Description

This machine is used for personal machine learning / deep learning projects.

Keywords: GPU 1080 Ti, waterblock, overclock, temperature, machine learning, budget build

CPU choice: i3-7100. Since Intel has this latest 7 series, why not choose the 7 series 14nm FinFET technology? Also, because machine learning / deep learning tasks are mainly operated in GPU, for CPU, two cores with four threads are totally enough. No demand on CPU overclock as well, then the choice comes to i3-7100, low price, enough performance.

GPU choice: Nvidia GTX 1080 Ti. This is the best performance best price GPU for personal machine learning tasks till now. Titan x (pascal) or Titan xp is too expensive but that extra money does not equal to the small performance gain compared to 1080 Ti. For full-load machine learning / deep learning tasks, it seems the usage of GPU is averagely higher than gaming, so the temperature will be crazily high. I made the stress test with founder edition fans, at 100% speed of the fan (manually tuned to 100%), temperature can be 62~65 degree C, sometimes to close to 68 degree C if the system ambient is hot. The other problem is manually tuned 100% fan speed is too loud like a jet going to fly high up into the sky. Therefore a watercool is a must for ML/DL applications. Lower temperature always means better performance and long life-time duration.

GPU performance with watercool system: 10.5 TOPS/s ~ 11.5 TOPS/s (operations per second), it depends on the main GPU frequency setups. Can be overclocked to quite stable 1760MHz ~ 2050MHz (cannot get more than 2050MHz). Temperature at idle: 24~28 degree C. Temperature at full-load with ML/DL applications with OC 2050MHz: 42~45 degree C (depends on ambient, can be lower, also depends on watercool fans, I didn't use many of them.)

Motherboard choice: H270. No need to CPU overclock, GPU overclock can be done by Nvidia-settings (with cool-bits setups). This works perfectly with i3-7100. Price of this motherboard seems to guarantee a robust performance, I mean no system glitches from motherboard at least. Not a big fan of doing motherboard BIOS, so didn't choose any 170 based motherboard. For long term, 1080 Ti will be taken out from this machine, and this machine will become some other personal project's server. So only 1 pcie slot does not matter. H270 will be enough for other maybe raspberry pi or similar thing related home projects.

Memory choice: 2*8GB. For 11GB GPU memory, total 16GB system memory is quite enough.

Storage choice: low budge SSD. 240GB is enough. HDD using 3TB to storage big data.

System power choice: 650W. This is enough for current system build. Current power consumption will be no larger than 500W, including full-load GPU.

Case choice: Manta. Just aim for sexy looking, nothing else.

Other choices: two monitor is a must, good keyboard for coding is a must, good headphones to enjoy music and videos is a must.

Long term upgrade plan: Expand the system, or say, taking the 1080 Ti out, go to four 1080 Ti all watercooled, with maybe x299 motherboard (not existing yet) and a good 6 or 8 core CPU. A large monster case to put all these in. And a much more powerful system power supply.

The build process was not that smooth because of the small Manta case, but definitely worth it, Manta is sexy. A lesson is learned that never make the screws too tight till the last assemble step for everything. Whole machine took roughly about 4 hours. My lovely wife helped build GPU watercool for me:) The tubing for GPU watercool is the most headache part, I offered labor, of course.

Comments

  • 25 months ago
  • 7 points

I sense a...

love the stock cooler lol

  • 25 months ago
  • 2 points

bottleneck? at least not for me, I got the full power of what I need, hahahaha

  • 25 months ago
  • 1 point

really-wow!

That speaks well for the power of the i3 lineup still

  • 25 months ago
  • 1 point

slightly lolol

  • 25 months ago
  • 2 points

damn! 1+! i have never seen anything like that. a custom cooled gpu and a stock cpu cooler. And an i3 with a 1080Ti? i think you just won the "case more expensive than the cpu" award. ;D nice build anyways

  • 25 months ago
  • 2 points

Hahaha, isn't this build funny! Thanks for the award from you!

  • 25 months ago
  • 2 points

you are welcome :D

  • 25 months ago
  • 1 point

You mentioned OC via coolbits - are you using Linux?

