FAQ FAQ  Forum Search   Register Register  Login Login

PC For LONG term, heavy workloads

Post Date: 2018-01-05

 Post Reply Post Reply
Author
  Topic Search Topic Search  Topic Options Topic Options
TechSY730 View Drop Down
Newbie
Newbie


Joined: 05 Jan 2018
Online Status: Offline
Posts: 3
  Quote TechSY730 Quote  Post ReplyReply bullet Topic: PC For LONG term, heavy workloads
    Posted: 05 Jan 2018 at 4:10pm
I am getting a PC for gaming and heavy scientific workloads.
How heavy?
Days, or even weeks straight with all CPU cores and all GPUs maxed out at 100%. (No, I'm not building a mining rig; I would build it myself if I was going for a mining rig)

Though I will interrupt it when I want to play games that my laptop can't handle. :)

I understand that I will need a pretty beefy cooling solution to be able to handle that kind of load.

At a minimum, I want an latest generation Intel i7 with 8 cores (or equivalent sorts of power for AMD) and 2xNVIDIA 1080tis and 32gb of RAM.

The monitor I plan on getting is a ASUS ROG SWIFT PG278QR 27" 2560x1440 1ms 165Hz G-SYNC Eye Care. However, if cost becomes an issue, I may drop down to the 1920x1080 model with mostly the same specs in the other stuff.

Yes, my desired graphics card(s) are a bit overkill for a monitor like that, but again, I am using this for computing as well as gaming.

My biggest questions are:
* Will mid-size towers (Lumos, Velox, etc.) be able to house a cooling solution (air or liquid) that can handle this sort of workload, or will I have to go with a monster sized full-size case (Aventum, etc.)?
* Is liquid cooling of CPU and GPUs basically mandatory to keep up with this workload (given I am not going to have a datacenter like level of AC available)?
* What motherboard should I choose (I may add more non-GPU cards later)?
* How much overclocking can I get away with without sacrificing long-term reliability and computation correctness?



Edited by TechSY730 - 05 Jan 2018 at 4:20pm
Back to Top
DST4ME View Drop Down
DS ELITE
DS ELITE

Email address used to purchase matched with forums account email.

Joined: 14 Apr 2008
Online Status: Offline
Posts: 36758
  Quote DST4ME Quote  Post ReplyReply bullet Posted: 05 Jan 2018 at 4:51pm
Im a bit confused as how this can't be mining, if we are talking 24/7 load on cpu /gpu then mining or not it us all the same. Cpu /gpu don't care about what is stressing/loading them, they are effected by 24/7 work load.

Lol anyways, there are some issues but I will first answer your questions:

1. Velox is full tower, you gonna want big case refer to issue #1.

2. What is your ambient temp? If ambient temp is good then mild oc on cpu you would be fine but if not then you need lc and we are not talking regular kind here, get ready to drop some $$$ on lc, refer to issue #1.

3. Highest mobo we can get here or we need to custom order, depending on your needs.

4. Oc is dependent on heat and vcore, you want the highest oc with lowest vcore, if both vcore and heat are good your cpu will outlast system.

Now issues:

Issue #1. Few years back someone was able to convince everybody that nobody's GPU and CPU runs at full load at the same time, therefore there is no need for a CPU to have its own loop and GPU to have its own Loop, leaving you with one Loop to cool both CPU and GPU with assumption that only one will be under full load at a time.

Personally i find this foolish, in your case the above will not work very well.

You are going to want one loop (pump/block/rad/res) for cpu and separate loop for gpu to get the best cooling.

If your temps run on high side (around 80c) 24/7, expect pump failures, that is just how it is.

For dual loop you gonna want a big case might have to special order.

Issue #2. You may not have enough pci slot for everything.

So my question to you is this, is the second and first being used at the same time but separately for work?

Reason i ask is because if i removed one,and did not tell you, in gaming you would never know the difference.

Edited by DST4ME - 05 Jan 2018 at 4:53pm
Back to Top
TechSY730 View Drop Down
Newbie
Newbie


Joined: 05 Jan 2018
Online Status: Offline
Posts: 3
  Quote TechSY730 Quote  Post ReplyReply bullet Posted: 05 Jan 2018 at 6:10pm
Ambient temp averages around 70°. Not the most ideal for cooling, but certainly no "cooling in midday summer temperatures" level of cooling demand.

Ideally, I wouldn't want a very large case, but if that is what it takes, then that is what it takes. Which is why I was asking if something smaller, more mid-sized would do (which it sounds like it might not)

Yes, when not in use for haming, both GPUs will be in use for computing, but when gaming, I'll have one be running the game and one still doing computation, as so few games currently parallelize well enough for SLI to really help at all (they'll still be in SLI, because why not, but I won't count on that helping performance)

I was hoping to avoid a dual loop setup, but yeah, if the CPU alone could make the loop ineffective at cooling the GPUs, then I may have to.
What's the advantage to dual loop over single loop with two radiators (Like CPU -> radiator 1 -> GPUs -> radiator 2, with pumps and reservoirs added where needed)

EDIT: Oh, could I get away with some of this cooling overhead if I went air cooling? I understand it would be like louder than an airport and the heatsinks would have to rival the Monolith in size, but could it will work and take still take up less space than dual loop liquid cooling?

