Desktop for Work

[Cadelin]

I have been informed that I have £1121 in my expenses account that needs to be spent before 31st December or I will lose it! I have decided to get a new desktop computer for my office. I will be claiming back some travel expenses so I will probably have about £750 to spend. I need to get a monitor and mouse + keyboard too. I will be running Scientific Linux on it.

My work involves large numbers of complex mathematical calculations. My key requirements are:
1) Reliability
2) Storage space (I can easily use 100 GB per day)
3) Processing Power + RAM

Any suggestions as to what I should be looking for? I also need to get this in Switzerland so if there are any companies that deliver internationally and are good that would be great. Thanks for any advice in advance!

 

[nath]

Is building it yourself out of the question? For £750 you could put together a beast of a system for processing and storage/RAM given that (I assume) you've no interest in a potent graphics card.

 

[TdC]

is the program you're calculating with multi-threaded? would you benefit from a quad core cpu? (or two)

 

[Cadelin]

Building it myself is not out of the question but I am not an expert so I would prefer to buy it unless there really is no company that ships these kind of things.

is the program you're calculating with multi-threaded? would you benefit from a quad core cpu? (or two)

Yes the calculations can be split. We have some computer farms and for large calculations I normally split my calculations across around 100 computers and then get the results sent back to me. This is why storage is the most important thing for me. However for smaller calculations and testing things you do need your own computer and the more cores the better. Most of the computers we use currently are dual cores.

 

[TdC]

my current workstation is a Q6600 @ 3GHz with 8GB ram. it has 4 750GB disks in RAID 5 as storage. it's gfx card is a gforce 8800GTS, though I am guessing you will require something like a Quadro or one of the bigger Matroxes for your purposes.

you can build my workstation for a less than 750squids atm, and spend the difference on beer.

 

[Cadelin]

my current workstation is a Q6600 @ 3GHz with 8GB ram. it has 4 750GB disks in RAID 5 as storage. it's gfx card is a gforce 8800GTS, though I am guessing you will require something like a Quadro or one of the bigger Matroxes for your purposes.

you can build my workstation for a less than 750squids atm, and spend the difference on beer.

There were too many words in this post that I have heard off but don't really understand. I am not a computer expert. I do mathematical calculations that require a computer. I just want the computer to be there and work. I don't think I need a graphics card and I don't know what the best way to configure multiple harddisks is.

I had been hoping I could get something like:
21" Monitor or there abouts
Some sort of Quad core processor.
8 GB RAM
2TB+ Harddisk space.
A graphics card that is good enough to support the monitor.

 

[Jupitus]

I'm just fascinated to know what your work is !!

 

[Cadelin]

I'm just fascinated to know what your work is !!

I work here!

I am working on B-Physics

I also help align the Inner Detector

 

[Ch3tan]

Freeman??

 

[Jupitus]

I work here!

I am working on B-Physics

I also help align the Inner Detector

Ooooo nice one - that's kinda cooler than cool

 

[inactionman]

If you do complex mathematical calculations (I assume you are given what you have said about your work) you probably want a decent graphics card, say a 4850, as it can be used for all double-precision fp, and it will cope with it much better than the processor. Although it may not be that simple to actually do this depending on what software you are using.

 

[old.user4556]

Freeman??

my thoughts

 

[Chilly]

If you do complex mathematical calculations (I assume you are given what you have said about your work) you probably want a decent graphics card, say a 4850, as it can be used for all double-precision fp, and it will cope with it much better than the processor. Although it may not be that simple to actually do this depending on what software you are using.

thats not a bad idea you know. the grunt/$ ratio you get in graphics cards is far better than general purpose CPUs. It does require coding to specific libraries etc but it's doable with some effort + learnings.

Failing that I'd just go for a big lump of cheap ram, a mid range i7 quad core cpu and some cheapo disks. Dell will prob have something at good value. I cant seem to get onto their site atm to have a look tho, sorry.

 

[Cadelin]

If you do complex mathematical calculations (I assume you are given what you have said about your work) you probably want a decent graphics card, say a 4850, as it can be used for all double-precision fp, and it will cope with it much better than the processor. Although it may not be that simple to actually do this depending on what software you are using.

Most of the important code is written in Fortran, although slowly some things are being moved into C++. The framework we use is C++ with python scripting but most of the functions are still fortran functions with wrappers. You can read about it here if you want!

I am pretty sure that the software is not optimized to use any graphics cards and I don't see much point in getting anything other than a low spec one to support duel monitors if I wanted.

As I said the majority of the calculations are done on farms of computers, and then the results get sent to me. I will only be using my own computer really to compile/test new software.

 

[TdC]

There were too many words in this post that I have heard off but don't really understand. I am not a computer expert. I do mathematical calculations that require a computer. I just want the computer to be there and work. I don't think I need a graphics card and I don't know what the best way to configure multiple harddisks is.

I had been hoping I could get something like:
21" Monitor or there abouts
Some sort of Quad core processor.
8 GB RAM
2TB+ Harddisk space.
A graphics card that is good enough to support the monitor.

