special workstation build advice

[jkuebbing]

I have a special user case and was hoping for some advice

I realize this is incredibly long and I am sorry, I also apologize in advance I am a very bad writer with, spelling, grammar, and organization (dyslexia dose that) I am ok with correction if your ocd demandes it just don't be asholes about it . I am mechanical engineer student at UD looking to build a workstation hopefully with some outsider research funding. I have been interested in computers my hollow life and have been looking into stuff like this for a while i have built a couple of pc with friends including one water cooled system. I have been trying to find the perfect pc for CAD for two years by watching every thing i could and this odd user case for a year but the is very little info about this type of thing out there.

 

USER CASE

My user cases is so weird because this Workstation will be functioning as the brain for a couple of robotics projects. It will need to construct 3d images of the robots environment from multiple sources, including sonar, Ldar, and software i am working on that uses many normale 2d images from different perspective to build 3d images. It will also need to be able to run complex physics equations about interacting and moving about the world. last but not least this will need to function well for CAD and physics simulations as the robotics design is finalised, optimised, and the controlling software is designed. the robots will need to do the math fast on the fly and will learn and can make mistakes and will adjust in the future. But the rendered physics for simulations needs to be accurate. fell free to ask any other questions if it will help u answer.

 

This is where my questions begin.

 

1. CPU

   a.Intel vs amd (I know this has been a constant battle but for my use case specifically)

   b.i. If intel I think I want a xeon

   b.II. If AMD do i use a big 8 core, an APU, opteron, or opteron X2150 APU using crossfire to help?

 

2. amd vs.NVidia for GPU

   a. cuda vs openCL (mantle) (I have not begun to write the software. i did not want to get part of the way through only to find the other is better)

        a long time ago when I first started looking into this it looked like NVidia's cuda, sli ,quadro/tesla, physx, stole the show now i realise its more                     complicated than that and they seem to have fallen behind. 

   b. IF NVidia Quadro vs Tesla vs TITIAN

       I thought quadro and maybe a tesla but there has not be much improvement there in the last couple of years and NVidia came out at this years              technologie conference talking about using the titan z for rendering. I don't care about 10 bit color which from watching linus and tek syndicate is            the only benefit i am sure it still has, because the titans can do doable precision FLOPS as well.

   c. IF AMD S vs. V vs. W series

      I have not looked into this as much are the worth it?

   E. xeon Phi co processor and any AMD equivalent if aplicable

 

 

3. Motherboard

         I really need lots of PCIe as you will see as i continue, maybe m.2

   a.I. If intell i think i like the ASUS P9X79-E WS or  Z9PE-D8 WS, maybe GIGABYTE 7PESH3 SSI EEB

    a.II. if AMD i have no Idea and cant even seem to find anything comparable.

 

4. should I wait for X99 because of PCIe 4.0 and DDR 4.0, and will there be WS boards on that line?

 

5. Ram seems straight forward for this as much and as fast as possible.

 

6. STORAGE (the really complicated and user specific question) sorry for any rambling this is where i have many idaes and the answers seem to be the least straight forward.

 

Because of always needing to work with and have access to massive, 3d image, objects, protocols, as well as previous outcomes of actions physics and other things used by the robots and for their design I need some very fast storage and I need somewhere to archive things that are no longer relevant. So I was thinking maybe OCZ PCIe SSD, are they worth it? Or should i use M.2, Or hybrid, SAS, weird Raids(but those use the CPU)

 

quick storage to store searchable files that can help the robot find the relevant file. (thinking low quality 3d images of places its been to compare if it find one that matches bring up full quality compare if similar enough use and add the new data from sensors to to make even better then when no longer relevant send back away) (it would also need to remember how to move and what happened last time to help anticipate and correct errors in balance) (think of a powered exoskeleton that could "watch"/sense a person move then get used to how they move and help.) (or a self driving car that could build models anticipate whats ahead get used to how other drivers drive remember what happened in such and such situation last time and act accordingly.

