Linux VS Windows; Does OS matter for F@H?

[TVwazhere]

Currently planning my build for INTOS (IN THE NAME OF SCIENCE) and I've had a lot of suggestions to use a Linux variant like Ubuntu instead of windows. 

 

Is there a bonus for using Linux? Or does it just simply perform better based off the hardware? I've never used any form of Linux OS's so this will be new to me and I'm wondering if the First-Time-User experience will be worth the learning curve, or if it's only going to net me 1% differences (I'm also reading about how different Overclocking is on Linux)

[LtStaffel]

It performs better in general because it has significantly less bloatware than Windows does. Just look at the installed-on-disk sizes: Linux=1-10GB (depending on distro) while Windows=20GB+

[TVwazhere]

Just now, LtStaffel said:

Linux=1-10GB (depending on distro) while Windows=20GB+

Okay but does that translate into measurable differences in PPD though? Tha'ts my question in a nutshell

[LtStaffel]

1 minute ago, TVwazhere said:

Okay but does that translate into measurable differences in PPD though? Tha'ts my question in a nutshell

Not sure what PPD is but I'm assuming it's similar to a hashing rate. In that case no, there is no inherent "bonus" to using Linux. However, if you use an open source program and it is well optimized for Linux, then it would perform better than a Windows port. When it comes to how something performs on certain hardware it very quickly becomes a matter of drivers and software, with the only difference the OS making is the amount of overhead (or in Linux's case, lack of huge overhead) it introduces.

[Dissitesuxba11s]

@Gorgon will know more on this topic.

 

From my understanding, Linux takes less amount of bandwidth on the PCIe bus than Windows, therefore more throughput, and PPD, can be passed through from the GPU and PCIe slot. (Somebody, please correct me if I'm wrong.)

[LtStaffel]

Just now, Dissitesuxba11s said:

@Gorgon

From my understanding, Linux takes less amount of bandwidth on the PCIe bus than Windows, therefore more throughput, and PPD, can be passed through from the GPU and PCIe slot. (Somebody, please correct me if I'm wrong.)

I'd be interested to see a source on this. I don't think it's impossible, but I've never heard of it and don't see an obvious reason why this would be the case. It would have to be a matter of drivers. OP, note that Nvidia does NOT play nice with Linux with their drivers. AMD is only okay because there are open source solutions (like ROCM); when I tried to use their proprietary stuff it was difficult to get working and I never succeeded (regardless of how comfortable I am in Linux. I even edited their install script to make it more compatible with my distro but that only solved problem 1/Unknown).

[TVwazhere]

6 minutes ago, LtStaffel said:

 OP, note that Nvidia does NOT play nice with Linux with their drivers. 

That's an important note because I'm using a GTX 970 and a 750ti. 

 

Granted, this build will do literally nothing other than Folding. So performance outside of Folding is not applicable for my usage (but could be for others, so discussion on that matter is still welcomed)

[Dissitesuxba11s]

8 minutes ago, LtStaffel said:

I'd be interested to see a source on this.

Unfortunately, I can only find forum posts on this topic like this one and this. I'm not going to try to interpret what they are explaining as this is beyond my swim lane.

[LtStaffel]

20 minutes ago, Dissitesuxba11s said:

Unfortunately, I can only find forum posts on this topic like this one and this. I'm not going to try to interpret what they are explaining as this is beyond my swim lane.

From my cursory inspection of those two posts they all tend to agree that Nvidia cards perform marginally better on Linux, though the exact reason for this is not always tied to PCIe differences. Most people on those threads think it has to do with drivers.

[LtStaffel]

26 minutes ago, TVwazhere said:

That's an important note because I'm using a GTX 970 and a 750ti. 

