• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Is Bulldozer good for 8 thread genetic algorithm problems?

H

hp79

n00b
Joined
Oct 5, 2011
Messages
6
I'm wondering if AMD's Bulldozer is good for genetic algorithm / evolutionary computation problems. (ParadisEO)

I already bought a i7-2600K, and it takes about 50 hours to solve 4 thread problem, and about 80 hours if I run 8 thread with hyper threading. I never thought that HT would really work this well.

The program I am running only uses minimal memory (less than 1GB for 8 threads), and minimal disk activity. I have 8GB in the system, and a SSD which is probably a waste, but RAM was dirt cheap, and I had a SSD laying around.

It's the raw computing power I need, but does that relate to floating point calculation? I don't know what kind of benchmarks I should be looking at. Definitely not gaming benchmarks.

I'm going to buy a Z68 motherboard tonight because now I really want to overclock to 4.5GHz. I currently have it on a H61 board.
 
you may want to look into the 6200 series opterons if thats what you want to use the processor for. running a single processor or 2p G34 setup you should be able to do 3-4 times the amount of data flow.

btw why would you waste money on a z68 motherboard just to overclock? the p67 boards overclock just as good. the only benefit the z68 really has is that stupid msata iSSD or what ever the hell its called.


if you have a way to allow some one that has a bulldozer system already setup to do a time test with what you are trying to do it might save you some save you money or tell you its you need to switch. i'm sure you can probably get a couple people here to help you out. if anything you could try talking to tobit or jebo in the Distributed computing forum. they may be willing to help test it for you since they both are playing around with 8120's right now on F@H.
 
No money for LGA2011 :(

AMD's Bulldozer looked very poor from all of the review sites I can find. I thought maybe it is useful for other purpose not shown in benchmarks.

I don't know what kind of computation power from the CPU is the important factor for this purpose. I read some papers that CUDA gives really good speedups (up to 20), but I'm a noob in C/C++ and don't want to get my hands on reprogramming the whole thing.

I bought a Biostar z68 board because I don't want to add an extra graphics card on it. That's the only reason for it. The graphics card will suck up at least 10W just sitting idle, and since I'm keeping it powered on for a long time whenever it's on, I want to save some power too. The H61 board system was using 100W on full load, I'm expecting a little higher (around 120) when I overclock to 4.5GHz.

The z68 board isn't really expensive compared to P67 boards if you can find the right deal. I paid $105 for the biostar TZ68A+ at Frys which has four dimm, two x16 pci-e slots, 2 pci, 1 x1 pci-e, usb 3.0, uefi bios. When not using for computation, I'll use it as a HTPC since I can use the HDMI and sandy bridge cpus have a very low idle power (20-25W). I'm waiting for the new cpu cooler to come this afternoon. Then I'll rebuild the system.

So not going into server cpus such as Opteron or Xeon, what would be the better consumer cpu for F@H? I think F@H might require similar computation as evolutionary algorithm stuff.
 
If you are concerned about power usage and will be overclocking BD is likely not for you. There is no on die GPU, you will need to buy one which is only going to add to the wattage.
 
hp79 said:
No money for LGA2011 :(

AMD's Bulldozer looked very poor from all of the review sites I can find. I thought maybe it is useful for other purpose not shown in benchmarks.

I don't know what kind of computation power from the CPU is the important factor for this purpose. I read some papers that CUDA gives really good speedups (up to 20), but I'm a noob in C/C++ and don't want to get my hands on reprogramming the whole thing.

I bought a Biostar z68 board because I don't want to add an extra graphics card on it. That's the only reason for it. The graphics card will suck up at least 10W just sitting idle, and since I'm keeping it powered on for a long time whenever it's on, I want to save some power too. The H61 board system was using 100W on full load, I'm expecting a little higher (around 120) when I overclock to 4.5GHz.

