Aerodynamic simulations, Total War Attila PC build $1200

[SteveGrabowski0]

Greeting Master Builder! 

 

Guided by Linus Tech Tips team I was led to this forum to present my build before buying components in shops. 

This is my very first build so I am very sorry for obvious mistakes I have made during the process, well you will be the judges.

 

Main Goal of this PC: Aerodynamics model simulation software and Total War Attila gaming.

 

I want to play on best possible settings for Total War Attila ( I know it is not very well optimized even for a high end components but I want to get as close I can to best settings) and Warhammer in the future. 

 

This will be my main work for engineering studies. - So soft like Catia, Matlab or Inventor will be use non-stop here. 

I would like to play also Planetary Annihilation - I am generaly strategy type player if that information is of any significance.   

 

 

Budget: $1200

 

​Components: (color theme: Black-Red)

 

 

CPU: i7 4790K $340 https://pcpartpicker.com/part/intel-cpu-bx80646i74790k

CPU cooling: Cooler Master Hyper 212 EVO $30 https://pcpartpicker.com/part/cooler-master-cpu-cooler-rr212e20pkr2

GPU: Gigabyte Radeon R9 390 Gaming G1 $340 https://pcpartpicker.com/part/gigabyte-video-card-gvr939g1gaming8gd

Motherboard: GIGABYTE GA-Z97X-SLI LGA 1150 Intel Z97 HDMI SATA 6Gb/s USB 3.0 ATX Intel Motherboard $120 https://pcpartpicker.com/part/gigabyte-motherboard-gaz97xsli

RAM: Kingston Hyperx Ddr3 2X 4Gb 1600Mhz Cl9 Xmp Savage (Hx316C9Srk2/8) $50 https://pcpartpicker.com/part/kingston-memory-hx316c9srk28

SSD: Samsung SSD Evo 850 250GB (MZ-75E250B/EU) $90 https://pcpartpicker.com/part/samsung-internal-hard-drive-mz75e250bam

HD:WD Caviar 1TB (WD10EZEX) $50 https://pcpartpicker.com/part/western-digital-internal-hard-drive-wd10ezex

Power supply: Corsair CS Series 650W Modular 80+ GOLD - CP-9020077-EU $100 http://www.amazon.co.uk/Corsair-CP-9020077-UK-Builder-Series-Semi-Modular/dp/B00GN8VZ7U

Case: Nzxt SOURCE 340 - USB 3.0 \ (CA-S340W-B1) $75 https://pcpartpicker.com/part/nzxt-case-cas340wb1

 

Total: $1195

 

And here things I would like to ask you about:

 

1) How would you score this build from 1-10, and how I can improve it to get 10/10 in the desired price range?

2) Is 650W enough for this build knowing that I wish to overclock as much as cooling system and other components lets me?

3) Is any of component a big overkill for this build or quite contrary I have cheapen out on something?

4a)And the biggest of them: I do not fully see big difference in GPU creators - I know I have some R9 390 for MSI and Sapphire and Gigabyte so I choose cheapest (with neat look) of them but I do not know If this is wise approach.

4b) Even when I choose developer of the GPU there are still version of it and again I am not sure how significat the differences are. Can you explain that to me? ( I tried to research some benchmark but I didn't find any )

Thank you so much for anybody who would like to aid/comment/help/rebuke me 

 

Don't get an R9 390. Get a GTX 970 instead since MATLAB supports CUDA.

[SteveGrabowski0]

I wonder if even a 780 or 780 Ti might make more sense. I haven't paid too much attention since I bought my 970 strictly for gaming, but supposedly Kepler has better performance on double precision arithmetic than Maxwell does. I'd imagine the CUDA could really help with all the fluid flow differential equations you'll be solving in MATLAB studying aerospace though.

[SteveGrabowski0]

@TheChris

 

For a purely gaming/recreational build honestly an i5 would sufficient and you might consider the i7 purely because strategy games tend to like threads.  I personally went with a X99 motherboard because of the render work I do and the fact the professional usage justified the expense.  The sooner tasks complete, the sooner I game.  I think that in terms of future usage the X99 option makes the most sense for anyone looking at a workstation that also can game right now.  You have the most headroom in terms of growing your processor if you need it (The 8 core 5960X and a number of Xeon options) out there, plenty of PCIe lanes, etc.  Professionally speaking there is no replacement for displacement. With my 5960X I can have a pair of Linux VMs up and doing stuff, a render job running on the host OS, and still have processing power to fart around on the web, watch videos, etc.  

 

A slightly weaker graphics card now will still deliver over 60 FPS in most circumstances and do just fine for at least a year, likely longer.  You can look at the high bandwidth memory next cards that should appearing near the end of 2016 and then decide when to upgrade the GPU. 

 

This is a great point. If there was ever a time to punt on GPU, it might be now with Pascal coming at 16 nm in a year. Hell, a $250 B-Stock refurbished EVGA GTX 970 might make sense here, or a used 780 off eBay.