 

Granted, this build will do literally nothing other than Folding. So performance outside of Folding is not applicable for my usage (but could be for others, so discussion on that matter is still welcomed)

It's not impossible to do with Nvidia cards, it's just that it will probably be more painful than AMD cards would be. I've gotten Nvidia cards running on Linux and I've gotten AMD cards running on Linux. Nvidia took more work but still happened. I don't know how experienced in Linux you are, but if you are pretty new then I'd consider just sticking to Windows if possible.

[TVwazhere]

51 minutes ago, LtStaffel said:

 don't know how experienced in Linux you are, but if you are pretty new then I'd consider just sticking to Windows if possible.

Well...

1 hour ago, TVwazhere said:

I've never used any form of Linux OS's so this will be new to me and I'm wondering if the First-Time-User experience will be worth the learning curve, or if it's only going to net me 1% differences (I'm also reading about how different Overclocking is on Linux)

[Gorgon]

55 minutes ago, LtStaffel said:

It's not impossible to do with Nvidia cards, it's just that it will probably be more painful than AMD cards would be. I've gotten Nvidia cards running on Linux and I've gotten AMD cards running on Linux. Nvidia took more work but still happened. I don't know how experienced in Linux you are, but if you are pretty new then I'd consider just sticking to Windows if possible.

Good advice. As you point out no seems to know why it is so, but people empirically observe that Linux systems outperform Windows in Folding. What does seem to be consistent is that these differences are less apparent when running on high-end hardware.

 

I'm comfortable in Linux but yet had to spend a couple of months to get a good feel for how to tune Linux for best performance in my use case. With a competition under-way it is likely not the right time to be experimenting but to be focusing on just producing as much as we can.

 

Of course, not experimenting and not trying to compete to the best of our ability against fellow team-mates is likely a foreign concept to the typical LTT forum member!

 

With my recent purchase of a RTX 2070 for folding I replaced a 1060 in my Linux Folding Dev. system. I've moved that 1060 to my Windows 10 Pro 64 system but I haven't had the time to get Zabbix running on the Windows system yet to actually get some measurements on folding performance to compare the the two.

 

Even then any comparison will be skewed by the fact the Windows system is my Daily Driver and is also CPU folding on 6/8 threads and is being used for other tasks where the Linux system just folds.

[TVwazhere]

2 minutes ago, Gorgon said:

Even then any comparison will be skewed by the fact the Windows system is my Daily Driver and is also CPU folding on 6/8 threads and is being used for other tasks where the Linux system just folds.

I think I'll stick to windows for now, and then after the competition month is over, try using Linux (since i'll have a 3 week average at that point) and see if the average scores are any different

[Gorgon]

5 minutes ago, TVwazhere said:

I think I'll stick to windows for now, and then after the competition month is over, try using Linux (since i'll have a 3 week average at that point) and see if the average scores are any different

That would be a really useful comparison.

[FloRolf]

1 hour ago, TVwazhere said:

I think I'll stick to windows for now, and then after the competition month is over, try using Linux (since i'll have a 3 week average at that point) and see if the average scores are any different

Mention me in the thread once you have your results

 

It would also be nice to see the differences at low end hardware like your 750ti compared to a more mid spectrum, the 970. Please also get a RTX2080Ti so we can see what Ubuntu does for the highest end you can get

 

[TVwazhere]

1 minute ago, FloRolf said:

Mention me in the thread once you have your results

 

It would also be nice to see the differences at low end hardware like your 750ti compared to a more mid spectrum, the 970. Please also get a RTX2080Ti so we can see what Ubuntu does for the highest end you can get

 

Lemme just spend $1200 for some data, sure  

[Gorgon]

2 hours ago, FloRolf said:

Mention me in the thread once you have your results

 

It would also be nice to see the differences at low end hardware like your 750ti compared to a more mid spectrum, the 970. Please also get a RTX2080Ti so we can see what Ubuntu does for the highest end you can get

 

Mention me too - I'd be interested.