 

these use made me think of a couple solutions the important pieces of which are PCIe SSD hybrid hdd in raid 0 which seams like it would make more of a difference since 3 or so good ssd in raid can saturate the bus but as ASUS demonstrated when they compared four and 5 devices in raid PCIe has much more bandwidth.

 

maybe even an idea i had while watching linuses review of the HAF Stacker was to  max out the main PC and have a second small pc in one of the other sections with a raid card and a NIC to run Raid 5 or maybe 60 without taking away form the main pc performance and still having some where to store all the data safely when its not relevant or being used.

 

If i use the ASUS boards i was thinking of my set up would look something like the following, this is what i came up with after months of using my free time to scour the web watching ever relevant linus, tek syndicate, o3cd, New egg, NVidia tech conference, tiger direct, AND talk, or any thing else i could find and was reliable, relevant, and i could bear to watch. So this is what i came up with please make comments or suggestions.

 

CPU

XEON

MotherBoard Z9PE-D8 WS

GPU already asked

RAM as much as possible and ass fast

storage

1 to 3 PCIe SSD

2 SSD in raid 0 on the native 6GB/s

the rest filled out with hybrid drives or slower larger SSD in the best raid arrays I can make.

Use the Native LAN to connect to a smaller PC

 

7. water cooling or no

 

smaller pc

maybe an CPU with on board graphic so it can do some less important or time critical work in the background

ssd cache

a NIC card

Raid card with huge raid 6 or 60 array

Lan card to go back to the main

also any internet connection would be moderated through this computer

 

this computer could also monitor the data used and the speed at which it can be accessed and make suggestions as to where to move data instead.

another uses would be to uses this to monitor temps and make adjustments so the big pc never has to "think" about it and if on top could even use its psu and fan controls to run them leaving more power for the cards and pc and less scaled power usage.

 

thinking of putting it all in a mountain mods case or the HAF stacker

 

8. cases suggestions

 

9. Do i have to get a special version of windows or are there gread cad  and server capabilities for linux it would be easier to program in i could cut out parts to make it less of a burden to run in the background and this is the one and only place i am wanting to make cuts for price.

 

 

 

 

 

 

 

 

 

[CoolBeans]

I can't answer your questions until you tell me your budget

[wng_kingsley7]

Looks like this might be a big one xDD! Might take a little bit of time @jkuebbing I'll re-edit this post to give you my thoughts & a spec list and just for reassurance I'm gonna follow the thread.

[Dredgy]

Asus Z9PE-D8 would be the motherboard I would use,and stick two Xeons in it. It only supports 64GB of a RAM so if you need more than that maybe look at other manufacturers. The particular board is kind of a hybrid between consumer level and full blown server board, which I think is an advantage - you may not.

For storage, a couple of SATA drives should be fine. I would be unorthodox and use a 240-500GB SSD as a scratch disk, a 1TB SSD as a boot/application drive and have at least 2 high capacity, high quality HDDs in a RAID 1/10 array for mass storage and archiving. Cache that array on another SSD if you want. At. The consumer level, the Samsung 850 Pro and SanDisk Extreme Pro are the only SSDs worth considering.

I wouldn't watercool, it's unnecessary and EEB cases probably won't be friendly to it.

I'd probably still is a Nvidia GPU solution, but what software your using depends on exactly what I'd get.

[jkuebbing]

also I was thinking water cooling  on the GPU simply to fit all of what i want. and use pcie fan blowers on the ssd to get some are out of the case and the rest blowing in as much as possible for positive pressure to to keep dust out i was also looking into High performance PSU that maybe even blow backwards on startup to keep dust out this is going to get a loot of use and i would like it to last a while.

 

also just want to check that you can have multiple different raided solutions and be able to manage and maybe even write software that dynamically manages what data goes were so i can write to my ssds less and get higher speed on what i need.

