15

Dear Mathematica users

I am looking into buying a new computer and Mathematica is one of the main applications I run. To help me assess the overwhelming welter of choice out there:

  1. I'd like to know whether there are any system performance metrics, along the lines of those measured by Geekbench http://browser.primatelabs.com/processor-benchmarks, defined specifically for working with Mathematica, and whether there are any tables on the web comparing currently sold machines.

  2. It would also be very helpful if these metrics came with detailed descriptions of what they mean. I haven't been able to find a clear definition of what Geekbench numbers mean - e.g. does a measurement of 2000 mean a machine that runs roughly twice as fast as one measuring 1000 and what exactly was it doing when it was running.

Many thanks in advance

Selene Routley
  • 259
  • 2
  • 7
  • The best CPU-benchmarks available (in my opinion) are http://www.cpubenchmark.net and Cinebench... Althouth this is a very interesting question, I'm afraid I could be closed because it's not directly related to Mathematica... – Rod Jul 07 '13 at 02:13
  • 1
    From my limited experience of Geekbench, the odd choice of weightings for the different components makes the result rather meaningless as a practical performance metric. This is a general problem: any real workload emphasizes particular aspects of performance more than others, and unless you can define the workload very well, you will not be able to make reliable inferences from benchmark scores. For Mathematica this is a particularly serious problem, since everyone uses it in a different way. The built-in benchmark is mostly reasonable, but might not be fully representative of your usage. – Oleksandr R. Jul 07 '13 at 02:17
  • 2
    This question appears to be off-topic because it is not about using Mathematica software – m_goldberg Jul 07 '13 at 02:37
  • Needs["Benchmarking\"]; Benchmark[]` Now if I can find a webpage that has a list of results for different machines... – Eric Brown Jul 07 '13 at 02:54
  • 1
    @EricBrown http://www.xbitlabs.com/articles/cpu/display/core-i7-4770k_9.html – Oleksandr R. Jul 07 '13 at 02:54
  • @OleksandrR. That does nicely! My Macbook Air i7 gets 0.898. Maybe this will give some perspective of the general performance of portables. – Eric Brown Jul 07 '13 at 03:05
  • 4
    Whichever computer you buy, try to get as much RAM as possible - Mathematica eats RAM for breakfast. One questioner here has 128 GB - that's a good start... :) – cormullion Jul 07 '13 at 07:29
  • Unless you plan to process movies in memory I found 8 GB is generally sufficient. – Sjoerd C. de Vries Jul 07 '13 at 08:07
  • @SjoerdC.deVries true, we don't know the OP's requirements. I have 12GB, and frequently run out on graphics-related work... – cormullion Jul 07 '13 at 08:31
  • @cormullion Due to IT restrictions at my work, I'm running Mathematica in a virtual machine there. It has 2 GB only. – Sjoerd C. de Vries Jul 07 '13 at 08:39
  • @EricBrown My results (3930-K, 8GB RAM, 128 GB SSD) for comparison: {"MachineName" -> "6-node homogeneous cluster", "System" -> "Windows-x86-64", "BenchmarkName" -> "MathematicaMark9", "FullVersionNumber" -> "9.0.1", "Date" -> "July 7, 2013", "BenchmarkResult" -> 3.802, "TotalTime" -> 65.531}. – Rod Jul 07 '13 at 13:12
  • @RodLm beware, though, that the parallel benchmark score is not directly comparable to the ordinary one. As far as Passmark and Cinebench are concerned: ray tracing is a reasonable benchmark but it has very particular characteristics that are not generalizable; Passmark doesn't appear to describe the benchmark components properly (and from what description there is, they seem too naive to be useful). SPEC CPU is generally considered to be the CPU benchmark, especially the GCC component of it. (Other components tend to be "broken" by excessive tuning in vendor compilers.) – Oleksandr R. Jul 07 '13 at 14:20
  • @OleksandrR. That's interesting... almost all CPU benchmarks I've done involve parallelization. I wonder why Mathematica is not a "serious" competitor in this benchmark market, once it has (IMO) all necessary attibrutes to test and report CPU's, in single or parallel mode... – Rod Jul 07 '13 at 22:17
  • I second @cormullion! MMA eats RAM for breakfast, lunch, dinner and still goes to bed hungry. I have 16 GB RAM and a 128 GB SSD disks dedicated for the swap file. – Ajasja Jan 09 '14 at 16:54
  • Oh yeah, and guess what's for dessert? That's right MORE RAM :) But I guess it depends on what you use MMA for. – Ajasja Jan 09 '14 at 17:06