My RTX 2070 is currently being driven by a 10-year old Athlon II X2 2.8GHz e/w 3GB DDR3 and a PCIe2x16 and is only producing on average 1.3MPPD

 

So I can say at this point that new GPUs will definitely work on old hardware under Linux. I could try installing Windows 10 Home 64 on that mobo when were done this month but I think that might be an unfair comparison.

[FloRolf]

9 hours ago, Gorgon said:

Mention me too - I'd be interested.

My RTX 2070 is currently being driven by a 10-year old Athlon II X2 2.8GHz e/w 3GB DDR3 and a PCIe2x16 and is only producing on average 1.3MPPD

 

So I can say at this point that new GPUs will definitely work on old hardware under Linux. I could try installing Windows 10 Home 64 on that mobo when were done this month but I think that might be an unfair comparison.

Nice! My 1080 is only sitting at 800k.

Apparently the upgrade is worth it

 

[Gorgon]

24 minutes ago, FloRolf said:

Nice! My 1080 is only sitting at 800k.

Apparently the upgrade is worth it

 

I'm cranking out 890-940,000PPD on a EVGA 1070 Ti SC Black under Linux so even with a 1080 FE you should be able to get more PPD out. I would estimate the average yield is about 915kPPD

 

EVGA 6 GB GTX 1060 Shortboard under Win10 Pro 64, 11719WU 421,326PPD

EVGA 6 GB GTX 1060 SC Gaming under Linux, 11720 WU 492437 PPD (This is a nice WU, average PPD is more like 420-440kPPD)

 

If I ever get the same WU running on the 2 1060s I'll post the results with the clocks for comparison.

 

Honestly if your 1080 is doing its job still I'd hold on until you can take advantage of the new features. I only pulled the trigger as I was planning on getting another 1070Ti for folding and hoping the prices would tank but I missed that window and when I saw the inferred PPD performance for the 2070 from the single-precision FAHBench results I pulled the trigger. Probably would have been better to wait for Black Friday in hopes of another $50 off or whatever but then there was this contest

[FloRolf]

1 hour ago, Gorgon said:

I'm cranking out 890-940,000PPD on a EVGA 1070 Ti SC Black under Linux so even with a 1080 FE you should be able to get more PPD out. I would estimate the average yield is about 915kPPD

 

EVGA 6 GB GTX 1060 Shortboard under Win10 Pro 64, 11719WU 421,326PPD

EVGA 6 GB GTX 1060 SC Gaming under Linux, 11720 WU 492437 PPD (This is a nice WU, average PPD is more like 420-440kPPD)

 

If I ever get the same WU running on the 2 1060s I'll post the results with the clocks for comparison.

 

Honestly if your 1080 is doing its job still I'd hold on until you can take advantage of the new features. I only pulled the trigger as I was planning on getting another 1070Ti for folding and hoping the prices would tank but I missed that window and when I saw the inferred PPD performance for the 2070 from the single-precision FAHBench results I pulled the trigger. Probably would have been better to wait for Black Friday in hopes of another $50 off or whatever but then there was this contest

whaaaat? Hows that possible? Is it really only the GPU making 890-940k?

What clock speeds is it running at? I have my 1080 at stock, 1885 Core and 4500 memory. i just gave it a little OC to 1950/4600 and i'm now getting 840k instead of 800-820k but that could be the WU. 

 

My highly OC'ed 1050ti is also at 210-230k with 1960/4500 speeds.

 

acc. to this speadsheet the 1070Ti generally seems to do better than the 1080 for some reason but i've got one of the best 1050Tis at least, lol. 

Your 1060 are killing it though with almost 500k compared to the average of 370k on this list. thats insane.

 

https://docs.google.com/spreadsheets/d/1vcVoSVtamcoGj5sFfvKF_XlvuviWWveJIg_iZ8U2bf0/pub?output=html#

[Gorgon]

1 hour ago, FloRolf said:

whaaaat? Hows that possible? Is it really only the GPU making 890-940k?