  • 25 months ago
  • 1 point

Yes, Ubuntu. There is how to do this if you search on website. Set coolbits to 4 if you want to control GPU origianl fan speed. Set to 8 if you want to overclock it. For my case, I set 8.

  • 25 months ago
  • 1 point

Love seeing fellow Linux users around here!

So how do you get your overclock settings to persist across reboots? I haven't had good luck with that.

  • 25 months ago
  • 1 point

I haven't tried yet, I assume it can be done by setting the coolbits to be 16 (Enables overvoltage using nvidia-settings CLI options) and write a auto startup bash commands to tune it.

  • 18 months ago
  • 1 point

"nvidia-smi" changes the power limit on the card and is separate from "nvidia-settings"

You can query your maximum power limit with the following command:

nvidia-smi -a -d POWER

For my GTX 1060, If I wanted to set my card to it's maximum power limit, (180W) I add this to my startup script along with making nvidia-smi run sudo without a password using visudo:

sudo nvidia-smi -pm 1
sudo nvidia-smi -pl 180

Unfortunately, trying to set "GPUOverVoltageOffset" results in nvidia-settings telling you it's a read-only attribute even with Coolbits enabled. Pascal cannot be voltage tweaked in Linux, only the power limits can be changed.

  • 25 months ago
  • 1 point

If u r using Linux, don't u need another video card to light the system up? since I think under Linux you can't take full use of GPU if that GPU is used for video output. Or u just ssh to that rig?

  • 25 months ago
  • 1 point

GPU is smart enough to separate the system 'light up' and the 'computational work'. But the system light up will use about 500MB memory of GPU.

  • 25 months ago
  • 1 point

Yep, so I am considering about adding one more 'light up' card

  • 25 months ago
  • 1 point

May be you can try just use CPU, the intel CPUs has integrated graphic module? And just let CPU work, disable the GPU video output.

  • 25 months ago
  • 1 point

That thing looks awesome! Nice to see a build that isn't geared towards gaming!

  • 25 months ago
  • 1 point

Thanks! Yeah, I didn't see many machine learning targeted builds here. So wanna share the experience with people looking into that. This build is under 2000 watercool GPU, call it a budget build for ML/DL.

  • 25 months ago
  • 1 point

weird parts that caught my eye but then i read the description and understood what u're going for and now I knw abt machine learning , more like Dota learning ! m interested in the frame rates that u're getting with i3/1080ti how does it perform/look with max settings @1080p ? and u could've gotten a Ryzen 5 1400 and a good 90$ B350 mobo for slightly pricier than a i3/h270 combination but with so much more performance gain, in the end it's still an awesome build +1

  • 25 months ago
  • 1 point

Did you see the DOTA2 icon on my screen? It's a long time since I played dota. That's just to confirm this GPU is working well:) I didn't test how many frame rate it can get form the i3/1080Ti, maybe for big games it cannot get the full 1080Ti potentials. Not familiar with AMD setups, I could do a better research...

  • 25 months ago
  • 1 point

yep didn't miss it , and yeah of course it's a great game along with being one of the most optimized games for Linux , a good choice for testing the gpu (with the reborn update the game now craves a good gpu more so than a good cpu ) u def should check out AMD Ryzen 7 (8 cores cpu 300$ to 500$ range ) also there's Ryzen 5 or even wait for ryzen 3 (a 4 core unlocked cpu for the price of an i3) for now your i3 gets the job done stick with it ~

  • 25 months ago
  • 1 point

Yeah, looks good, AMD surely has lots of potentials, the 8 cores should be enough for my future plan of building another big system with 4 GPUs. But I will need to study about the PCIe lane number when going to that direction... Thanks for your comments and encouragement!

  • 25 months ago
  • 1 point

Good build! I have a question also and that is, what do you mean by "deep learning"?

  • 25 months ago
  • 2 points

You can refer to https://en.wikipedia.org/wiki/Deep_learning If cannot see that, just think about simple linear algebra and matrix multiplications:)

  • 25 months ago
  • 2 points

ok thx!

  • 25 months ago
  • 1 point

Now here's a loop build that isn't a major gaming PC. +1 for the loop and hope your deep learning goes well. It's different, in a good way. :)

  • 25 months ago
  • 1 point

Thanks! Thinking and doing differently has lots of fun:)

  • 25 months ago
  • 1 point

Nice build but why not liquid cool cpu as wel?

