does let the GPU cores and the OS do their integer-based work while FAH has the FPUs saturated. I hadn't really tried that, myself.

How was I to know it's a well-established behavior...

[7im]

That does like like an "or what."

FYI, Bruce is a forum Administrator, with a massive post count, and one of the oldest user accounts (user # 8).

When he makes an affirmative statement, it is mostly likely a true statement. He was only giving full disclosure that he had not tried it personally. But there are several threads where people have reported their results, so bruce's statement was from observed empirical evidence of the affirmative.

But in light of that background information, I can easily with where both sides misunderstood the other side. No big deal. Everyone understands each other now. Fold on.