New PC: biggest bang for the buck?

[DarrenS]

hi

looking to upgrade my pc and I dont know which cpu to pick.

primary use is for pixinisight, what else.

two processors at the top of my price range are

Intel CPU Core i5 Unlocked 2500K Sandy Bridge Quad Core Processor

or

AMD CPU Phenom II 1090T Black Edition

any thoughts on these and which motherboard/chipset - memory configuration to go for?

thanks
Darren

[Carlos Milovic]

This is funny... I was experimenting with OC... I was able to increase the processor's freq up to 3.5 for a while, and the benchmarks went down in linux from ~18secs to ~14secs (non-parallel) and nearly 1 sec down (~7secs) on the parallel one. Unfortunately, the temp went up too fast, so I can't keep that setup
Anyway, I just repeated the tests with no OC on the windows partition (same computer), and to my surprise it was faster than in fedora!  Maybe windows is managing better AMD's frequency boost when performing non-parallel tasks.

I'll repeat the tests with my current soft OC setup (it is supposed to run at 3.2GHz without boost and 3.3GHz with it) and post them.

[Carlos Milovic]

Ok, here are the measurements for my current setup:
        Fedora 15      Windows 7   
         Single   Parallel   Single   Parallel
Average   16.682   8.515   11.526   6.694
Std. Dev   0.361   0.107   0.282   0.108


That is a huge difference! I'm very surprised.

This is my current CPU speed as stated in the BIOS:
Base: 3045 MHz
Boost: 3349 MHz

In Fedora, with the "dmesg | grep MHz" command I get a lecture of 2851.730 MHz, which is consistent to a x14 multiplicator over my current 203 MHz FQC (In the BIOS I use x15 and x16.5 multipliers ). Could it be that Fedora is using the processors at a lower speed?

[Juan Conejero]

Hi Carlos,

There may be different factors here:

- You have to optimize your system. There are lots of unnecessary services that come enabled by default. See:
http://www.harald-hoyer.de/personal/blog/fedora-15-8-services-you-can-most-likely-disable
http://www.mjmwired.net/resources/mjm-services-f14.html

- KDE 4 is slow. GNOME 2.x is faster and makes better use of the available resources. However, Fedora 15 comes with one of the worst things that have happened in the Linux world for a long time: GNOME 3. If you use KDE, there are also some features that are mostly useless and can be disabled.

- Fedora 15 is slower than F14. See for example: http://www.dedoimedo.com/computers/fedora-15.html

Anyway, the differences you've got are very big. There is something in your Linux configuration that is very wrong. Or Fedora 15 is really slow ... 

So far I still have F 14 on all my Linux machines. I guess this will continue the same for some time ...

[Carlos Milovic]

Hi Juan
I just disabled cpuspeed and I came very close to the windows tests (only half a sec slower). I'll try to disable more services, and try again.

[georg.viehoever]

To have a fair comparison, you would have to disable unnecessary Windows services as well.... 

[Carlos Milovic]

Like windows itself? Since I almost don't use it, I don't have even an antivirus installed. So, it is almost as fresh as a new installation.

[Carlos Milovic]

After reboot the gain was higher. And I got the feeling that maybe now the OC would allow larger amounts, without being unstable. It seems that I was right I set the FSB to 240Mhz, and disabled the boost. Now I have ~3360MHz always, on every processor.

The new number are:

   Fedora 15      Windows 7   
   Single   Parallel   Single   Parallel
   10.3   6.502   11.26   6.556
   10.65   6.508   10.72   6.646
   10.35   6.495   10.8   6.411
   10.73   6.537   10.89   6.351
   10.45   6.514   10.64   6.418
Average   10.496   6.511   10.862   6.476
Std. Dev   0.187   0.016   0.241   0.121


As you can see, there was a huge imrpovement on Fedora! Right now it is faster than Windows (and these numbers are also smaller than the previous, so the OC worked well ) on the normal benchmark, and almost the same in the parallel one.

Now I'm happy.

[Juan Conejero]

Very nice results. I think you can squeeze even more performance by fine tuning services and KDE features.

Now I have ~3360MHz always, on every processor.

Make sure your machine is properly cooled.

To have a fair comparison, you would have to disable unnecessary Windows services as well.... 

Like windows itself? Since I almost don't use it, I don't have even an antivirus installed. So, it is almost as fresh as a new installation.

    

[georg.viehoever]

Hi,

all in all, fair  benchmarking CPUs gets more and more difficult. Today, you have to factor in things like TurboBoost, and the latest CPUs even have something like short term Boost taking advantage of thermal inertia, and have to take into account the thermal budget of the on-chip GPUs as well. Simple things like the type of Desktop manager can make a large difference.

Any measured difference <10% is barely noticable anyway in practice.

Georg

[mmirot]

I just tested my new CPU on 1.7
Windows 7, 8 Gigs Memory,  Intel i7 2600
Regular 7.9
parallel 4.8

Does not seem possible but that what it says

Wow??

[georg.viehoever]

I tested PI 1.7.5.779 with latest updates on Amazon's EC2 cloud, on a t1.micro instance (the smallest system they offer, $0.035/h, 615 MB RAM, 7GB free disk space) with OS Windows_Server-2008-R2_SP1-Language_Packs-64Bit-Base-2012.03.13 (ami-9b8fb6ef).

- Benchmark_M74.psm: does not load into latest PI
- Benchmark_M74_parallel.psm: 345 seconds

Not fast, but it works...
Georg

[georg.viehoever]

Same for c1.xlarge system ($1.140/h, 7 GB RAM, 1.690 GB harddisk space):

- Benchmark_M74_parallel.psm: 16.2 seconds

Thats reasonable power...
Georg

[Nocturnal]

I forgot about that benchmark. It runs in 12.9 seconds on my i7-2630 (2G) laptop. The first time it took 2 seconds longer. Did you try running it a few times? There seems to be some lazy loading of processes happening.

[georg.viehoever]

Yes, ran a couple of times and used the median value.
Georg