Newer, better desktop computer shows same performance as my old laptop!

dule1701e

Honorable
Jul 11, 2013
6
0
10,510
We recently bought a new desktop computer for Matlab/Simulink modelling (i7 3970X, ASUS P9X79 Pro MB, Memory Corsair Dominator Platinum 16GB (4x4GB) DDR). Previously we did our modelling on Toshiba laptops (Portege R830). Our work is mostly processor and memory intensive and does not include a lot disk read/writes. While laptop (corporate computer) is full with bunch of software and low in disk space and memory, new computer is stand-alone with only Windows / Office and Matlab installed. Both operating systems are 64-bit.

We expected our simulations to be at least 2x faster, but are roughly the same speed (while new computer's cpu has 4 times CPUbenchmark.net points than laptop's). Does anybody have any ideas why this is so?
 
Solution
dule1701e,

I've never seen or used MATLAB / Simulink, so my comments are wild speculation, that I hope will cause outrage among others here who will actually be able to diagnose the situation>

1._ The problem may arise from the nature of these programs in that they are running a different sequence for every task. That there is variation in the functions / variables contained in the performance compared algorithms / vectors / matrices / arrays, and etc. such that the task sequences are of unequal complexity. And /or there may be weakly typed objects / sequences for objects that contain variables and constants that are not of equal complexity. In summary, that the tasks being compared on the two systems are not directly...
While I understand it is processor and memory intensive, bottlenecks can happen anywhere that can still affect things. Whether or not a lot of disk read/writes occur, they still occur and add up quickly to cause bottlenecks. In your case you should probably have at a bare minimum a 7200RPM Hard Drive (normally 5400RPM is sold). I personally would suggest a SSD + HDD combination as the best likelihood to remove any possible 'lag' for loading heavy programs like this (temp files on the HDD, SSD just OS and the Modeling software).

As this is a modeling simulator, I noticed among Nvidia is two different brands of video, the normal consumer model and the Quadro lines. Someone tested the Quardos (which I seen in laptops before) with games and said the same sort of thing about 'gaming'. When I looked into it, Quadros were made for Modeling which won't perform the same as games. If you use a 'consumer' video card to do modeling, whilt it can play video games quite nicely, the modeling crawls. The source apparrently is in the design of the software which has specific dependencies to achieve the goals set out (modeling or gaming).

Long and short of it, if your doing modeling, what is the video card in both the Protege (btw is no longer even listed at Toshiba) and the new 'rig'? If you don't have a specified Video Card in the new system optimized for 3D modeling, you will not see as large improvement as you may anticipated if you have a 7200RPM HDD.
 
dule1701e,

I've never seen or used MATLAB / Simulink, so my comments are wild speculation, that I hope will cause outrage among others here who will actually be able to diagnose the situation>

1._ The problem may arise from the nature of these programs in that they are running a different sequence for every task. That there is variation in the functions / variables contained in the performance compared algorithms / vectors / matrices / arrays, and etc. such that the task sequences are of unequal complexity. And /or there may be weakly typed objects / sequences for objects that contain variables and constants that are not of equal complexity. In summary, that the tasks being compared on the two systems are not directly comparable. Perhaps you've already tried something along these lines, but if an identical simulation parameters and sequence set could be entered on both systems in exactly the same way and then the run times compared, if there was the 2X factor in the run times, then I would suggest, that's indicating a hardware issue.

_A> Download and run Passmark Performance Test- there's a free 30-day trial - and compare results by doing an "advanced" baseline search entering all the system parameters possible- CPU, GPU, motherboard. If any result is widely different- especially system rating, CPU, Disk, or Memory scores, there's a place to start.

_B> If there's indication of a hardware issue, my first suspect would be RAM memory incompatibility and you might also run memtest or other test, especially if the Passmark results show low results on memory. Also, check the ASUS site and compare their list of tested memory for the P9X79 board with the exact model of the Corsair used to verify compatibility.

2._ If hardware seems to be performing properly in general_ that is does well on Passmark Performance Test, there's then possible various software configuration and sub-routines that may have conflicts, or that operate poorly on certain hardware >

_A > Verify if there are error / validation subprograms or that require double precision that may be inoperative or throttled due to hardware configuration , i.e. incapable of double precision and/or error correcting RAM. If so, consider running the test sequence on a Xeon > ECC RAM > Quadro system for comparison. If there is access to a Xeon E5-1650 system, that is the closest equivalent CPU to i7-3970X.

_B > For this kind of intensive computational system, you might consider increasing RAM to 32 or 64GB which is the maximum supported by the i7-3970X.

_C > If the CPU is overclocked, consider resetting that to standard clock speed just to eliminate that as a factor.

I apologize for these rambling speculations. Again, I hope someone more knowledgeable will comment.

