Yep, so as I said you should be able to find a screaming processor for your box

NewEgg has some combo deals that look promising.

Thanks for your input.

My $0.02.

AMD's got some 45W CPUs (Dual-core, I've got 2 of them currently) - both support virtualization, Intel's lower-end CPUs don't all support it. (They were making it one of the distinctive features, trying to get folks to buy the higher-end CPUs) Newegg has a search value for Virtualization Support. Last I'd looked, the 2.6GHz dual-core 45W part was about $60. AMD now has a 65W quad-core available for almost $200, rumor is they're working on a lower-clocked 45W quad-core sometime in the next 6-9 months.

RAM is cheap (give or take, $10/GB), get 2GB sticks. (Pair of 2, or a set of 4, your choice - you'll just have to go 64-bit to take advantage of more than 4GB)

Motherboard, if you're not looking for high-end graphics, on-board should be fine. AMD's Radeon HD3x00 series and Nvidia's 8x00 are both relatively capable, but Nvidia's drivers are still better under Linux. (AMD's working on getting ATI's drivers better...but it's taking longer than expected...) Intel's *work* but don't provide much power. (GPU is pretty much the one thing that's sometimes tricky under Linux, for standard components, if you care about 3D support)

Hard drive, these days 500GB drives are down to about $50. New drives are coming with 500GB platters - so multiples of that. 500GB, 1000GB, 1500GB, 2000GB. You can save $10 off the 500GB and get a 160GB, but I'd spend the $10 for newer tech and triple the capacity.

DVD burner, I personally just go with whatever Samsung's got on Newegg, usually about $27.

Power Supply, I'd look at 80% efficient or better (Silver, Bronze and Gold, each with slightly better efficiencies)

Random things:
Most of the on-board IDE controllers are NOT bootable anymore.
AMD's current socket (AM2+) can accept AM3 CPUs. AM3 sockets cannot accept AM2+, so you could get a current dual-core, and then upgrade to a quad when those come down in price and if/when you can use it. (I'm considering just that, for my gaming rig)
Intel's sockets aren't always so compatible, but I haven't built an Intel-based rig in awhile. Just what I've heard.
Your existing rig (if you buy a cheap case and move it over) could become a router, or always-on shell box, if the new rig wouldn't become that.

I do not, for the life of me, understand why anyone would make an on-board IDE controller that was not bootable.

I've already ordered the board, and went back and checked, and the IDE controller was not bootable. I'd planned on booting from my existing IDE disks, and put off moving to SATA disks until after my next paycheck.

So, I either :

1. Buy three SATA removable drive racks, four removable drive slides, and a USB external drive case, plus four SATA drives, and reconstruct my current setup in SATA.

2. Buy one SATA drive, mount it internally, copy just enough on it to boot from the IDE drives.

3. Buy a PCI IDE controller board that allows booting from the drives it controls.

I can't do #1 right now without dipping into savings, and if I wait until I can afford the extra expense I'll be out of the 30-day return guarantee from NewEgg.

Any recommendations for #3?

To save $0.05 on a controller or something like that. (Since "the majority of folks" build new boxes with newer hard drives, they'll save that nickel on all 10,000,000 boards they produce this year. Compared to the percentage of customers that care about IDE booting, it's worth the cost savings to them)

I've already ordered the board, and went back and checked, and the IDE controller was not bootable. I'd planned on booting from my existing IDE disks, and put off moving to SATA disks until after my next paycheck.[/quote]

Glad I mentioned it then.



Most new boards support booting off USB-based devices. Perhaps toss a small flash drive in, and put your bootloader on there? (Then have that point to your IDE drives. Bonus: secure the machine by taking said drive with you)
I know Microcenter has drives starting at like $6 or $7 for a 2GB Flash Drive. (At the checkout, their house-brand ones - not the fastest on the market, but they're very cheap and work well - could even store a spare OS for recovery stuff, if the primary goes)
EDIT: Could use Flash, SD card, whatever's handy and has a bootable adapter (not all SD adapters are bootable) - or find a second-hand SATA drive out of a laptop or desktop that's otherwise useless to someone.

