Speccing out a number-cruncher for microarray analysis

C

ceolstan

Weaksauce
Joined
Sep 19, 2004
Messages
72
My kid sister is in the data-analysis stage of her PhD research in molecular biology. Her research involves, among other things, analyzing 80 microarray samples. I don't have the kind of knowledge that it takes to appreciate the sheer amount of data, but apparently these can easily have about 30000 lists associated with them. The software is TIGR, an open source suite of programs designed for microarray. At any rate, attempts at running the analysis brings her 4-year-old P4 machine to a crashing halt. The machines in the lab are also brought to their knees. The sponsor of her research has given her the go-ahead to spec out a machine that can do this analysis.

Here's what I've been thinking for more bleeding edge:

Motherboard: Gigabyte GA-X38T-DQ6
CPU: Intel Core 2 Extreme QX9650
RAM: Corsair 2x2GB 240-Pin DDR3 SDRAM
Case: Antec Nine Hundred
PSU: PC Power & Cooling Silencer 750W
HDD: Western Digital 750G HDD
Optical Drive: ASUS 20x DVD Drive
Graphics: EVGA 8800GT
Vista 64-bit system (not decided on flavor--Home Premium?)

This is probably the starting point, with the probability of going lower. The hardware support for the 45 nm technology isn't yet mature, and DDR3 is quite pricey. One of the P35 boards with DDR2 could probably do her a good job, too.

This is just the high end. The first criterion is number-crunching muscle. The new cpu + the faster fsb should give her the kind of beef she needs to get through the large amount of data associated with this kind of research. The amount of RAM is also there to help with the number crunching demands.

I am considering that perhaps a RAID array would be the smart solution for guarding against failures. I've specced for a single harddrive, but perhaps 2 drives of 500G would be better.

I want a really good power supply. I know the newer processors and graphics cards have lower consumption, but since the data will be important, I thought it likewise important to have a solid psu driving the system.

The case seems as if it has good cooling. I'm somewhat concerned about heat if the processor will be under load.

I've not included aftermarket cooling. This system will not be overclocked, but I would definitely want to think about some aftermarket cooling.

Graphics aren't a crucial part of this system, but since it'll be running Vista, it makes sense to have a DX10 card included. I haven't put in the monitor, but she says a tilt/swivel monitor is iimportant and I thnk a24-inch one would be great.

Any other comments or suggestions on ways either to improve the system or to offer a lower-budget version that would get the job done?

Thanks!

--ceolstan
 
So you need high processing power (dual quad-core xeon?), tons of ram (8GB+), fast HDDs (RAID5 SAS array, since its database driven), etc. Vista isn't an option, since TIGR doesnt run on it (according to a quick google search). I still can't tell if the 8800GT is needed or not. If there's any 3D rendering being done, a good graphics card like a FireGL or Quadro would be preferred, if it was using OpenGL. Though, it looks like its only 2D images, so a high powered vidcard may not be needed at all.

Unless this is a personal system, go with Dell or HP -- or whatever OEM the institution gets a good discount at.

Whats the budget?
 
enginurd said:
If there's any 3D rendering being done, a good graphics card like a FireGL or Quadro would be preferred, if it was using OpenGL. Though, it looks like its only 2D images, so a high powered vidcard may not be needed at all.
Click to expand...

That brings up a good point. How is the data visualized? Or does this system even need to go visualization at all? A midrange or value-range card might be more than enough.
 
I think a midrange video card would be fine, too. I just specced out higher end.

In fact, I think the real answer would be to throw this kind of analysis at a computing cluster, and I'm not sure why that hasn't been investigated. Budget is $7K, but that's because the sponsor is coming from the project on the medical side of things and hasn't a clue about the details on the molecular side of things. And no one really knows about the computer demands.

I do think the more cores, the better, but my concern has already been raised: if the software can't take advantage of the multiple cores, then more won't necessarily be better.

I do think the question about asking TIGR what kind of hardware they recommend might be the ticket. I've poked a bit for bioinformatic systems, and the last one I saw had just a vanilla Intel dual core processor. It came with a hefty price tag, though--about $5K US, which seemed to be a bit much for what the system included.

Thanks a bunch. I'll go poke around a bit and report back more as I find out more.

--ceolstan
 
I managed to corner my sister at a time when she wasn't dealing with her toddler.

After poking about more on the TM4 site, I suggested she do the following: 1) check with the IT support people at both her university and the medical center (primary funding source) to see if she can have access to the unix clusters, and 2) contact TM4 directly to see what they believe is optimal.

Initially, I was given the impression that this was going to be a standalone desktop for use at home--hence the desktop components. However, it seems as if it will be housed at a hospital, so a xeon server running linux/unix may be the best bet, as the machine will be available for other research-based projects. Presumably, she could get remote access, which means that she'd not have to have anything spiffy at home or in the office.

I've asked her to ascertain what kind of imaging will take place. If they're going to be doing 2D only, then she won't need much in the way of graphics. If there is any 3D imaging, then she'll likely need a decent mid-range card. The best kind of description that I can get from her is that their images are like those in Scientific American or National Geographic.

I'd like to thank folks for their patience. I've learned a lot of the right questions to ask, and have concluded that she needs to do some additional research in order to learn the vocabulary she needs to know in order to talk to her IT people. It looks less and less like a DIY project (though I'd love to be able to put together a really high-end machine!) and more and more like something that can best be handled by talking to IT support and to OEM server vendors.

--ceolstan
 
Good job. Keep us updated.

If you were building a standalone desktop system, you were on a good track, but I still would have gone the xeon route.
 
You must log in or register to reply here.
Back
Top