[Dredgy]

also I was thinking water cooling  on the GPU simply to fit all of what i want. and use pcie fan blowers on the ssd to get some are out of the case and the rest blowing in as much as possible for positive pressure to to keep dust out i was also looking into High performance PSU that maybe even blow backwards on startup to keep dust out this is going to get a loot of use and i would like it to last a while.

You can buy single slot workstation cards. Water cooling a card won't actually save you that much space. PCI-E risers are probably a better option - they're what I use.

For PSU, I don't know my around workstation ones, though I'm sure you can find ones that are dust proof.

 

also just want to check that you can have multiple different raided solutions and be able to manage and maybe even write software that dynamically manages what data goes were so i can write to my ssds less and get higher speed on what i need.

You can do that, seems like a lot of effort though, get a good SSD and you can write as much as you like to them. The ones I mentioned are high end consumer products and are warrantied for 10 years. SSDs, particularly good ones, can have huge amounts written to them.

[jkuebbing]

as to the question of budget i am applying for several large grants for  but for the biggest ones I Give them my budget once figuring out what i need I have to tell them what i need but after the big ones that are focused on the computer hardware. I would expect and like to be able to get the budget some were under $7,000, but could go up to 10,000

 

are there any other boards that support full four PCIe 3.0 X16  weather that be sli or not, and have more ram and I would not have to go to full server capabilities.

 

thanks for all your help I am great at physics and designing real stuff and a pretty good programer but really don't know much about this type of hard ware and am making the transition to actually thinking about parallel processing vs just html and java. I don't know if I will win the grants this year but if i ever do get the resources to build this i owe you guys a tour.

[jkuebbing]

as for storage I have been thinking about finding the perfect blend and way to store it all. I do need allot of storage in the order of 50 tb (not all up front i will be archiving lots data once its been processed) so i can just make one big array to do that and

I need some very fast stuff but i only need about 1TB of that 

 

any thought on the simpler just having 2 RevoDrive 3 X2 all the programs and OS and then the other for the robots cache

or similar performance raid solutions.

[brob]

@jkuebbing,

 

On the Intel vs AMD cpu debate I think it is fairly clear that Intel LGA2011 cpu offer the most computational power for the environment described.

 

Asus P9X79-E WS can run 4x16 or 2x16+5x8.

 

Go with x8 PCIe ssd like Intel SSDPEDPX800G301, Intel SSDPEDOX400G301, OCZ RVD350-FHPX28-960G or OCZ RVD350-FHPX28-480G.

[Dredgy]

Asus P9X79-E WS can run 4x16 or 2x16+5x8.

He will most likely need 2 CPUs,

Also, does your university not have an array for you to store your results, or will you actually need to buy 50TB of Storage? Because you're going to have enough trouble keeping this thing affordable as it is.

[brob]

He will most likely need 2 CPUs,

...

 

It seems to me that a distributed processing model is needed. Perhaps pre-processors for initial data acquisition and/or an analysis engine in addition to the data aggregator. Distributed models scale a little better than monolithic systems and allow for the asymmetric use/application of resources.

[jkuebbing]

I think i can calm down on storage I can figure it out an may not even need PCIe SSD Raid may give me a better solution and easier to odd load to and extra storage none usable like a separate nas

 

does anyone know about the other questions

 

to correct other things i said now that i double checked them I Need about 1 to 2 TB of fast storage and basicly as much accessible storage as possible without being too slow (i am thinking hybrids) then i can add storage capacity with a Nas or home server as necessary.

 

as far as budget I should revise and say that if you think its worth it i can go to 10,000 without any problem that leaves plenty of room for some random extra stuff to happen my real no mess top price is  close to 13,000, but i am only willing to spend that on something like a couple of teslas if it really will benefit from it. I just wanted to low ball it incase there was anything i was forgetting now after checking i am more sure about what i can spend on it.

 

last what i really am most confused about is the quadro vs. titian vs. tesla choice after this GPU tech conference talk by NVIdia on the titian z.

