Message boards :
Number crunching :
Building a Computer - Advice Needed
Message board moderation
Author | Message |
---|---|
Poolhall Junkie Send message Joined: 16 Aug 16 Posts: 3 Credit: 3,618,898 RAC: 0 |
I'm a "computer dummy" and need advice. I just joined and am currently using an old Pentium computer to crunch numbers 24/7. I intend to build a new PC which will be used exclusively for distributed computing projects and need advice. My plan is to leave the computer on 24/7 crunching data for SETI or similar projects. Here are the things I know I need: · Monitor, keyboard, and mouse for basic interaction with the computer. · Internet access to get / receive data from distributed computer projects. This new computer is currently a black box to me. This is where I need advice: · Overall strategy: Is it best for me to buy the best and fastest (hence expensive) hardware available in a single ‘box’, or is there a way to buy older CPU technology at a huge discount, then ‘wire it all together’, throw it into a server rack, and get a bigger bang for the buck? · Computing power: What type of CPU should I get? Multi-core? How many cores are possible? Should I overclock? · Memory: How much RAM needed for fastest crunching? · Storage: How much hard disk storage capacity needed? · Energy consumption: I want to run as cool and efficient as possible. · Cooling: How to keep this monster from melting? · Budget: The cheaper the better, but I’d probably be willing to go as high as $2,000 overall cost. I already have keyboards, mice, and monitors, so that doesn’t impact my $2,000 budget. Any advice would be greatly appreciated. Thanks!!! |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, $2000 Good budget. Do the GPUs have to be the newest or greatest? If not, you could do something like this. http://pcpartpicker.com/list/rcvJf8 This would give you a 12 core, with a 980TI Hybrid. If you don't want the 980Ti, you could go for other GPUs but I like the fact hybrid run cooler and you shunt most of the heat out of the case. This board will allow for 2 GPUs but you might have to get a bigger Power supply for it if you go for 2 GPUs Of course the issue with any radiator is making sure they have mounting point. For 2 radiators, not a problem but if you go for 3 then you have to start getting creative. Zalster |
Poolhall Junkie Send message Joined: 16 Aug 16 Posts: 3 Credit: 3,618,898 RAC: 0 |
Thanks so much for spec'ing out a system! Super helpful. Follow-up questions: 1. Based on the link, how does this create a 12-core? Were you assuming I'd buy 2 CPUs, or does the expensive graphics card play a role here? 2. Being computer hardware ignorant, I'm surprised a graphics card is so expensive, compared to the actual CPU. Does the graphics card do more than just throw graphics to the monitor? Thanks again for your help! -Michael |
Rich Send message Joined: 27 Oct 14 Posts: 4 Credit: 49,285,910 RAC: 0 |
Pool: The processor is a 6 core processor and with Hyper Threading it looks like a 12 core processor to the operating system, 12 logical processors. The GPU (graphics processor unit) can be setup to process work units also. Reading some of the posts will help you get familiar with how to do that. Rich |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Thanks so much for spec'ing out a system! Super helpful. No problem. This is a Intel physical 6 core, but you can turn on hyperthreading in the bios and it will create 6 more virtual cores that can be used for crunching. AMD does hyperthreading as well but I've found the intel work better (personal choice) So when I speak of 12 cores, I mean 6 physical plus 6 virtual on that single CPU chip. The expensive GPU is where most of the analysis of the Data get done now a days. You can analyze (crunch) the data on the CPU but they tend to be much slower than doing the work on the GPU. Depending on the Graphic card, you will be able to do more work or more complex data..example GPUGrid is another project with very complex and very large data (work units) They specify that unless you have a certain generation of graphic cards, you probably shouldn't try to crunch their work units due to the length of time it will take to do so. Here at Seti, for a long time, the applications that are used to crunch the work units compensated for older GPUs. But lately, we have newer data that required a more indepth analysis so it places a greater amount of stress on the hardware of a computer. As such, older and slower components are having a harder time dealing with those work units. Why did I pick that specific graphic card? First, I have several and they are still good producers of work. Second, they have the ability to run more than 1 work units at a time Third, they are hybrid so they run much cooler than non hybrids Fourth, since the 10X0s have come out, the prices on the 900s series have started to drop. Hope this helps some Zalster |
Poolhall Junkie Send message Joined: 16 Aug 16 Posts: 3 Credit: 3,618,898 RAC: 0 |
FANTASTIC! Super helpful information. Thanks so much, and I'll read other posts to educate myself. Hopefully I'll build out a much more powerful computer in the next month. Just doing my part to help with the search! -Michael |
ML1 Send message Joined: 25 Nov 01 Posts: 20291 Credit: 7,508,002 RAC: 20 |
So when I speak of 12 cores, I mean 6 physical plus 6 virtual on that single CPU chip. Small nit-pick on the detail: I would describe that as "6 physical CPU cores that are 'multiplexed' by 'hyperthreading' to implement 12 logical ('virtual') cores". What the hyperthreading does is to share the processing cycles of one physical core between two processing threads. You never get anything for nothing and so that in effect gives you two 'logical' cores running at half the clock speed of the actual one physical core. However... You get a performance gain if one process thread becomes 'stalled' waiting for hardware or some other CPU resource whereby the other process thread can take advantage of otherwise wasted CPU cycles whilst the first process waits. In practical terms, this means that the two 'hyperthreads' run at anything from approximately 50% of full core speed to about 65% of full core speed. Overall, although each process is running slower, you can get up to about 130% throughput over that of not utilizing the hyperthreading. However, you sacrifice some of the system 'responsiveness'. The expensive GPU is where most of the analysis of the Data get done now a days. You can analyze (crunch) the data on the CPU but they tend to be much slower than doing the work on the GPU. Is not the best advice to go for the very latest mid-range GPUs for both optimum processing power and optimum power consumption/expenditure? And go for a small-ish SSD on the leanest machine that you can install!! Look up the Lunatics ;-) Happy fast crunchin'! Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
ML1 Send message Joined: 25 Nov 01 Posts: 20291 Credit: 7,508,002 RAC: 20 |
FANTASTIC! Super helpful information. Thanks so much, and I'll read other posts to educate myself. Hopefully I'll build out a much more powerful computer in the next month. Just doing my part to help with the search! Be very welcome onboard. Note also that a big part of our search is in learning how best/fast to search... Keep searchin' (in all respects and all ways) And have fun! Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
The expensive GPU is where most of the analysis of the Data get done now a days. You can analyze (crunch) the data on the CPU but they tend to be much slower than doing the work on the GPU. You misinterpreted this ML1. I didn't mean the most expensive GPU is where most of the work is done, I meant the GPU vs CPU gets better productivity. ;) therefore almost any GPU would be better productive than just a CPU cruncher. I picked the hybrid as he wanted it to run cool, and short of a full watercooled system, the hybrid are an easy alternative. Zalster |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.