Cheers,

BambiBoom

[ Dell Precision T5400 > 2X Xeon X5460 quad core @3.16GHz > 16 GB ECC > Quadro. FX 4800 (1.5GB) > WD RE4 / Segt Brcda 500GB > Windows 7 Ultimate 64-bit > AutoCad, Revit, Solidworks 2010, Sketchup Pro, Corel Technical Designer, Adobe CS4 MC, WordP Office, MS Office ]

 
Solution
Hi guys, and thanks for your prompt reply.

Thanks for good ideas there - I will also download the test software and post results soon. Both systems are 64-bit. We are doing power system analysis, which is basically number crunching and does not depend on graphics. Once the start-up conditions are loaded into the memory (roughly 200MB) from there on it is only processor time and processor - memory interaction. I am running the same model on both computers, using the same software package.


 
Hi gents,

I've attached a screenshots of benchmark results from both computers - they show drastical difference in performance, but my programs are still running at similar speeds. I should also probably note that desktop (better computer) has a very cheap, $35 video card since it does not need any specific graphic performance (while I'm running my programs and simulations, nothing is actually changing on the screen).

Thanks
 
dule1701e,

The overall system rating of 1629.3 in Passmark Performance Test 8 demonstrates there is a definite performance issue with your system- or it may be anomalous based on test parameter weighting .

And the mystery is which is wonky. The score for every parameter > CPU, memory, and disk are superb. If you go to Passmark > Manage Baselines > Advanced Search, enter "i7-3970K" for CPU and ASUS P9X79 PRO for motherboard and the search will yield 13 results. Of these results, 12 are valid and the lowest score is anomalous as the disk score of 796 with that CPU and motherboard is hardly possible even with a creaky mech'l drive. The ratings range from 4712 to 6439 > dramatically higher than your 1639. My elderly dual Xeon 4-core manages 1859 with a CPU score of 8528 to your system' (13422) , Mem =729 (2755) , Disk = 929 (2539). (My disk score of 929 is a Western Digital RE4 mech'l which is the reason I discount the system with a 729 score)

This odd, low rating leads me to believe that the system performance is actually extremely good except for what must be very poor graphics performance. I didn't see in the report you posted any graphics scores, and I'm wondering what those scores were? There is a lot of weighting on these scores as the graphics performance as a key measure for the system performance. While you test showed that your system has an extremely high capability in CPU parameters utilized for METLAB simulation, and memory and disk scores were extremely good > you would need a dual Xeon 6 or 8 core to improve them.

However, the system in the search reveals all have graphics cards at the very high end and in the 13 systems, there are only 4 cards > GTX 670, 680 (3), 690 (8), and Quadro 6000, producing 2D scores = 800 > 998 and 3D = 1371 > 6200. If you had one of these cards, your system rating should be more in the 5000 range.

Again, your system may be doing what it's supposed to be doing extremely well, but is simply making a low score based on graphics weighting in the test.

My suggestion is to run identical MATLAB simulations on both systems and if MATLAB has a task timer, compare the overall timing. If the new i3-3970K is not dramatically faster then I could only conclude that the simulation is in some way benefitting from CUDA coprocessor acceleration. Then I would borrow a CUDA based GTX or Quadro card and if the Passmark rating improves substantially, you might consider a Tesla C2075 > which has a 1X DVI video output or K20 CUDA coprocessor to in effect unlock the processing power. If those devices are out of budget ($3,300), you could buy a GTX 670 to have video outut and add a pair of used C2050's with 448 CUDA cores for about $300 each. Again, that strategy would only be valuable if your simulation software can use the CUDA coprocessing. If that's the case, you may benefit from simply adding 3X GTX 670 (SLI) for 4,032 Cores. I have very little understanding of MATLAB except that it addresses weakly typed objects in arrays in a very fluid way that (I may be mistaken, but believe) demands / relies on single and double precision for performance, and I would think that if your running particle or gas simulations that could be even more important. In that case, a Xeon > ECC RAM > Quadro (if CUDA-based) system would be best. But, still, your system does so very well in floating point, integer and other critical CPU functions.

I apologize for rambling on, but the results for what should be an extremely high performance system is possibly based on the hardware having a performance orientation different to that of the software- of which I know practically nothing.

I would be very interested to know what you discover on this matter.

Cheers,

BambiBoom

[ Dell Precision T5400 > 2X Xeon X5460 quad core @3.16GHz > 16 GB ECC > Quadro. FX 4800 (1.5GB) > WD RE4 / Segt Brcda 500GB > Windows 7 Ultimate 64-bit > AutoCad, Revit, Solidworks 2010, Sketchup Pro, Corel Technical Designer, Adobe. CS4 MC, WordP Office, MS Office ]