[jkuebbing]

you are correct i don't need 50 TB of storage now I just want to be able to scale to that this project is while i am still attending school but is my own intellectual property when i graduate this is stuff i plan on continuing to work on. 

[jkuebbing]

It seems to me that a distributed processing model is needed. Perhaps pre-processors for initial data acquisition and/or an analysis engine in addition to the data aggregator. Distributed models scale a little better than monolithic systems and allow for the asymmetric use/application of resources.

 what type of hard ware would i need for that? I was planing on only having a small processor in the robot that send all the data back to the big one. then the big one process all the data and informs the small one on how to act or if the small one has already been given an answer it just acts and the big one goes on trying to think of other problems build the models of the environment and anticipate what is up next. Is this what you are talking about. If not please explain as i said computer hardware is not my thing and im just ok at software but am learning fast.

[brob]

I think i can calm down on storage I can figure it out an may not even need PCIe SSD Raid may give me a better solution and easier to odd load to and extra storage none usable like a separate nas

 

does anyone know about the other questions

 

to correct other things i said now that i double checked them I Need about 1 to 2 TB of fast storage and basicly as much accessible storage as possible without being too slow (i am thinking hybrids) then i can add storage capacity with a Nas or home server as necessary.

 

as far as budget I should revise and say that if you think its worth it i can go to 10,000 without any problem that leaves plenty of room for some random extra stuff to happen my real no mess top price is  close to 13,000, but i am only willing to spend that on something like a couple of teslas if it really will benefit from it. I just wanted to low ball it incase there was anything i was forgetting now after checking i am more sure about what i can spend on it.

 

last what i really am most confused about is the quadro vs. titian vs. tesla choice after this GPU tech conference talk by NVIdia on the titian z.

 

Given the patterns described, hybrids are not going to seriously affect performance. I expect that maxing out memory on the motherboard and running 3-4 hdd in RAID 0 will do more.

[jkuebbing]

Given the patterns described, hybrids are not going to seriously affect performance. I expect that maxing out memory on the motherboard and running 3-4 hdd in RAID 0 will do more.

thanks thats help full

[jkuebbing]

@jkuebbing,

 

On the Intel vs AMD cpu debate I think it is fairly clear that Intel LGA2011 cpu offer the most computational power for the environment described.

 

Asus P9X79-E WS can run 4x16 or 2x16+5x8.

 

Go with x8 PCIe ssd like Intel SSDPEDPX800G301, Intel SSDPEDOX400G301, OCZ RVD350-FHPX28-960G or OCZ RVD350-FHPX28-480G.

thanks that is very help full the SSD suggested seem perfect and should be able to keep up with the rest of the build you guys are suggesting I think im going to go with less space and go with smaller ones offered and save the money for the GPU and CPU setup I will need. i can add a second if i need to latter once i get this all up and running which could be a while. That also allows me to go with a couple of hdd in raid  for the big storage.

[brob]

 what type of hard ware would i need for that? I was planing on only having a small processor in the robot that send all the data back to the big one. then the big one process all the data and informs the small one on how to act or if the small one has already been given an answer it just acts and the big one goes on trying to think of other problems build the models of the environment and anticipate what is up next. Is this what you are talking about. If not please explain as i said computer hardware is not my thing and im just ok at software but am learning fast.

 

Presumably you have a rough outline of the software system. Break the "big one" into different parts. One part builds the environment models. Another part does the analysis. And yet another part takes the analysis and determines what, if any direction needs to be passed back to the robot. One might even break up the model building into various parts, depending on what is actually being done. The tricky part with this approach is the communications between sub-systems.

 

While we always think that speed is measured in bps. That is not the case in a complex system. Transmitting information instead of raw data is much faster and requires much less bandwidth. So, for example, the robot sends IR data to the IR processing sub-system. That unit can simply compare the new data to what came before and pass on only key differences/changes. The model builder can adjust the model, etc. (Don't get hung up on how inaccurate my characterization is. The principal is the important part.)