The z68 board isn't really expensive compared to P67 boards if you can find the right deal. I paid $105 for the biostar TZ68A+ at Frys which has four dimm, two x16 pci-e slots, 2 pci, 1 x1 pci-e, usb 3.0, uefi bios. When not using for computation, I'll use it as a HTPC since I can use the HDMI and sandy bridge cpus have a very low idle power (20-25W). I'm waiting for the new cpu cooler to come this afternoon. Then I'll rebuild the system.

So not going into server cpus such as Opteron or Xeon, what would be the better consumer cpu for F@H? I think F@H might require similar computation as evolutionary algorithm stuff.
Click to expand...

ParadisEO is just a framework...what problem exactly are you trying to solve?
 
bwang said:
ParadisEO is just a framework...what problem exactly are you trying to solve?
Click to expand...

Network connection (node and wires) optimization problems.
Once I give the parameters, it runs for 10000 iterations or more. I saw each iteration takes about 10-30 seconds. I run 4 threads, or sometimes 8 if I can wait for a little longer. Hyperthreading really helps when running 8 thread. I wasn't expecting much from HT because experiences from Pentium4 day, but when I tested it out, it actually works well, provides at least 30% performance boost. Instead of 4 jobs for 2 days two times, I can run 8 jobs and finish it in little over 3 days.

I'm just wondering how efficient or powerful AMD's Buldozer is for this kind of work. There's no way I'll switch to AMD though. My last AMD was Athlon II X3 435, and I didn't like the gaming performance in HL2, also energy consumption looked bad compared to a Celeron e3300.
 
I was involved with computational bioengineering a short time ago. I'd advise you to seek out any GPU-based solution you can find and move your work over to that platform. You'll see literally an order of magnitude in improvement.

I'm assuming your task is highly-parallelizable (it is, right?).
 
If your not wanting to switch then whats the point? curiosity? Anyhow, even in properly threaded applications, BD 8-core architecture still proved severely inefficient and trailed the Sandbridge i7. I agree with your idea to OC and stick with the chip ya got now, its a hell of a chip and aside from moving to a dually rig, id stick with it.
 
wabbitseason said:
I was involved with computational bioengineering a short time ago. I'd advise you to seek out any GPU-based solution you can find and move your work over to that platform. You'll see literally an order of magnitude in improvement.

I'm assuming your task is highly-parallelizable (it is, right?).
Click to expand...

Well, it might not perform all that well on the GPU either. From my limited exposure, genetic algorithm code tens to be very branchy, and that's exactly the kind of code that GPUs are not so good at.

Intel has the best branch predictor you'll find.
 
Bulldozer should be a good fit if your code is integer math heavy. Else stick with the sandy bridge.
 
I can say, with reasonable confidence (*points at signature*) that a 2600k is going to out do the FX by a huge margin, regardless of what you do with it, and for a very competitive price. BD only makes sense, if you're trying to avoid the expense of a platform switch (new mobo+cpu because you already own a 900-series board), but otherwise, Intel is the better value for any new build and FX flat out useless to change to, if you already have a sandy bridge.

Honestly, though, for such workloads, if it's that important, GPUs are more appropriate, and you should try to find software that will use them, or strongly encourage your dev team to start developing in that direction.

EDIT: On overclocking, I second (or 3rd, 4th, 5th, etc) that you don't need a new board, purely for the chipset, to overclock the sandy bridge. Unless your motherboard is really crappy, any p67 will overclock just as well. Just squeeze whatever you can out of what you got now, and profit. But, again: for these sorts of apps, GPU is the rout to take... CPU is not really cut out for such tasks. Genetic algorithms are going to be hugely FPU and GPUs are (more or less and for lack of better explanation/definition) balls out FPUs.
 
Last edited:
People keep fishing for what bulldozer might actually be good for, but there's no fish.
 
If I wanted something to do my encryption and my compression I will use my Gpus which is way way way way way x10 faster
 
You must log in or register to reply here.
Back
Top