[Hassoon]

Hello everyone,

Since this is my first post, a mini-introduction should be in order

I am a graduate student at San Diego State University. I have been studying Computer Science for nearly six years now, but I have only recently cultivated a keen interest in the field of Computer Graphics.

I am currently doing research in improving the algorithmic efficiency of techniques like Photon Mapping, Global Illumination and real-time Ray Tracing.

As such, I felt compelled to dig up my 2 year-old Mac Pro and start some coding. The machine still performs extremely well considering today's standards, but I have some cash I'd like to use to upgrade the machine, specifically the GPU and the RAM.

The machine currently has an nVidia GeForce 7300 GT. I'd like to upgrade to nVidia's Quadro 4800:

http://www.newegg.com/Product/Produc...82E16814133281

As for the RAM, it currently has a paltry 1 GB. I'd like to beef it up to 8 GB:

http://www.newegg.com/Product/Produc...82E16820146783

The machine currently runs on 2 Intel Dual-Core Xeons @ 2 Ghz, which isn't too bad. Hopefully no bottleneck issues.

My current budget is $2500, which is just enough for the GPU and the RAM. My goal of course is performance, but I have to keep the budget in mind.

I considered selling my current machine and upgrade to an 8-core Nehalem based Mac Pro, but I don't think the machine is currently worth anymore than $1000 (best estimate). $3500 will not be enough to cover a brand new Mac Pro, a top-of-the-line Quadro and 8 GBs of RAM.

I would really appreciate your input and/or opinions

Thanks guys,

- H

[MfA]

Wow, not your average grad student.

Just curious, are you planning to do this research on the GPU? If so wouldn't it be better to save some of the outrageous amount of money for that graphics card so you can build 1 or 2 server boxes to play around with with the next generation of cards? (In 6 months you can probably build something with 2 GB next gen cards from either AMD or NVIDIA for <500$.)

[Hassoon]

Originally Posted by MfA

Just curious, are you planning to do this research on the GPU?

Yes

Originally Posted by MfA

If so wouldn't it be better to save some of the outrageous amount of money for that graphics card so you can build 1 or 2 server boxes to play around with with the next generation of cards? (In 6 months you can probably build something with 2 GB next gen cards from either AMD or NVIDIA for <500$.)

You're right, however:

1- My thesis submission deadline for defense is July 2010. Waiting 6 months for next-gen cards is out of the question

2- I don't think nVidia is going to release its next high-end Quadro card for any less than $2500-$3500, let alone <$500. AMD does have cheaper alternatives, but not all are Mac Pro compatible. I'm also programming in C, so nVidia's CUDA SDK is something I'm looking forward to using.

[Tim Murray]

I thought the two-year-old Mac Pros didn't work with GT200 cards?

[MfA]

Originally Posted by Hassoon

2- I don't think nVidia is going to release its next high-end Quadro card for any less than $2500-$3500

Sorry, I'm having a hard time adjusting to your way of thinking. When I said server box I meant a linux or windows machine with commodity parts. I expect commodity HD5000/GT300 parts with 2 GB to be available for < 300$ in 6 months ... in non Mac land you can build a < 500$ server around that

In the mean time you can just get this:
http://www.newegg.com/Product/Produc...82E16814130500

PS. nevermind doesn't work ... as Tim said, the same goes for the Quadro. From the newegg page "MacPro2,1 are not compatible". Mac's are such wondrous machines ... I guess your best option is to build a non Mac for your research

[Hassoon]

Originally Posted by MfA

Sorry, I'm having a hard time adjusting to your way of thinking.

That's alright I failed to mention that the reason I'm leaning towards the Quadro because it's specifically designed for research/design purposes:

http://www.nvidia.com/object/product...or_mac_us.html

Whereas the mainstream GT 200/300 line is designed with high-performance gaming in mind.

That said, thank you SO much for bringing up the compatibility issue. I just found out from the link above that the Quadro card I wanted to purchase isn't even compatible with Mac Pro 1,1 (only Mac Pro 3,1 and above). You saved a TON of headache

Originally Posted by MfA

I said server box I meant a linux or windows machine with commodity parts. I expect commodity HD5000/GT300 parts with 2 GB to be available for < 300$ in 6 months ... in non Mac land you can build a < 500$ server around that

Now your idea makes perfect sense. Even though they're a fraction of the cost of a Quadro, I think one or two cards from the GT line may suffice for CUDA coding...I'll have to ask around this forum or nVidia's forum.

Originally Posted by MfA

are such wondrous machines ... I guess your best option is to build a non Mac for your research

Agreed...the whole compatibility issue was very disappointing.

The server option seems more exciting since you get to build the whole thing from scratch. Again, thanks

[Davros]

Go for a pc much better bang for buck

[Blazkowicz]

Nothing stops you from running CUDA code on a geforce, a Quadro is meant for expensive industrial engineering software, niches like geophysics (oil firms), etc; it lost much relevance as professional 3D packages support gaming cards (even using Direct3D), 3Dlabs is dead, etc.

Why want a $1600 8800GTX? It's outperformed and outfeatured by cards 10% the price.
Check on what nvidia has done : the GT200 isn't crippled for GPGPU, at first they wanted to (reserving DP support for the quadro and tesla lines).
If you want the bigger memory amount, there are GTX260 and GTX275 with 1792MB.

the GT200 GPU is also a bit better for CUDA (bigger register files and DP support)

if your Mac still hinders compatibility with a GTX 260/275 then sell it, get a core i7 920, the new Asrock X58 mobo (170$/€), 6GB ddr3 and a circa 500W PSU. You get a new GPU, big amount of ram, a CPU better than the two Xeon, all for about half the budget you wished to spend.

You can then add 6 more GB if you feel 6GB isn't enough (it's huge already), replace the CPU with a six-core variant during 2010 if you feel the need (cheaper alternative to the i7 xeon)

[ShaidarHaran]

Blaz:
Quadros are designed for 3d modeling and the like. Tesla is for the scientific apps.

[Freak'n Big Panda]

As others have said your best bet is probably to buy commidity components for a new PC. As blazkowicz said you can get a really nice core i7 system for waaay less than $2500.

And don't waste your money on a Quadro, completely uselss as they offer no added cuda functionality.

[digitalwanderer]

Can't you just mod a regular GF into a Quadro with some driver tom-foolery? Or have they changed that?

[Freak'n Big Panda]

not sure about the current cards but I remember back in the day you could do that with geforce 3s

[Hassoon]

I want to thank everyone for their responses and suggestions so far.

I went with MfA's suggestion and built a server box using commodity parts from scratch. Not only was the solution a ton cheaper, but I also had some left over cash that I used to get a great deal on a Quadro 5800

Originally Posted by Freak'n Big Panda

And don't waste your money on a Quadro, completely uselss as they offer no added cuda functionality.

In addition to my thesis work, we have two Teslas in the department that are being used to solve seemingly NP-hard mathematical formulas. While the Quadro doesn't offer the parallel computing capabilities of a Tesla, its GPU calculating capability will definitely come in handy.

I'm also a hobbyist Zbrush modeler

[Davros]

which part of "Don't buy a Quadro" didn't you understand