I have $7K I have to burn to on a 'scientific' computer. That is a computer for running simulations, doing AI shit, potentially running VR, and other scientific computing shit.
And by burn, I mean I have to spend it right now or it goes away. What gets me the most bang for my buck?
pic semi-related
4 Titan Z's
Quad processors. Xenos BTW
OP here, currently looking for a build with 32 cores, a decent GPU(none of us understand GPUs), and a fast as fuck SSD.
>>52448990
>ssd
>>52448891
You can't use 4 titan z's idiot.
>>52448990
dont forget to buy a 10 meg pipe
>>52448820
Don't scientists nowadays connect to clouds or data cenceters for processing power?
You really don't need anything more but a decent computer to do scientific work... even most software are resources efficient or connect to other servers.
Do you think you're going to simulate the universe in your computer? lol
>>52449122
>>Don't scientists nowadays connect to clouds or data cenceters for processing power?
Yes, but for stupid bureaucratic reasons we have to spend $7k on a computer or else it goes away. Our campus has a cluster, but for reasons I don't know, we can't use it.
Something, something VR might need extra low-latency or something which can't be done in the cloud.
>>52449207
Can't you spend like 4k and not mention where the other 3k goes?
>>52448820
Will you use CUDA at all?
Here you go buddy.
http://pcpartpicker.com/p/hhxkVn
>>52449279
>this fucking build
>>52449279
dem headphones lmao.
>>52449279
Kek 70k; does the motherboard even fit that many pci-e devices?
>>52449122
Hopefully not.
You generally have to do paper work* for using resources outside the department.
* You may have some small amount of "free" time, but this is often not enough for all the jobs you CBA doing paperwork for.
>Cuda
Generally not.
>>52449329
Yup
OP, if you are still here I have a legit build for you: pcpartpicker.com/p/bh8yjX
>290
For basic GPGPU usage. Can upgrade in the future if you need to.
Get a couple QUADRO cards or whatever Nvidia's computation cards are called nowadays.
At like $2000 a pop they should fill out a budget nicely.
>>52449279
OP here I said $7K not $70K. Kek though
>>52449362
That's pretty good. None of us know how to do GPGPU so that card is probably good.
>>52448990
You need lots of RAM for calculations. Or else you need to spend time coding so there is enough RAM. I tried using a old laptop just for a little data science. And it was the RAM that gave me trouble.
You might also look into using cloud computing like amazon for example
>>52448820
>And by burn, I mean I have to spend it right now or it goes away
give the credit card number back pyotr
>>52449487
no its not good you stupid motherfucker, nvidia has the scientific market cornered, so if you want sciency numbercrunching get the most expensive titan you can.
if you want vr get 2 titans, anything less won't do and you will be ridiculed by your colleagues when your shit gpu chokes on everything vr related you try like a young, overzealous faggot at a gangbang.
tl;dr GPU: nvidia only, 2 titans if possible.
>>52449954
>Titan
>not a tesla or quadro
It's a workstation, not a gayming shitbox.
>>52449985
>vr
it is gayming you retard, and titan is the only option to get middle groun between gayming and compute. 980ti would be a disaster in gpgpu scenario but great in gayman. So the titan line is the only option for op.
>>52450035
VR is only one of the things he's "potentially" doing though
>>52450042
vr is about to explode this year so they won't be able to resist. And since he said noone knows shit about gpus its more likely they won't get into really heavy numbercrunching where error correcting hardware like in teslas would matter, and for everything else the titan will do fine. With 2 and directx12/open sores equivalent (as if anyone would be retarded enough to attempt realistic rendering on linucks) they're getting future-proofed real good.