  • 25 months ago
  • 1 point

I didn't use cpu very much or high-load, so currently temperature of cpu is something like 32~35, so it's ok.

  • 25 months ago
  • 1 point

great buiild, im trying a similar build with the 1080 ti also. Anyone mind checking it out and see if im all set and ready to go lol https://pcpartpicker.com/user/Lightsand/saved/422GXL thanks!! :D

  • 25 months ago
  • 1 point

If you want to do the watercool for 1080 Ti, be sure you get pre-calculations of what kind of pip work you need in the case, because pip works may conflict with big cpu fan, 212 is not small. Otherwise, you're good.

  • 25 months ago
  • 1 point

i see, i do want water cooling but later lol. Thanks for the input :D !

  • 25 months ago
  • 1 point

I changed my build a bit https://pcpartpicker.com/list/KJnWCy needed it to be the budget of 1500 sadly. Do you have any idea if im good with this build because im planning on running 3 monitors at least. also idk how the ram works is it compatible ? i may want to overclock in the future. Sorry for the bother just really want this right :( thanks for your time :D

  • 25 months ago
  • 1 point

Now your issue maybe the motherboard. You can search the details of your choice, that only supports DDR4 2400/ 2133 MHz quoted from their website. Also, do you really want a Intel B250 chipset? B250 has only 14 PCIe lanes, that means you cannot fully utilize your very-good CPU and good-GPU, these two both support 16 PCIe lanes. If you want to control the budget, reduce it from the SSD. Buy a smaller SSD and a OK size HDD. Also, I see some website is doing the i3-7100 discount, that means if you choose a just ok i3, your price will drop another 100+ dollars. Also, no need for 32GB RAM, unless you have good justification that you really need them. In short, I suggest you downgrade CPU, Memory size, SSD size, upgrade motherboard.

  • 25 months ago
  • 1 point

Oh wow, yeah honestly the motherboard i dont really know much about i should upgrade. and the Ram also idk if im going to need that much worth downgrading as you said. I will heed your advice thanks for everything ! :D

[comment deleted]
  • 25 months ago
  • 1 point

So I love the idea of this build and am looking to build a similar PC for prototyping at home. However, I wanted to try out some modifications and wanted to get your opinion.

1st, I had some concerns over data loading performance. How has you experience been with loading larger data sets (especially image data)? Along this line of thinking I wanted to replace the i3 with an i5 with a higher core count and upgrade the SSD and Ram for slightly better performance on this end. For RAM i was currently looking at one of the Corsair 2x8GB models and for ssd i was looking at the samsung pro 256 gb. However, if you are not seeing data loading latency, it doesn't make sense to forcefully upgrade these.

Another reason to bump the processor up is that Matlab recommends an i5. Do you think I would need a CPU fan for this setup with an I5 and would it fit in the miniItx case. I saw that an i3 worked for you purposes but I was wondering how hard it would be to use an i5 (+ fan?) for this build

  • 25 months ago
  • 1 point

Thanks for your comments and great to see you want to do the similar ML applications at home. Yes I do see the data load latency because of SSD. A good SSD will be enough, CPU and RAM has little to do with that. But if you are going to use Matlab, which mainly work with CPU? You'd better go with i5 or even i7. I do see some kind of throttling on GPU because of CPU (an easy test is benchmark software scores slightly lower than what I can find on web, because these benchmarks utilize CPU a lot as well), but this impact is not much for GPU calculations. If your program do need lots of CPU operations, besides of the GPU operations, you will need to upgrade to i5 or i7. As for the CPU fan, i5 or i7 worth a water cool fan, find an all-in-one water cool fan, especially if you choose overclockable CPU (with 'k' name on the CPU). That kind of AIO fan is not large, choose a 120mm radiator, it is good with mini ITX case.

  • 25 months ago
  • 1 point

Also, what's your thoughts on the new Hydro GFX GTX 1080 Ti . I dunno if I feel confortable with installing a watercooling solution myself if there is a comparable method for only 100$ more. Are the FE cards better than the MSI ones?

