Best Processor for Single Thread Application

[GaetanoL]

I am working with an application written in MatLab programming language. The application is very math intensive, but it does not take advantage of multi-threading. What is the best cpu for executing a single thread? To be clear, multiple cores won't help me here except to split out minor OS processes.

What do you think is the fastest processor out there for single core execution?

Thanks in advance,
Guy

 

[Idontcare]

What do you think is the fastest processor out there for single core execution?

Intel 980X

 

[veri745]

Intel 980X

LOL, so you point him to a 12-thread CPU. You're right, but I still find it funny.

 

[Markfw]

If you overclock, and I7 socket 1366 would probably also equal that.

 

[Triskain]

What do you think is the fastest processor out there for single core execution?

A Clarkdale cranked up to 4,5 GHz.

 

[khon]

I'd go with an i5-655k and do some overclocking.

 

[dbcooper1]

So are you looking specifically for floating point performance?

 

[tweakboy]

Pentium D @ 3Ghz and hyper threading

 

[yh125d]

Highest clocked current intel you can find. i5 670 prolly

 

[LoneNinja]

You'll want the fastest I5/I7 you can get. They currently offer the most performance per clock on a single thread.

 

[Markfw]

Pentium D @ 3Ghz and hyper threading

Not even close......I7 920 @ stock with no HT will beat that ! Let alone with HT@ 4 ghz.

 

[faxon]

Pentium D @ 3Ghz and hyper threading

a core 2 duo would be twice as fast at that clock speed lmao

 

[alyarb]

I guess there's a chance your program can fit inside a 4MB L3 cache. If that's the case, an unlocked clarkdale should get you 5 GHz on air cooling.

If you need low memory latency you have to get an i5 750 or i7.

 

[Cogman]

I guess there's a chance your program can fit inside a 4MB L3 cache. If that's the case, an unlocked clarkdale should get you 5 GHz on air cooling.

If you need low memory latency you have to get an i5 750 or i7.

The program will never be loaded into cache. Cache is for data, not instructions. (ok, there is an Icache, but it usually ranges around 128kb)

Data, however, can often be in the MB of size, so having a large cache does help.

IDC gave the best answer. Top of the line intel offering will give you the fastest single threaded execution.

You should consider breaking your app into multiple threads if possible. This can USUALLY be easily accomplished with most math problems (not all of course, however, most problems that you are crunching numbers on a computer are generally the "do this function on this set of data" Which is easily split up by saying "do this function on this half of the data and at the same time the other half of the data.)

 

[GaetanoL]

So, here's my situation. I have two systems: an Acer Aspire AS5740-5255 LX.PM902.131 Notebook PC with Intel Core i3-330M 2.13GHz, 4GB DDR3, 320GB HDD, Windows 7 Home Premium 64-bit and a Gateway DX4831-05 PT.GAJ02.013 Desktop PC with Intel Core i5-650 3.2GHz, 8GB DDR3, 1TB HDD, eSATA, FireWire, Windows 7 Home Premium 64-bit.

And, for some reason my files are getting run through the i3 more quickly than the i5! This is the complete opposite of what I expected when I purchased the i5, but I thought that maybe the i3 handles single thread better than the i5.

The files are large (~100MB) and the math is pretty complex, so it can take about 24 hours to run a file through the MatLab program. I'm not sure if there is some difference in the configuration between the 2 computers to explain what I'm seeing.

Anyway, any opinions from the experts are appreciated.


Thanks,
Guy

 

[Cogman]

Hmm, that is interesting. The only thing you might look at is your memory bandwidth. Its possible your i3 has a faster memory bus speed then the i5. If the problem involves a lot of data movement, that may be the determining factor (though, in most applications, memory performance means nothing. This may be one of the rare cases where that isn't true)

How about virus scanning software? Are they the same?

 

[Idontcare]

A 2.13 GHz i3 should not process matlab executions faster than a 3.2GHz i5.

First suspect is that something is borked with the i5 setup. Ram bandwidth alone cannot account for X being faster than something which is 1.5X. Does not compute.

