Once every decade build for R&D div. at my company

[college_student]

Hi folks,

The R&D branch at my company just got my $5000 budget approved to upgrade our ~10 year old system. We need a number cruncher that a few of us will SSH into and run a lot of video processing and analysis algorithms.

We deal with TBs of video data and I already have a 20 TB zpool of Samsung 850 Pros (10X 2TB disks), so these will not need to be bought.

CPU: Intel 5960X
Ram: G.SKILL Ripjaws V Series 128GB (8 x 16GB) DDR4-3000
Mobo: ASRock X99 Extreme11 LGA 2011-v3
GPU: 3-way SLI GTX 980 Ti
PSU: SeaSonic Platinum-1200(SS-1200XP3) 1200W
Cooler: Corsair H100i 77.0 CFM Liquid CPU Cooler
M.2 OS drive: SAMSUNG 950 PRO M.2 512GB
Peripherals: Logitech Wireless combo MX800
Monitors: TWO Dell Ultra HD 4k Monitor P2715Q
Case: Fractal Design Define XL R2 FD-CA-DEF-XL-R2-BL

Please rip this apart and give suggestions. Its not my money, but don't want to be wasteful. I considered going Xeon, but the extra cache on the 5960K is super useful for my application.

Thanks!

 

[ronbo613]

3 GTX 980Ti video cards? I am interested in what all this video processing power is needed for.

 

[frowertr]

I know you said R&D but it sounds like this is an important server for your company's livelihood. If so and its basically a production server, why would you want to whitebox this? You can get GPU servers from Supermicro or Dell. That way you get a warranty with support if something breaks.

 

[Insomniator]

Get a Dell.

 

[MustISO]

Get a Dell.

or HP.

This isn't something you should be building yourself. You want same day support and hardware replacement.

 

[nerp]

Huge mistake to build this yourself. For R&D and a company, get enterprise service.

 

[cbrunny]

Just curious, what is the existing system?

 

[Homerboy]

I know you said R&D but it sounds like this is an important server for your company's livelihood. If so and its basically a production server, why would you want to whitebox this? You can get GPU servers from Supermicro or Dell. That way you get a warranty with support if something breaks.

Get a Dell.

or HP.

This isn't something you should be building yourself. You want same day support and hardware replacement.

Huge mistake to build this yourself. For R&D and a company, get enterprise service.

All quoted for truth. This is NOT A DIY project.

 

[dave_the_nerd]

Agreed.

Although the system you should buy for this will be a lot more than $5k.

 

[frowertr]

Agreed.

Although the system you should buy for this will be a lot more than $5k.

+1
A 3-GPU server. Yowsa!

 

[college_student]

+1
A 3-GPU server. Yowsa!

Thanks everyone for responses and cautions.
General beuracracy at company prevents me from purchasing things from Dell with that large of a price tag. Without going into too many details, our dept. "speeds up" the buying process by having us build things like this (making several smaller purchases instead of large 'capital' ones). I know this might not be a satisfactory response to many of you, but I've learned to live with it over the years.

Now, we process really large (long?) video streams from multiple camera systems and typically we have three streams. Our current code base allows us to run one stream per GPU, and still takes like 10 hours to process. Thus three GPUs. We've thought about re-writing some of this to make it more efficient, but the one day turn around hasn't been a problem so far.

Also I learned that the haswell-E only takes 64GB of RAM, so I'll update that in my build. Also, the 5960X might be overkill, so maybe lower that to 5930K and cooled over air with Noctua.

Thanks.

 

[tynopik]

Also I learned that the haswell-E only takes 64GB of RAM, so I'll update that in my build.

I was wondering about that

Why not an E5-16xx Xeon?

If this is truly something you want to last for 10 years, start loading it with ECC LRDIMMs and you can expand up to 512GB if needed

http://www.newegg.com/Product/Product.aspx?Item=N82E16813182958
http://www.newegg.com/Product/Product.aspx?Item=9SIA24G3RH7940
http://www.newegg.com/Product/Product.aspx?Item=N82E16819117512

 

[mvbighead]

If you're limited to three GPUs in one build... why not 2-3 GPUs in two builds? That way, you have something of redundancy if one system fails. From the sounds of it, the GPU is most important, and if you can run 4-6 streams instead of 3, you can probably get more done, no?