  • 25 months ago
  • 2 points

Alight so I have a build that I wanted to throw your way since you seem pretty knowledgeable. I basically have the same build as you with some part upgrades

https://pcpartpicker.com/list/6bGyTH

  1. Upgrade to Samsung 960 EVO 250 GB SSD. For large datasets that wont fit in memory, I will probably use this disk as an intermediate store while loading from memory.

  2. I went with a Kaby Lake CPU just incase I get an itch for more performance and decide to upgrade to the i7-7600k like all the other builds have. Honestly the I5-7500 was just as cheap as the i5-6500 so I thought might as well go Kaby. Its nice that its good with the stock cooler but I chose a mid-tower case instead of the mini incase I want to add a cpu cooler some time in the future

  3. The PSU is rated at 850 watts (same price as the 750 watt). I wanted to get higher than 650 incase I go for a 2 GTX 1080 TIs in the future.

  4. For the GPU, I was thinking of going with the Hydro GFX GTX 1080 Ti. It runs peak 40-50X for only about $899 so I thought that it might be a good solution for someone who is very hesitant to look at the 1080ti much less install a cooler.

  5. I had to get a motherboard that supports the z270 chipset and I honestly don't know too much about motherboard selection. I just went with the ATX one that had no compatibility issues and seemed to work with the GTX.

Thanks for posting the build. Good inspiration for a first time builder.

  • 22 months ago
  • 1 point

Hi! As a complete beginner in PC building and someone who's building a rig purely for deep learning purposes- this is a big help. Thanks. We've already got a Titan X and are thinking of getting the rest of parts following your list blindly. If we've got some spare budget- and could allow a bump up in the processor, which one would you suggest?

  • 21 months ago
  • 1 point

This is a beautifull build man. I love builds that don't make much sense on paper, this thing takes it to the extreme. Must be fun. A watercooled 1080 ti and an i3 on stock cooling is very wacky. Im going to make a similar thing with ryzen and a passive cpu cooler; watercooled gpu as well. Thanks for doing somehing different, this is awesome. Good luck on that future upgrade idea, that sounds awesome.

  • 21 months ago
  • 1 point

Mind if I ask, what is "machine learning" ?

It's my first time seeing a PC with an i3 paired with a GTX 1080 Ti, so i'm simply wondering what use this machine has.

  • 20 months ago
  • 1 point

I also want to build a small, portable Deep Learning machine with a Mini ITX motherboard, so your system came up in my search.

What is a EK GPU watercool system? Do you refer to a kit like this one? https://www.ekwb.com/shop/kits

  • 16 months ago
  • 1 point

my i3 bottlenecks my gtx 1060 6gb so i can only imagine a 1080ti..

  • 12 months ago
  • 1 point

Isn't watercooling a little overkill for an i3 7100?

  • 23 months ago
  • 0 points

In case anyone is interested, I put together a similar build without water-cooling and going with an AMD system instead. https://pcpartpicker.com/b/trD2FT

[comment deleted by staff]
  • 18 months ago
  • 1 point

Comments like this will not be tolerated on our site.

[comment deleted]
  • 25 months ago
  • 1 point

haha, you know what, I didn't plan to go with that cpu cooler, I prepared one coolermaster 212 but that does not fit in the case with all those tubes easily. Anyway I don't need fancy cpu fan for my application:)

[comment deleted]
  • 25 months ago
  • 1 point

true but for what he is doing a good cpu cooler isnt requiered.

[comment deleted]
  • 25 months ago
  • 1 point

Thanks, I do check out your build using the 1080 Ti one, you mentioned 'The FE used here got up to 83C during stress-testing with fans set to 50%', true, I don't know why in Ubuntu by default it is locked to 50% fan speed, even hitting the frequency limit temperature of ~85C. I have to write a script and make up my own fan speed curve v.s. temperature to cool it. But 100% is way too loud for me. That's why I changed to watercool plan.

[comment deleted]
  • 25 months ago
  • 1 point

Before I wrote my own scripts to control that in linux, I did research a little bit, the answer is: simply no compatible version in linux. Now I don't need scripts because of watercool. Anyway you can search 'ubuntu gpu fan curve' in github if you need them, I do share the scripts there.