I'd have thought CERN would have some nice workstation all set up for you? also, using local storage when you could have EMC or something like that makes me think I should apply for a job there

Anyways,

• Acer TFT Monitor P241WD 24", DVI 1920 x 1200 pixels 400 candles brightness, 3000:1 contrast 271e
• Matrox M9120 512MB DDR2 PCIe x16 2xDVI-I (or any card that linux supports and can drive two or more monitors) 232e
• Intel Core 2 Quad Q6600 socket 775 (get a boxed one so you have the cooler intel supplies) 168e
• Gigabyte GA-EP45-DS3R, p45 RAID on board, firewire, sata, 2x gbit ethernet, sound (linux should support but you're best to check) 114e
• Corsair TWIN2X4096-6400C4DHX 4GB (2x 2GB) DDR2 57e (x2=114e)
• Samsung Spinpoint F1 HD103UJ, 1TB 90e (x4=270e)
• Samsung SH-S202N DVD/RW 20e
• Cooler Master Centurion 590 62e
• Zalman ZM600-HP 600W, nice and quiet. 90e

total=1341e


just to give you some indication ofc. some people here are a lot more clued in to hardware than I am, they could prolly dream up a better system, or a cheaper one get some tech to put it together, and you may have to recompile your kernel to optimise your performance. the P45 raid can boot large volumes, or you can use linux built in software raid to put your disks into some useful configuration. your tame techs should know about that too.

 

[Cadelin]

I'd have thought CERN would have some nice workstation all set up for you? also, using local storage when you could have EMC or something like that makes me think I should apply for a job there

You can buy computers from CERN stores. There are a couple of cheap desktops avaliable (Dual Core, 2 GB RAM, 250 GB HD etc). CERN stores is not designed for this kind of purchase. If I want a nice computer I can go and buy it in a shop. If I want a radiation hardened piece of electronics that will function reliably at -270C for 10+ years that what CERN stores was designed for.

The computing project that is going on at CERN is almost as impressive as the experiments. The ATLAS detector will produce 300 MB of raw data per second while it is running and we expect to have it running for around 1/3 of the year, so that's around 10 000 000 seconds. That means around 3 PB of data will be produced per year. Now this obviously needs to duplicated and other datasets produced from it that are easier to work with so it is expected that around 10-15 PB of storage space will be needed a year. We are aiming to get around 150 000 CPU in farms around the world to process all this data.

So no matter how powerful this desktop is it will be utterly insignificant compared to that. Most of the work I do will have to be done and stored remotely. However having a few TB of space available will allow me to store some stuff on my local computer so I don't have to always work remotely and this will vastly speed up my software development.

Anyway thanks for the suggestions, its given me alot of ideas. I am still more than willing to listen to ideas, but I will probably have to get the request form in by Monday/Tuesday next week to make sure my expenses account doesn't empty itself!

 

[TdC]

Certainly, my pleasure. I am aware CERN is growing the storage rigs btw, just for a laugh they will prolly surpass the company I currently work for, which is iirc currently the top user in Europe. My own requirements are much more in the OLTP field so we don't particularly have vast amounts of new data coming in at that speed.
I am fascinated more by your grid setups mind, we are implementing some of our own, but ar currently being conservative at something like 20k compute nodes for our risk analysis stuffs.

please be aware that the desktop I specced you is certainly not cutting edge, well the disks are, more or less but not the rest. You can certainly get a faster cpu, etc, but for a workstation this is (imho) a pretty decent one. with some bargins and tweaking of parts you can probably get it to bounce on the 1000 euro line.

 

[Cadelin]

I am fascinated more by your grid setups mind, we are implementing some of our own, but ar currently being conservative at something like 20k compute nodes for our risk analysis stuffs.

LOL! It would be fascinating if it worked at all well.

The best way to do it:
Build a couple of aircraft hanger sized buildings near CERN and stick 100 000 CPUs in them.

The politically approved way to do it:
Build a grid of small clusters of computers all around the world and have beautiful code fairies transport data and knowledge around the world!

Still we have 6+ months for the computer people to sort that.

 

[TdC]

beautiful code fairies my arse too bad you and I know that all is not as it seems in the magic land of data transportation o0

and six months is very short in project time, I hope someone was paying attention at the meeting that told people to get their arses in gear lol

ah well, at least you'll have your nice new workstation before then hehe!

 

[Zenith.UK]

Most of the important code is written in Fortran, although slowly some things are being moved into C++. The framework we use is C++ with python scripting but most of the functions are still fortran functions with wrappers. You can read about it here if you want!

I am pretty sure that the software is not optimized to use any graphics cards and I don't see much point in getting anything other than a low spec one to support duel monitors if I wanted.

As I said the majority of the calculations are done on farms of computers, and then the results get sent to me. I will only be using my own computer really to compile/test new software.

I think what inactionman is trying to say is that a decent graphics card can be used as a co-processor for floating point calculations that is orders of magnitudes faster than a CPU. Nvidia have been pushing alternative uses of their graphic cards for the last couple of years as all a graphic card does is process floating point maths VERY quickly.

A quick googling found that the programming language for Nvidia graphic cards is CUDA.
CUDA Zone -- The resource for CUDA developers
As you can see from that page, it has been used in a number of maths intensive applications.

 

[Cadelin]

I think what inactionman is trying to say is that a decent graphics card can be used as a co-processor for floating point calculations that is orders of magnitudes faster than a CPU. Nvidia have been pushing alternative uses of their graphic cards for the last couple of years as all a graphic card does is process floating point maths VERY quickly.

I am not doubting that this is possible, the problem is, is it possible for me to use easily. If it was simply a case of adding a graphics card and noticing that the time it takes code to compile and run drops hugely then I would be all for it. I am however fairly sure that this won't be the case and even if it is possible I don't have the time or the inclination to get it to work.

 

[Chilly]

not to mention the compute clusters wont have the same hardware so its pretty much a waste of time.

 

[Bahumat]

What about giving me the money to go find you one of those autistic number genius people?

yeah yeah I know that's close to the edge but meh

 

[gunner440]

What about giving me the money to go find you one of those autistic number genius people?

yeah yeah I know that's close to the edge but meh

But what about the 1000's that will be spent on nappies when they keep shitting themselves?