IGP Compute

One of the touted benefits of Haswell is the compute capability afforded by the IGP.  For anyone using DirectCompute or C++ AMP, the compute units of the HD 4600 can be exploited as easily as any discrete GPU, although efficiency might come into question.  Shown in some of the benchmarks below, it is faster for some of our computational software to run on the IGP than the CPU (particularly the highly multithreaded scenarios). 

Grid Solvers - Explicit Finite Difference on IGP

As before, we test both 2D and 3D explicit finite difference simulations with 2n nodes in each dimension, using OpenMP as the threading operator in single precision.  The grid is isotropic and the boundary conditions are sinks.  We iterate through a series of grid sizes, and results are shown in terms of ‘million nodes per second’ where the peak value is given in the results – higher is better.

Explicit Finite Difference Solver (2D) on IGPExplicit Finite Difference Solver (3D) on IGP

N-Body Simulation on IGP

As with the CPU compute, we run a simulation of 10240 particles of equal mass - the output for this code is in terms of GFLOPs, and the result recorded was the peak GFLOPs value.

N-Body Simulation on IGP

Matrix Multiplication on IGP

Matrix Multiplication occurs in a number of mathematical models, and is typically designed to avoid memory accesses where possible and optimize for a number of reads and writes depending on the registers available to each thread or batch of dispatched threads.  He we have a crude MatMul implementation, and iterate through a variety of matrix sizes to find the peak speed.  Results are given in terms of ‘million nodes per second’ and a higher number is better.

Matrix Multiplication on IGP

3D Particle Movement on IGP

Similar to our 3DPM Multithreaded test, except we run the fastest of our six movement algorithms with several million threads, each moving a particle in a random direction for a fixed number of steps.  Final results are given in million movements per second, and a higher number is better.

3D Particle Movement on IGP

CPU Compute Overclocking Results


View All Comments

  • UltraWide - Friday, December 13, 2013 - link

    On this page:

    What are the exact voltages each timing/subtiming? You only list the subtimings and peak MHz, but no voltages.
  • SorenLerby - Friday, December 13, 2013 - link

    Love my job, since I've been bringing in $5600… I sit at home, music playing while I work in front of my new iMac that I got now that I'm making it online(Click on menu Home)
  • JoannWDean - Saturday, December 14, 2013 - link

    my buddy's aunt earned 14958 dollar past week. she been working on the laptop and got a 510900 dollar home. All she did was get blessed and put into action the information leaked on this site... http://cpl.pw/OKeIJo Reply
  • teiva - Thursday, January 02, 2014 - link

    My pet cat's been earning $14823 a month just by eating Whiskas biscuits dipped in cream from albino goats that only sleep in a northerly direction lol Reply
  • jeffrey - Friday, December 13, 2013 - link

    Ian Cutress,
    Hello! This is another article stating 1866/C9 being the minimum for Haswell and to avoid 1600 or less. Even going so far as to say, "Any kit 1600 MHz or less is usually bad news."

    However, this ignores 1600/C8 modules. The 1600/C8 score a 200 on your Performance Index at stock timings. This is at your recommended 200 level. There are several kits of 2x4 GB 1600/C8 on Newegg that have memory profiles of 8-8-8-24 at 1.5v. I'll repeat, these 1600 8-8-8-24 1.5v kits score 200 on the Performance Index and hit the current memory sweet spot for most people of 2x4 GB. This scores within around 3% of the 1866/C9 kits which have a Performance score of 207.

    The reason I bring this up is that the 1600 8-8-8-24 kits are often less expensive than the 1866/C9 kits and offer essentially all of the performance.

    I enjoy reading your articles and appreciate how active you have been lately!
  • jeffrey - Friday, December 13, 2013 - link

    Here is an example of a great value for a 2x4 GB kit. This item has 500+ positive reviews and normally sells out when it goes on sale (as it is now).

    1600 MHz 8-8-8-24 1.5V
  • fractal9 - Friday, December 13, 2013 - link

    Where in the article does it state this? These days I want the most compatibility and reliability, and that means being in the JEDEC standard and at 1.5v. I don't care about a few extra fps or synthetic performance figures, I'll save my $80 and enjoy increased stability. Reply
  • Senti - Friday, December 13, 2013 - link

    "we are not going to set any records" with tRFC at 10? lol, sure we would! Do you actually understand the timings you are writing about? Real tRFC is the line above where the numbers are near 300. Now we have huge table of timings that is totally useless because we can't trust it at all...