 

These sub-systems can easily be individual pieces of software running on the same hardware. (Hopefully not simply parts of the same monolithic program.) Since the amount of actual data passing between the sub-systems is reasonably small, they could instead run on separate computers (or several sub-systems could run on a separate computer). This model is analogous to the human brain.

[jkuebbing]

what has been decided so far

CPU LGA2011 xeons

storage 

Intel 910 Series Ramsdale SSDPEDOX400G301

 

it has been brought to my attention that a raid set up will work for the rest Is it worth it to use a couple of hdd like the WD Velociraptor  with its work station specifications or should i go with more normal hdd at 7200 rpm where i can get 4x the storage for half the price and get twice as many to be able to put into raid 0 im thinking the second but tell me if im wrong. or maybe move the intel PCIe  to the slower OCZ and raid more Velociraptor that should be more durable?

 

Ram

with the motherboard i am thinking of i am limited to 64GB DDR3  example G.Skill F3-2133C10Q2-64GZM.

I was not thinking of server grade ram should i be?

 

Mother board

I'm still stuck with Z9PE-D8 WS I there anything else i should look at

GPU

with the most expensive set up we have discussed leave me with a budget of about 3,000 without going over 10,000 for GPU do i get 2 quadro 6000(PNY VCQ6000-PB) or 1 AMD W9100 or do i "break the bank" and get a quadro 6000 and tesla k20 or an AMD W9100 or S1000

 

As far a software i can use solid works or autodesk for the design optimization and for the robotics i can learn cuda or openCL

 

​any more suggestions you wonderful people i will look at it in the morning

[jkuebbing]

Presumably you have a rough outline of the software system. Break the "big one" into different parts. One part builds the environment models. Another part does the analysis. And yet another part takes the analysis and determines what, if any direction needs to be passed back to the robot. One might even break up the model building into various parts, depending on what is actually being done. The tricky part with this approach is the communications between sub-systems.

 

While we always think that speed is measured in bps. That is not the case in a complex system. Transmitting information instead of raw data is much faster and requires much less bandwidth. So, for example, the robot sends IR data to the IR processing sub-system. That unit can simply compare the new data to what came before and pass on only key differences/changes. The model builder can adjust the model, etc. (Don't get hung up on how inaccurate my characterization is. The principal is the important part.)

 

These sub-systems can easily be individual pieces of software running on the same hardware. (Hopefully not simply parts of the same monolithic program.) Since the amount of actual data passing between the sub-systems is reasonably small, they could instead run on separate computers (or several sub-systems could run on a separate computer). This model is analogous to the human brain.

I will have to look into this more you may have solved my problem this all sounds good any suggestions on the type of hard ware for any of those pieces or how they would share. THANK YOU

[jkuebbing]

MY answer has been broken up into two possibilities it looks like going @Beast solution will work best for me and I need to do some research into that type of thing to figure out enough to know the right questions to ask.

 

also looking at the one powerhouse choice  I realised i was looking at old fermi based GPUs for price pc part picker has not updated the ws options very much.

 

so really what i can get is 1 K6000 or W9100 I don't know much about AMD Graphics the W9100 is what i would want over the S10000 right

 

then go with the intel PCie and cheaper hdds or go with  velociraptors / equivalents and a small ssd  

 

that leaves more options for motherboards as i only need 2 pcie slots now (k6000 is a two slot card) and only want 4 (2X16 2X8)or so to leave room for expansion.

 

any suggestions on the GPU chosen the storage solution or the motherboard?

 

I am following this out more out of curiosity. I also what to have something to compare my "cluster" solution to and be able to tell the sponsor and or others what i saved how much faster it is and be able to compare it to something they would understand easier. Furthermore i would just like the info out there as i know i spent forever looking and I want to make it easier for anyone else looking.

