Using SMP without consuming alot of bandwidth, is it viable?

[Nantes]

Hi guys,

My system: Core i5 2500k 3.30 GHz (which I plan to overclock to 4.2 GHz+ when my Hyper 212 Plus heatsink arrives)
GeForce GTX 560 (to arrive in a couple days along with the heatsink, atm I have a GeForce 8800 GTS)
8 GB DDR3 RAM

I followed the installation guide for the SMP version of CPU FAH choosing big packets and 4000 MB memory size. After installation and reboot, the CPU had all 4 cores 100 active all the time. This would be fine, if I didn't also experience a 50% drop in my bandwidth (speedometer reports 5 mbps internet instead of 10 mbps). Since my internet connection is shared between multiple computers in my household, I don't think my family members want their internet halved in the name of science =D. Thus for now I have disabled the FAH service until I can find a way to reduce the bandwidth consumption.

I read in the FAQs that to reduce bandwidth I might want to use the unicore version instead. However that is a waste of perfectly fine soon-overclocked 4 cores. The FAQs also say the SMP requires big packets. Is this really true? Can I not take advantage of the 4 cores while using small packets to conserve bandwidth?

Aditionally, I tend to leave my computer on all day except when going to sleep. So it works for about 16 hours a day. Is this good enough for SMP or do I really have to leave it on 24/7?

Lastly, I do some heavy gaming from time to time, and I'm going to play alot of Skyrim on windowed mode until February. If I start using the GPU version when my new card arrives, will it impact my gaming, or will the client automatically stop using up the resources when I launch the windowed game? Same for the SMP client? The FAQs are rather ambiguous on this point.

Thanks!

[Jesse_V]

Welcome to the forum Nantes!

The SMP will only use the Internet at the beginning of processing a WU and at the end. For a quad-core, the data transfer there is comparable to the average webpage so it shouldn't make much of a difference. 99% of the time it will be using the CPU and not the Internet. Its possible that the processing somehow influenced your speed test, or it could be coincidence, but its very unusual that the SMP would affect your actual speed like that. So my guess is that the SMP is fine. Also, F@h uses more parts of your CPU and uses them a bit harder than a normal application, so overclocks that appear stable on other burn test things may cause Early Unit Ends on F@h.

CPUs do much better at backing off for other applications than GPUs do. F@h runs at an extremely low priority, so lag from the CPU is really rare. GPU software (drivers, etc) doesn't do as good a job with this. Its First In First Out, and its my understanding that F@h sends work to the GPU in little packets, so the GPU has to finish a F@h packet thing before it can render the screen or something. Usually this isn't a problem, but lower-end GPUs can lag up. So I'd recommend that you pause GPU folding when you game. If your really after super-high FPS, pause CPU folding as well.

Good luck.
FUS RO DAH!

[7im]

The SMP client (fahcore) uses the loopback adapter 127.0.0.x for intercore communications. This never hits your network, nor the internet, but some bandwidth monitors incorrectly include the loopback adapter in their speed measurements.

So either the SMP client was downloading a new work unit, at which point, the speed would go back up after the download (a few seconds to a few minutes later), or the bandwidth tool was being a tool.

[Nantes]

Welcome to the forum Nantes!

The SMP will only use the Internet at the beginning of processing a WU and at the end. For a quad-core, the data transfer there is comparable to the average webpage so it shouldn't make much of a difference. 99% of the time it will be using the CPU and not the Internet. Its possible that the processing somehow influenced your speed test, or it could be coincidence, but its very unusual that the SMP would affect your actual speed like that. So my guess is that the SMP is fine. Also, F@h uses more parts of your CPU and uses them a bit harder than a normal application, so overclocks that appear stable on other burn test things may cause Early Unit Ends on F@h.

CPUs do much better at backing off for other applications than GPUs do. F@h runs at an extremely low priority, so lag from the CPU is really rare. GPU software (drivers, etc) doesn't do as good a job with this. Its First In First Out, and its my understanding that F@h sends work to the GPU in little packets, so the GPU has to finish a F@h packet thing before it can render the screen or something. Usually this isn't a problem, but lower-end GPUs can lag up. So I'd recommend that you pause GPU folding when you game. If your really after super-high FPS, pause CPU folding as well.

Good luck.
FUS RO DAH!

You think it can cause an EUE to my processor even with the Hyper 212 Plus heatsink/fan? That thing in theory does not let the processor surpass 65 degrees or so even at full load...

So for the GPU version I'll have to install it non-service then, so that I can stop it whenever I want? Also, if I turn off the computer, does the SMP client resume from the point I had stopped before turning off, or do I lose some of the progress?

Thanks.

[Jesse_V]

My point was just stability, not so much heat. The admins have posted repeatedly that F@h works your CPU harder than some stability tests. The PG does not encourage nor condone overclocking, but simply takes the position that if you had thought your CPU was stable at a given overclock, you may discover that it causes computation errors for F@h. I'm not implying anything, just giving you information.

I've never installed my GPU as a service, but I'd say that you might need to go for a non-service install. There are some experts here on the forum that can give you some specific advice for your case, but I'd go for a non-service. You can put a F@h shortcut in your startup folder to achieve similar results.

F@h writes checkpoints. The default is to write on at least every 15 minutes. Normally when you quit and then restart F@h it will pick up from that checkpoint. Pausing should resume from the time you hit pause, but I'm not positive how it works in v7 with its different internal workings, but I think it follows this behavior. Basically, it shouldn't ever start over from the beginning unless there's some kind of problem.

[Zagen30]

Unless you're running XP, you can't install the GPU client as a service anyway.

[Nantes]

Well, I suppose all I have to do is wait for the new heatsink to reapply the F@H service and re-test my internet. Thanks everyone!

[Leonardo]

That thing in theory does not let the processor surpass 65 degrees or so even at full load...

No heatsink manufacturer guarantees their heatsinks for specific temperatures. The core temperatures your CPU will experience depends on several variables: CPU work load, ambient air temperature, air flow levels and efficiency within the computer case, heatsink efficiency, heatsink quality/engineering, voltage applied to the CPU, manufacturing variances of the CPU, optimal installation of the heatsink and the thermal interface material, and airflow through the heatsink. With that said, assuming you install your new heatsink correctly, your CPU's waste heat should dissipate much better than with the stock heatsink.

[PantherX]

...Pausing should resume from the time you hit pause, but I'm not positive how it works in v7 with its different internal workings, but I think it follows this behavior. Basically, it shouldn't ever start over from the beginning unless there's some kind of problem.

In v6, Pausing the Client would cause the FahCore to be present in RAM but not using the resources. When the user exits this paused Client, it would cause issues for some. In V7, when you pause a Slot, FahCore is actually terminated so even if you quit a paused Slot, it will not cause any issues.