What clock speeds is it running at? I have my 1080 at stock, 1885 Core and 4500 memory. i just gave it a little OC to 1950/4600 and i'm now getting 840k instead of 800-820k but that could be the WU. 

 

My highly OC'ed 1050ti is also at 210-230k with 1960/4500 speeds.

 

acc. to this speadsheet the 1070Ti generally seems to do better than the 1080 for some reason but i've got one of the best 1050Tis at least, lol. 

Your 1060 are killing it though with almost 500k compared to the average of 370k on this list. thats insane.

 

https://docs.google.com/spreadsheets/d/1vcVoSVtamcoGj5sFfvKF_XlvuviWWveJIg_iZ8U2bf0/pub?output=html#

The 1070 Ti is running pretty much continuously at 2014MHz with a +800MHz Memory overclock, Maximum Power Limit, 70% Fan speed, at 50C over ambient (23-26C), 180W average power. It is the upper card in a 2 card dedicated folding system in a PCIe3 x16 slot running at x8 in a Z370 Aorus Gaming 5, Pentium G5400, 8GB DDR4 2600 with Ubuntu 18.04.1 LTS Server. The chassis is a fractal Define S filled with a mix of 140mm Noctua iPPC and AF fans with 2 fans CPU controlled and the rest controlled by the GPU slot temperature with custom curves.

 

The average PCIe utilization as reported by the nvidia-smi utility is 48MB/s for RX and 62MB/s Tx.

 

All these averages are over a 24 hour interval as reported by my Zabbix server.

 

Averages in that spreadsheet should be taken with several spoonfuls of salt. Dig down into the by GPU tab and look for newer postings with current WUs and similar clockspeeds. Also note that there is no indication if the GPU was water cooled or, if air cooled, the type of heat sink and fan combination.

 

the 1060 in that rig runs at 2.03GHz with a +125MHz graphics and +800MHz Memory offset, 70% fan, 37C over ambient, Maximum power limit, 128W average.

[Gorgon]

And then Wild Bill Showed Up:

 

I'll take option 3, sometime in December for 10,000PPD please Alex

 

Moving to Folding Month ...

Edited November 10, 2018 by Gorgon
Whoops - wrong thread

[Gorgon]

I finally got my Windows Daily Driver monitored by my Zabbix server. Had to translate a bunch of scripts from Linux utilities to Dos commands - talk about a nightmare.

 

I'm going to have to some testing once Folding Month is over but see if you can spot the difference between Windows and Linux and spot any potential issues.

 

Linux:

Acer nForce iTX mobo, AMD Athlon II X2 2.9GHz, 3 GB DDR2, RTX 2070 in PCIe2 x 16

Windows:

SuperMicro X10SL7-F, e3-1231v3 Xenon, 32GB DDR3 Cruicial ECC, GTX 1060 in PCIe3 x16 (at x8)

 

Note: These values were capture using the nvidia-smi utility polling every minute

[Gorgon]

On 11/9/2018 at 1:37 PM, TVwazhere said:

I think I'll stick to windows for now, and then after the competition month is over, try using Linux (since i'll have a 3 week average at that point) and see if the average scores are any different

Once were into December start a thread up about switching to Linux and I would be happy to help you get going. I used a couple of different threads for advice and I'll dig them up and post them.

 

I went with Ubuntu which I'd never used before, I've been using SuSE at work for about 20 years and more recently work has switched over to Red Hat.

 

Ubuntu was _relatively_ painless for a Linux Distro install which is what Canonical, the company that supports it, was going for but there are several very subtle gotchas that will bite you in the ass getting folding working on a headless (i.e. no monitor or keyboard connected) folding system.

[TVwazhere]

So The folding month has ended, which means I'm going to be looking to start benchmarking Project ITNOS on a Window 7 HDD install, and maybe within a months time I'll try switching to Linux....

 

Man am I gonna need help when that day comes