[TheChris]

Okey, so shaken off a little by you guys informing me about CUDA technology I am facing my last dilemma:

 

R9 390 is doing better in almost all benchmarks, it has 8Gb of RAM vs 3,5Gb from GTX 970 but of course R9 doesn't support CUDA.

Question it:
1)What will be true and measurable differences with and without CUDA supporting GPU GTX 970 versus R9 390?

2)And if I should go with GTX 970 which power supply should I pick ( GTX 970 is much more power efficient. ) 
*Once again I would like to overclock my CPU and GPU?

3)Additionally: For my education - could my provide me with some benchmarks illustrating power/lack of power of CUDA technology?

 

[TheChris]

Hello again Master Builders!   
First components were ordered, and it is because of you guys. Thank you so much!   
 

-snip-

 

 

-snip- 

 

 

-snip-

 

 

-snip-

 

 

-snip-

 

 

-snip-

 

 

-snip-

Dear fellows could you aid me one last time with my final choice of the GPU by answering those questions:
 

1)What will be true and measurable differences with and without CUDA supporting GPU GTX 970 versus R9 390?

Okey, so shaken off a little by you guys informing me about CUDA technology I am facing my last dilemma:

R9 390 is doing better in almost all benchmarks, it has 8Gb of RAM vs 3,5Gb possessed by GTX 970 but of course R9 doesn't support CUDA.
 

2)And if I should go with GTX 970 which power supply should I pick ( GTX 970 is much more power efficient. )

*Once again I would like to overclock my CPU and GPU?

3)For my education - could my provide me with some benchmarks illustrating power/lack of power of CUDA technology?

+ additionally:

4th question: Is MSI Graphics Cards GTX 970 4GD5 OC  a good card? and how does it compare to Gigabyte GeForce GTX 970 OC (GV-N970WF3OC-4) and MSI GTX 970 GAMING 4G?

Thanks once again for you help!   

[Egad]

So for CUDA, it all depends on the application in question.  If the programmers did a good job, I'll see up to an 18% performance boost in run time with regard to the task.  For applications where the programmers have not optimized to CUDA, less.  This also applies in cases where the work being done simply can't be offloaded from your CPU onto your GPU (and then for that matter that your CPU can go do something else useful while the GPU does its thing).  This is a useful little report on CUDA vs OpenCL, you can just jump to page 8 if you want to see some graphs.  

 

I would also not say 18% is the ceiling, that is just the number based on the applications I use.  @SteveGrabowski0 might be able to expand on Matlab performance, etc.  Otherwise ask within your department to whoever specs the systems.  It might be worthwhile to ask around in your department and see how strongly people feel about CUDA being a must have for work.  Either ask the person who specs the system or the prof who has a giant workstation shoved under a desk.  If they went AMD and haven't felt any pain in their workflow, then CUDA isn't as worth it for you.  

 

For power supply, I'd hold in the 650 Watt range or even consider a 750 Watt one.  What I am going to guess is if you find yourself using modelling applications that can distribute work across multiple GPUs, you'll be sticking a second GPU in the case.  Maybe not right away, but once the price drops a little a second 970 might come along for the modeling and in game SLI.  So at least consider leaving yourself room to power it.  Matlab and a number of other applications have the ability to distribute the work across multiple GPUs and give you even better results.  If you think during your studies you'll be crunching big stuff, consider it.  970s will be getting cheaper at the end of 2016 when next gen is out and the 970 is now considered last gen.  I have a pair of 980Tis in my workstation.  When I work from home I distribute tasks over them and then I SLI them for games.  Works great.  If you see yourself considering this route, leave yourself headroom with your power supply.  

 

Also as Steve mentioned you could get a B-Stock EVGA at an attractive price.  When next gen comes you can stick in a better card, use that for gaming, and distribute workload across the GTX 970 and new card.  Modelling software is much less picky about matching cards than gaming.  

 

Asus, Gigabyte, MSI, and EVGA all tend to make good cards.  In your case I might consider the EVGA 970 blower cooler.  The general NVidia reference design, although the GTX 970 specifically lacked a reference design, is blower style and pushes its hot air out of the back of the case (right out the vents by the monitor ports).  Whereas the aftermarket designs such as the EVGA 970 ACX 2.0+ push air around inside the case and it is up to you to ensure your fans are venting it.  Typically speaking on cards intended to chug away on long tasks the reference design is favored.  You can't really overclock them much, if any, but they're also more immune to thermal throttling (where they get too hot and automatically reduce their clock).  You don't need to worry about hot air staying in the case since the reference cards just push it right out the back.  EVGA also has the best customer service rep of the bunch.  If you want the aftermarket coolers they're all good, personally I'm not a fan of the Asus Strix 970, I think the Strix cooler is the worst of the aftermarket designs, but of course that means worst of a good bunch, still good. I'd go EVGA SSC ACX 2.0+ myself (that means somewhat pre overclocked and with the ACX 2.0+ cooler on it).  You can then OC it some more.  

[hk29]

for your cuda question how just asking matlab. i dont think the IT guys can really answer that kind of question. are u a grad student or undergrad. i only seen heavy pc computations for grad students. like egad said find a prof doing the kind of work you are trying to build a machine for.