 

Thanks again for all the help you guys have already provided especially @Beast.

[brob]

I will have to look into this more you may have solved my problem this all sounds good any suggestions on the type of hard ware for any of those pieces or how they would share. THANK YOU

 

Hardware will be highly dependent on the tasks that must be completed. Some pieces will need to be able to perform highly parallel calculations. Others may only need to be able to do memory manipulation - network/tree searches and the like. Still others will only be interested in communicating with or to the robotic device.

 

Communications is always the difficult part of this type of system. What occurs to me initially is to use something like tcp/ip sockets for interprocess communications. This allows for easy movement of processing nodes to various pieces of hardware. One might start with everything running on a single piece of hardware. As more functionality is added and knowledge gained, nodes can be spun off to specialist hardware without rewriting. Separate pieces would be in a dedicated high-speed network.

[jkuebbing]

This is an update and thank you to all those interested. I am kind of just bouncing ideas of of you at this point because its more sane than talking to myself about it all and it helps me move forward if I "spell it all out". There are a few questions buried in this mess but you dont have to go digging for them. they are in bold or underlined if you really want to find out what i'm working on now. (Software, robotics, then preprocessor) My only real questions are. Is gigabit lan the best networking solution for me now right? This can all be done with C++ right as the software is not that far along and I probably should start from scratch as the code that is there is a garbled messy of what used to be the single piece of software. Do I need ubuntu server or can i use workstation for now. I will probably be back to this infrequently and don't have to have answers right away as your ideas have already mad it so I can get much more work done with the hardware I have. But I might post asking more questions as i look int the hardware when i can't stand to look at code any more.

 

Let the rambling to me self out loud commence. If you are not interested stop following. Warning TLDR!

 

Thank you very much I was attempting to do the same thing using parallel computing on one computer having different cores work on different projects. I was difficult to program, ran slow, and crashed often. Taking your advice i copied the different processes and made different programs to do each. Collect the data, turn the data into useful info, then tack it step by step from there. Its still not any where near perfect and i will need to be able to do all of the things at once on the fly so i will need better hardware. But for now this idea has made it so i can go a lot farther with the coded which still has a long way to come with the hardware I have now. I also won't have to have so much storage at such ridiculous speeds, that I would need to be constantly switching back and forth, Its more than any one computer can hold in its ram but by spreading it out i just might be able to.

 

I am going to be working on the coded as well as the actual robotics that I love. The robotics are actually how I got into all of this and what my actual project for school is. I am just hear and working on this now for fun, because i enjoy it, to learn about the hardware, how to solve problems. before school starts and i have classes and the robotics to work on. Which i love i just want to have this of my self to compliment which is why i wanted my own hardware, well that and the aerospace engineers hoge the server with simulations and diagnostics of there own proprietary designs. The founding will still be there i have unique enough idea (i haven't and won't share what is special about my robotics, the type of models i'm building and how they will be analyzed.) Also spreading out and dealing it will be better with no up front huge backer i have a better chance of getting this all done without it being stolen. and I have all the time i am just a going into sophomore year alought i have significantly more credits, and this dose not have to be done till middle of senior year. I'm just trying to get ahead and get more. really i just have to have some thing to write up and present by then I can keep working on that even after but free time as an engineer i fell will be few and far between.

 

I will just leave a couple of questions that I stumbled upon while I was looking into this style/ Idea.

 

Once I do get farther along with this code I can build each piece one at a time, get it up and running and add more "power" as necessary. Meaning I can hold of on some of the money and better manage it between computer hardware and robotics.

 

I do need a pretty good work station. So that I can model, optimize, and run physics tests on all the parts then the full assembly before building any more than the mostly functional leg I have now. This peace can then become the backbone of the actual modeling process. Without  needing to carry out so many other specialized tasks it can be much simpler.