How do other canned benchmarks look on the two setups? Single-thread LinPack results for example?

 

[Cogman]

A 2.13 GHz i3 should not process matlab executions faster than a 3.2GHz i5.

First suspect is that something is borked with the i5 setup. Ram bandwidth alone cannot account for X being faster than something which is 1.5X. Does not compute.

How do other canned benchmarks look on the two setups? Single-thread LinPack results for example?

The fastest processor in the world can't do squat without data to use. If the matlab program is doing only sequential data access and modification, it is possible that the ram bandwidth is bottlenecking the processor.

 

[alyarb]

Two OEM clarkdale machines should have very similar memory performance regardless of the core frequency. It's still dual channel 1066 running through that crappy off-die MCH.

 

[f4phantom2500]

So, here's my situation. I have two systems: an Acer Aspire AS5740-5255 LX.PM902.131 Notebook PC with Intel Core i3-330M 2.13GHz, 4GB DDR3, 320GB HDD, Windows 7 Home Premium 64-bit and a Gateway DX4831-05 PT.GAJ02.013 Desktop PC with Intel Core i5-650 3.2GHz, 8GB DDR3, 1TB HDD, eSATA, FireWire, Windows 7 Home Premium 64-bit.

And, for some reason my files are getting run through the i3 more quickly than the i5! This is the complete opposite of what I expected when I purchased the i5, but I thought that maybe the i3 handles single thread better than the i5.

The files are large (~100MB) and the math is pretty complex, so it can take about 24 hours to run a file through the MatLab program. I'm not sure if there is some difference in the configuration between the 2 computers to explain what I'm seeing.

Anyway, any opinions from the experts are appreciated.


Thanks,
Guy

well, it is a gateway...

shot in the dark here, but i guess it's possible that the program has issues when running on a system with more than 4GB of ram, even though it is on a 64 bit OS. it's also possible that, when gateway set up your desktop, they didn't configure the ram to run in dual channel mode. not sure how much of an impact that would have, but you should run cpu-z to check. might wanna try taking out 4GB of RAM from your desktop, and see if that changes the situation at all. you never know with computers.


also, try running superpi (or maybe hyperpi, i think is what the multi-threaded version is called) to see if this problem is consistent with other math programs or just the one you're using.

 

[Lorne]

Does the I5 system have add on gfx card? , If it does there is kinda a test you could try,,,,, open task manager on the I5 and leave it docked/minimized and run it again and see if times change.

Just a hunch.

 

[Voo]

The program will never be loaded into cache. Cache is for data, not instructions. (ok, there is an Icache, but it usually ranges around 128kb)

Data, however, can often be in the MB of size, so having a large cache does help.

Yeah and we always load the next few instructions from RAM into the L1 cache.. Just because you may not distinguish between data/instruction caches doesn't mean it's not there.

 

[Wizlem]

If the program is using a lot of loops with very few calculations per loop, I'd imagine ram performance is the most important factor. It always seemed to me that at the end of a loop, matlab writes all the data back to ram.

 

[extra]

LOL, so you point him to a 12-thread CPU. You're right, but I still find it funny.

Yeah, it is funny, but he answered the question 100% factually.

However in this case, I'd grab a i5 750 and overclock it a lot.

Resist the urge to grab one of the really high clocking 32nm cpus that don't have the memory controller on the die...

 

[taltamir]

LOL, so you point him to a 12-thread CPU. You're right, but I still find it funny.

Yes, he is indeed right... and it is indeed funny.

The nehalem architecture finally did away with the whole "lower core is better for less threaded app, higher core count is better for more threaded apps"... Largely thanks to turbo boost...

Also part of the benefit of the extreme CPUs is both binning and that they are unlocked (for max overclock)...

if willing to OC, then you should go with the above, or with the intel 875K.

If not willing to OC you can still get some solid performance from a nehalem with fewer cores

OP, please provide a budget and a bit more info on what you are doing, why, what for, etc.