1 Answers1

17

From Oleksandr's comments:

From my limited experience of Geekbench, the odd choice of weightings for the different components makes the result rather meaningless as a practical performance metric. This is a general problem: any real workload emphasizes particular aspects of performance more than others, and unless you can define the workload very well, you will not be able to make reliable inferences from benchmark scores. For Mathematica this is a particularly serious problem, since everyone uses it in a different way. The built-in benchmark is mostly reasonable, but might not be fully representative of your usage.

As Erik Brown remarked, the built-in benchmark can be used as follows:

Needs["Benchmarking`"];
Benchmark[]

It has results for some computer systems built-in as well:

$BenchmarkSystems

3.07 GHz Core i7-950 (8 Cores) (Windows 7 Pro (64-bit) Desktop)
1.73 GHz Core i7-820QM (8 Cores) (Windows 7 Ultimate (64-bit) Laptop)
2.4 Ghz Core 2 Duo Mobile T8300 (2 Cores) (MacBook OS X Snow Leopard (64-bit) Laptop)
2.80 GHz Core 2 Duo Mobile T9600 (2 Cores) (Windows 7 Pro (64-bit) Laptop)
1.6 GHz Core 2 Duo Mobile L7500 (2 Cores) (Windows 7 Pro (32-bit) Laptop)
2.93 GHz Core i7-940 (8 Cores) (Linux Ubuntu (64-bit) Desktop)
3.00 GHz Core 2 Duo E8400 (2 Cores) (Linux Ubuntu (64-bit) Desktop)
2.60 GHz Core 2 Duo Mobile T7800 (2 Cores) (Windows XP Pro (32-bit) Laptop)
2.13 GHz Core 2 Duo E6400 (2 Cores) (Windows Vista (32-bit) Server)
2 x 2.66 GHz Dual Core Xeon 5150 (4 Cores) (MacPro OS X Snow Leopard (64-bit) Server)
2 x 2.26 GHz Quad Core Xeon E5520 (8 Cores) (Mac XServe OS X (64-bit) Server)
2 x 2.00 GHz G5 PowerPC (2 Cores) (Mac OS X (32-bit) Desktop)
3.06 GHz Core 2 Duo E8435 (2 Cores) (iMac OS X Snow Leopard (64-bit) Desktop)
2.67 GHz Core 2 Quad Q9450 (4 Cores) (Linux Debian (64-bit) Desktop)
2 x 2.80 GHz Opteron 254 (2 Cores) (Windows XP Pro (64-bit) Server)

Mathematica graphics

Mathematica graphics

Sjoerd C. de Vries
  • 65,815
  • 14
  • 188
  • 323
  • nice, but note this isnt taxing memory at all (I see not even a blip on the windows task manager memory monitor). – george2079 Jan 09 '14 at 16:57
  • @george2079 Yes, that's good to take note of. It's primarily a CPU benchmark (peak load at my PC about 57%). – Sjoerd C. de Vries Jan 09 '14 at 17:03
  • maybe we could assemble a community table with benchmark results on different machines and OSs here on mathematica.SE? – shrx Jan 09 '14 at 18:24
  • @shrx If you post your result of Benchmark[][[1]] and a string describing your computer's characteristics (similar to the ones to the left of the histogram) I'll include them in the figure. – Sjoerd C. de Vries Jan 09 '14 at 18:55
  • Mr.Wizard probably has one of the fastest computers. But, since he has version 7, I don't think the comparison would be all that useful. – Oleksandr R. Jan 09 '14 at 22:23
  • @OleksandrR. You're familiar with his set-up? – Sjoerd C. de Vries Jan 09 '14 at 22:36
  • @SjoerdC.deVries not really. But I know that he's an enthusiast, and his timings are always much better than mine although my computer is not that bad. – Oleksandr R. Jan 09 '14 at 23:25
  • I'm going to have to put my overclock back on to take the stop spot back :) – s0rce Jan 11 '14 at 01:22