I think you underestimate how much scientific computing there is out there doing be done. ;)
I am going to be doing some artificial intelligence stuff and physics modeling, and whatever spare cycles are left over, I am donating to various scientific computing projects, which fills up the rest of the capacity. Yea, I may do some mining, but the majority of it will be either my work, games, or donated time to scientific projects. (Besides, if I was making something for profitable mining, I would be building it myself to save costs, and not be using 1080-tis)

Edited by TechSY730 - 05 Jan 2018 at 6:14pm
Back to Top
DST4ME View Drop Down
DS ELITE
DS ELITE

Email address used to purchase matched with forums account email.

Joined: 14 Apr 2008
Online Status: Offline
Posts: 36758
  Quote DST4ME Quote  Post ReplyReply bullet Posted: 05 Jan 2018 at 7:37pm
Scientific computing like haf? Not sure why you made the underestimate comment.

70f is a good ambient temp to have.

Ok so ambient is going to be help, next thing to determine temps is going to be cpu but more importantly what kind of an oc are you looking for on that cpu? I believe you said 8 core x299, what oc are you looking for on that cpu?

Not sure how your game will be if you are running something else gpu intensive, specially in sli, if they were separate (not sli) and you could isolate the gpu to do the work, then you may have the best chance of heavy gpu work and gaming at the same time but for best results it us best to do one at a time. Of course i don't know your apps and games so if it is something you can do already then you can do it with new system, if not play that one by the nose.


As for dual loop, temps and etc, first we need to know the range of oc, that will give us an idea of what kinda heat, from there we can see what options there are.

Separate loop will always have better temps due to less heat to dissipate, assuming we are using same exact rads.

My DST in my sig, has 4 gpus, the heat is a bitch and they have their own loops, but that us my fault, normally minimum you want single rad (120) per gpu, for best results 240/280 rad or higher per gpu. Triple rad for cpu

It all depends on how far you want to push things.

The Highend systems i have seen on line from benchers are mostly dual loop and setup the way i explained, the single loop you explained would have one less reservoir and possibly a pump, in short they are not that big of a deal as far as space, the big parts are going to be the 2 rads.

You could get noctua with triple fans, but no air cooling will match custom lc performance.

Im imagining this system doing fah 24/7 on both cpu and gpu, that is lot of work and heat, that being the case (your heavy workload being the same as fah) no way is aircooler keep up with 24/7 work load under oc.

Edited by DST4ME - 05 Jan 2018 at 7:38pm
Back to Top
TechSY730 View Drop Down
Newbie
Newbie


Joined: 05 Jan 2018
Online Status: Offline
Posts: 3
  Quote TechSY730 Quote  Post ReplyReply bullet Posted: 09 Jan 2018 at 7:25pm
I made the "underestimate" comment because you wondered how this could be anything but mining. :P

I'll probably just turn the SLI off most of the time so I can do one GPU doing gaming and the other doing other computation work. I'll only turn it on for the few times where I want to go maximum show-off mode in the few games that can take advantage of it.

Let's see, the Intel i7-7820X goes up to 3.60 GHz (without turbo boost, which usually wouldn't do anything for me anyways since I plan to tie up all cores frequently). Basically, how far can I push it without complicating the cooling needs much? Even a tiny overclock, like to 3.8 or 3.7 would be enough for me.
In fact, I am not too hung up on overclocking at all. If the best answer for keeping cooling needs from getting too complicated would be stock speeds, then I could take that.
Part of this is my inexperience in knowing how much overclocking you can do before it starts pushing the limits of cooling solutions. So that's what I am asking; how much can you overclock while not needing to go crazy with cooling solutions given workloads with CPU and GPU at 100% simultaneously?

EDIT: If it makes any difference, I will have a software solution that will pause the workloads if temps start getting too high.


Edited by TechSY730 - 09 Jan 2018 at 7:42pm
Back to Top
DST4ME View Drop Down
DS ELITE
DS ELITE

Email address used to purchase matched with forums account email.

Joined: 14 Apr 2008
Online Status: Offline
Posts: 36758
  Quote DST4ME Quote  Post ReplyReply bullet Posted: 10 Jan 2018 at 12:11am
My comment was how can this not be a mining system, I didn't say "anything but a mining system" .

How far you can push oc depends on cooling, but your question is a bit weird, as soon as cpu runs at higher speed heat starts, how much heat depends on cpu itself, two 7820x can oc and have different temps, each chip is different, one can run 80c at 3.8 the other could run 80c at 4.3 lets say. So it is hard to give you a temp, it all depends on chip lottery.

Then there is the workload, some apps like the ones they use here to test don't push the cpu too hard, prime95 for example that we ocers use pushes the cpu way harder and thus the test temps with prime are going to be higher than these other apps they use here. Now if your workload mimic prime then your temps are going to be higher than your test temps here.

So it is hard to tell you temps.

Now 7820 has turbo boost at 4.0/4.3/4.5, which is over your 3.8, with h115i you could look at 85c possibly. Again too many factors.

But oc is the biggest performance gain you can get, you are down playing the difference between turbo boost and stock speed. It will make a huge difference in completing time of the workload.

If you know anyone with new cpu that runs Fah on cpu ask them to share completion time with same workload with and without oc, make the difference is as close to stock 3.5 and oc 4.3. Keep in mind that your cpu time difference is going to be even faster then your friends cpu since newer cpu = faster cpu.


Edited by DST4ME - 10 Jan 2018 at 12:14am
Back to Top
 Post Reply Post Reply

Forum Jump Forum Permissions View Drop Down



This page was generated in 0.046875 seconds.