Fastest single core computer

NeedMoreSpeed

Honorable
Aug 9, 2013
10
0
10,510
I'm running computationally intensive simulation software that is still in the dark ages of single core processing. The bottleneck for this software appears to be the CPU. What would be the fastest CPU to run this software (for a computer that could be built for around £1000) and why?
 
Solution


your best way is an unlocked CPU , and overclock it to 4.8 GHZ .. for that the i5 "K" is the cheapest one. and use a very VERY good cooler to keep it stable .

SNA3

Honorable


your best way is an unlocked CPU , and overclock it to 4.8 GHZ .. for that the i5 "K" is the cheapest one. and use a very VERY good cooler to keep it stable .
 
Solution
I don't think any single core cpu chips are made today.
Currently, the haswell cpu chips have the best efficiency per clock.
They are all quads today.
I might suggest the 4670K with as good of an overclock as you can manage.
The 4770K is more expensive and has a larger cache which possibly might have an impact on your particular app.
The higher basic multiplier is irrelevant since the 4770K will probably oc to about the same high levels as the 4670K.
 

SNA3

Honorable


AMD suck in performance per core compared to intel , and thats why they use more cores to catch up to intel . thanks :)
 

NeedMoreSpeed

Honorable
Aug 9, 2013
10
0
10,510
Thanks for the swift responses.

The CPU wouldn't need to be single core. Multi core would be an advantage as it would allow multiple instances of the software to be run simultaneously, or the operating system etc. to run on a seperate core from the simulation.

I currently use an i7-3520m 2.9GHz cpu running Windows 7. This runs the software faster than any Amazon web services instance. We're looking at getting a dedicated desktop to run the software.

I was looking at the i7 4770k 3.9GHz, suspecting that it could do the same calculations in 25% less time than my laptop.

Overclocking seems tempting. However I've read that it can cause unstable operation. The simulation results need to be reliable. I also need to use this in an office environment where a noisy fan would not be appreciated. Would a small amount of overclocking (low risk, low noise) get better performance than the i7-4770K?
 

COLGeek

Cybernaut
Moderator

Another factor to consider, especially running multiple instances of the tool, is the amount of memory your system has.

Are you using a custom app, or something commercially available for your simulation? Will the app actually support running multiple instances?
 

SNA3

Honorable


the good thing about it , you will loose nothing .. you can turn the overclocking off any time ..

GET the i7-4770K , and gor silent cooler , get the Noctua NH-D14 it is one of the most silent coolers and the most effective air cooler around.

 

NeedMoreSpeed

Honorable
Aug 9, 2013
10
0
10,510


The software only uses 32-bit memory. However I'm hopeful that might change during the lifetime of this PC. I routinely run multiple instances of the software on different cores.
 

NeedMoreSpeed

Honorable
Aug 9, 2013
10
0
10,510


Very informative. Thanks!
 

COLGeek

Cybernaut
Moderator
Are the instances running in Virtual Machines?

What are the specs for the current machine? We know the CPU, but what about the memory and storage (both could be limiting factors). Better yet, what make/model of laptop are you using?

I only ask to assess how much of a performance gain is possible. We could help you build a beast of a rig and the end result could be that the simulation would not really be able to take advantage of the new tech. This is not really a question of raw horsepower, but of optimization of the simulation environment (that may or may not take advantage of said horsepower).
 

NeedMoreSpeed

Honorable
Aug 9, 2013
10
0
10,510
currently running on HP EliteBook 2570p i7-3520M CPU@2.9GHz 8GB Ram, Windows 7 Pro 64bit. Please let me know if there are any other relevant details.

these single thread simulations take 2+ days

I might need to invest in some new dedicated hardware anyway, as I might need to set four simulations running at once, and whilst they're running continue to check emails, manipulate spreadsheets etc. It would be very useful for a business case (to invest in new hardware) to have a reasonable estimate of how much faster I could expect the new rig to run simulations than the laptop runs a single simulation. I was thinking along the lines of in 100 seconds a 2.9GHz would do 290G cycles, a 3.9GHz would do the same number in 75 seconds and thus would take 75% of the time to run a simulation.
 

COLGeek

Cybernaut
Moderator
Do you mean to say that a single instance of a single thread simulation takes 2+ days to run?

How large is the dataset being manipulated? Is the dataset all loaded in memory? Or, is it stored (and accessed from) on the HDD and processed as the simulation runs?
 

SNA3

Honorable


here is a comparison between the 2 CPU , get the I7 4770K , it has also double the cache , and can reach 4.8G overclocked. you will notice alot of difference .

http://ark.intel.com/compare/75123,64893

 

COLGeek

Cybernaut
Moderator
I think SNA3's i7-4770K recommendation is on the right track for processing power, along with (at least)32GB of memory. A SSD for the OS and application (or at least with a fast HDD) would help as well with loading and accessing data.

How much storage does this system require?

 

8350rocks

Distinguished


That's speculation and wildly inaccurate. Thanks :)

@NeedMoreSpeed:

Honestly, if you're going to overclock...the i5-2500k would be your best bet...if you're not going to overclock...the i5-3570k would be better. There is no reason to look at a 4670k, as they don't provide any increase in performance worth discussing at all, when compared to the higher price vs. the 3570k.

If you want to look at AMD, and you wanted to overclock. For a real budget minded machine, you could go to a Phenom II X4 965 and overclock to over 4 GHz and get solid performance. If you want more cores, you could look at the 1090T 6 core and overclock equally well.

In the FX line, your best bet would likely be something like the FX 6350 (6 cores) and overclock to around 4.7-4.8 GHz. (You could run this one stock @ 3.9 GHz as well)

Keep in mind, the AMD solutions will easily be quite a bit cheaper than the intel ones and offer you a lot of bang for your buck.
 

SNA3

Honorable


Whats the not accurate ? per core at the same clock AMD is waaay behind Intel.
 

COLGeek

Cybernaut
Moderator
The more I think about this thread, the more I see how this simulation might run well on a server just churning in the background. Anything that takes 2+ days to run now is a huge dataset and that machine should not be used for other general computing tasks while crunching numbers. A Xeon or Opteron based server with plenty of memory could grind through all of this and free up other systems for general use.

Just another consideration for NeedMoreSpeed to think about. This really doesn't sound some like a typical gaming beast (but a gaming beast certainly could handle this task) sort of scenario.
 

8350rocks

Distinguished


That's a logical fallacy...you can't compare separate architectures at the same clock speeds. Compare them where the manufacturers distribute them in stock format, or overclocked equal amounts to get a fair comparison.

Your logical fallacy is false equivalency...
 

rgd1101

Don't
Moderator