What I have now is the ability to boot off of whichever drive I choose, simply by changing which drive bay is powered, and which drive is jumpered as master. I don't want to lose that. I used to spend quite a lot of time configuring boot loaders, swapping drives is a lot easier.

I plan on moving the whole setup to SATA eventually, so I could buy a single drive and not cinsider it a waste.

I'm guessing I could configure the SATA drive's MBR to boot the IDE master drive, but I've not done that sort of thing in a very long while. I see IDE controller cards selling for under $20.

Thought I'd give an update.

The new CPU and motherboard is up and running, mostly. I can boot Linux, connect to the net, access my external USB drive, etc.

Don't have the DVD installed, my old Windows XP disk crashes on boot, don't have sound working in Linux. But we're mostly there. I have a cryptography program that is typical of the sort of thing I've been waiting on. It runs three times as fast on the new hardware as did the original, and the multi-threaded version runs ten times as fast, since it can take advantage of the multi-cores.

Cool.. Congrats

While I'm not familiar with hillclimbers, I have executed some john runs (lookup john on sf.net) on my own data using openMPI, and I have used excess capacity on other machines on my home LAN to get something on the order of a 30x increase in processing. So, my question to you is this: can you parallelize your number crunching across multiple machines using openMPI? If so, then harness the power of the existing machines and don't upgrade your Sempron.

For example, my main workstation is an Intel Core i7 (quad core hyper threaded) 2.93 MHz, 12GB RAM Linux system - Linux sees and uses all that RAM plus 8 CPUs and 48,000 bogoMIPS of capacity. This machine is capable of testing about 1.2 million ciphers/sec of john tests. I throw into the mix all of my other linux-based workstations and servers and can test over 35 million c/s with the openMPI enabled application.

Bottom line: If this "need-for-speed" is limited to seldom-run processing, try putting together a compute cluster. If this "need-for-speed" is a frequent processing need, look into upgrading to Intel's or AMD's current top-of-the-line processor(s), but find the price-point you can afford.

There is a reason that the top supercomputers in the world run massively parallel clusters.... You, too, should be able to make use of them.

Caveat - I've also analyzed, designed, programmed, tested, and run many cryptanalysis trials, and I do know the value of compute-intensive processing.

If I was doing this for work, where there are dozens of machines to spread the load over, I might well have looked to a distributed package. I'm at home, and there are only two machines available - my desktop and an arthritic old laptop I bought used, five years ago, because I wanted to make sure I always had a machine with a real serial port. (For robotics experimentation).

In any case, I upgraded to a 3.0 GHz quad-core Phenom II.

I had a rather simplistic cryptarithm brute-forcer application that I decided to use as a benchmark app. A cryptarithm is a math problem - multiplication or long division or square or cube roots, where every digit is replaced by a letter.

My app generated every permutation of the ten letters in the puzzle, and tried each to see if it was correct. My first version used the University of Exeter recursive algorthm for generating perms. Running it against a set of nine cryptarithms, it would solve the bunch in 190 seconds.

The recursive algorithm wasn't easily partitionable into threads, so I switched to a permutation decoding method using an algorithm I found on Wikipedia's page on permutations. I'd loop through every integer between 0 and (n-1)!, and for each I'd generate the permutation and test. Then I split it into four loops, from 0 to x-1, x to 2*x-1. etc., where x = n!/4. Then I ran each loop in a separate thread. Total time for the set of nine cryptarithnms on my 1.2 GHz single-core Sempron: 230 seconds. Not a surprise. The threading added overhead, and there was still just one core.

So, on my new MB: the first, single-threaded program ran in 65 seconds - about three times faster than on the Sempron. But the multi-threaded program ran faster yet - 19 seconds, almost exactly ten times as fast as the original on the old processor.

Truth is, the old processor was fast enough, for base-10 cryptariithms. 20 seconds isn't a horribly long time to wait. But base-11 and base-12 cryptarithms were another story. On the old machine, base-12 problems would take most of an hour. On the new one they take just a couple of minutes.

I think you made a very good choice.