 

I am Thinking a normal LGA 2011 i7 a board and case with some room for growth a ssd for software and a hdd for storage (is a velocorapter 10000rpm drive worth it) or should i go with more storage keep in mind there will be a separate module to hold and store data securely latter.. Does that follow in the hardware world?

 

the frist other peace I was going to build was the pre-processor i was thinking for the beginning keep it cheap and use a AMD APU it will mostly just be receiving the data turning them into the useful information and doing some preliminary analysis hence the onboard graphics. Does that follow in the hardware world?

 

​next question is on transferring the data should I use the on board gigabit lan, go ahead and get 10gbit which is more than i need for now but may be called for much farther down the line, should I get a gigabit nic card, or should i get motherboards that have two lan slots that support teaming (i know it dose not double spreads but i would often be sending or receiving info from/to multiple sub processes. Finally most of the process can be broken down but a good deal of it is sequential. Pictures are brought in analysed next a model is built. then the model is analyzed. Several of these analyses  can be broken down and worked on in parts and in different ways. Which is why this is an improvement. But i am never sending or receiving data from more then two system. Would I still need a switch? or could i add that later as i need to send data to more places. I there any other way to transfer data that i should be looking into.

 

last I have a couple of weird little questions i came up with while looking into things.

When I went looking for motherboards with two lan connection I found a lot more with old 1150 sockets and none with amd processor compatibility was i looking in the wrong places, and general thoughts on cpu not of the 2011 verity, I also found a asrock board that "natively" supported a lot of SAS of the 1150 variety.

Do you have any suggestions for a small form factor motherboard or hole computer with a little bit of pci capability and would have high throughput of data for lan.

Thoughts on using apus in the little machines for doing less intensive work, just to free up some of the memory in larger more complicated systems.

[brob]

As far as I am aware the only real difference between Ubuntu desktop & server are the various addons. So you can start with a desktop installation and add server parts as/if needed or vice versa.

 

I would choose a software development language/platform that was readily available on all the hardware I anticipate using. Managing different languages needlessly complicates things. C++ is fairly universal, but one has to be careful of local dialects.

 

Gigabit LAN is very common. It is not a bad place to start. On can usually add a NIC if something faster is needed.

 

I am not a real fan of Velociraptors. But they are fast for hdd. I would take a hard look at the amount of fast storage that is required. Anything less than 900GB and I would look seriously at an ssd. Anything that does not require fast storage can be stored on 7200 rpm hdd. One might even have a task that "aged" data, moving it from fast storage to longer term, regular storage.

 

Presumably you want to use an iGPU for parallel computations. APU have a stronger iGPU than Intel, but weaker CPU. I suggest you try to dig up some benchmarks of work similar to what you anticipate to see if there are any significant differences. When you get down to it the system cost differential is marginal at best.

 

The RAID support of the Asrock Z87 Extreme 11/ac is rather limited. You might be better off with a separate RAID card. With a RAID adapter one would have the added advantage of being able to move it to new/different hardware.

 

Platform is a difficult question. LGA1150 has a good range of processors & motherboards to choose from. LGA2011 has more powerful cpu. AM3/AM3+ has a good variety of cpu but motherboard selection is rather limited. LGA2011 or AM3+ are good choices only if a system is going to be running a large number of concurrent tasks/processes. (Which is better will depend on the work being done, but likely Intel will be the preferred choice.) Otherwise LGA1150 makes the most sense as it has fast cpu and excellent motherboard selection.

[jkuebbing]

I was only looking at fast hdd because of the amount i will writing to them i will be aging the data in most of the systems and the pre-processor will not be holding any thing for very long esenaly i need a scratch disk.

 

I will do some more looking at the apu's to see if its the right choice.

 

thanks for the answer as to platform. there is piece that will be limited to one cpu based task and i needed the larger motherboard selection.

 

C++ was chosen because it is the language that i am most familiar with that was supported by my hardware directly and form some research it looked like java would have been a poor choice. 

 

Thats all the questions i have for now untill i get more work done with this software thanks for all your help.