    Btw, your spam filter is awful – looks like it blocked my (dynamic) IP for unknown reason.
  • Hairs_ - Friday, December 13, 2013 - link

    Oh dear god. Not this again. It is time for a thorough debunking, based on the data presented, because clearly it is not going to stop.

    Established from Anandtech's own testing data:
    There is no distinguishable, patternable difference between 8GB and 16GB memory kits. In some cases, there is a swing one way or the other, but nothing is consistent.
    I think that most enthusiasts would recommend, if asked by a friend on a midrange budget, a 1600 8gb kit. The lowest price for that at C9 timings on newegg is $55. Not the $150 - 250 being bandied about for these moronic kits.


    An 1866 C9 16GB kit comes in at ~$115. (It is safe to assume that *some* kit will always be on offer.)

    So let's look at the conclusive data:

    From all the graphs, we can conclude the difference between the Best "stupid" kit and a reasonable 1600 c9 kit is:

    Average 2.78%
    Max 13.27%
    Min 0.00%

    The best results for the stupid kits are:
    13.27% running Explicit Finite Solver (3d only, the 2d result was 3.3%) on an IGP. Anyone running this is not using an IGP to do it, I suspect.
    12.91% on USB3 copy. The page for those results states they are pulled from the Motherboard reviews, so there's no indication this speed isn't motherboard specific rather than memory specific (as some AsRock boards have USB boost which allegedly seems to work). One result on this same test had a 34.55% lead, but it had the same lead (or larger) on "faster" kits than the 1600 kit, so I am treating this as an outlier.

    The vast majority of the results are just not in the stupid kit's favour.

    Test Difference Best Stupid Kit 1600 C9 Kit Testing
    Graph 1 5.67% 12.86 12.17 IGP
    Graph 2 1.66% 9.2 9.05 IGP
    Graph 3 4.65% 9 8.6 IGP
    Graph 4 1.29% 105.39 104.05 Direct
    Graph 5 1.47% 62.94 62.03 Direct
    Graph 6 1.28% 47.4 46.8 Direct
    Graph 7 0.41% 49.55 49.35 Direct
    Graph 8 3.14% 212.5 206.03 Weird triple unbalanced which won't ever occur
    Graph 9 3.13% 123.93 120.17 Weird triple unbalanced which won't ever occur
    Graph 10 1.04% 91 90.06 Weird triple unbalanced which won't ever occur
    Graph 11 5.86% 113.85 107.55 Weird triple unbalanced which won't ever occur
    Graph 12 1.00% 54.834 55.383 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 13 12.91% 47.33 53.44 CPU - From MOBO reviews so no indication if same testbed was used or if MOBO had USB speed boost.
    Graph 14 0.00% 47 47 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 15 1.56% 64 65 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 16 3.71% 190.32 183.51 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 17 1.23% 49.57 48.97 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 18 2.13% 4.8 4.7 CPU - From MOBO reviews so no indication if same testbed was used
    Graph 19 0.10% 132.06 131.93 CPU Compute
    Graph 20 1.59% 755.04 743.25 CPU Compute
    Graph 21 0.01% 38.793 38.791 CPU Compute
    Graph 22 0.45% 1354 1348 CPU Compute
    Graph 23 1.32% 613 605 CPU Compute
    Graph 24 1.33% 152 150 CPU Compute
    Graph 25 3.31% 1780 1723 IGP Compute
    Graph 26 13.27% 222 196 IGP Compute - Explicit Finite Difference Solver (3d), not something going to run on an IGP…
    Graph 27 2.49% 150.653 146.993 IGP Compute
    Graph 28 1.93% 50.731 49.769 IGP Compute
    Graph 29 2.60% 7103 6923 IGP Compute
  • Hairs_ - Friday, December 13, 2013 - link

    The "Weird triple unbalanced" above refers to the idea that there is some lunatic out there running a 4 generation old Dual GPU card in crossfire with a different, single GPU from the same generation but a different spec. This is referred to as "esoteric" but I think it's safe to assume that the only person doing this is someone who has video cards sitting in a pile next to a testbed because they're a video card reviewer.

    Suffice to say I don't think this test is representative of any real world use pattern, but then neither are a lot of the tests where the stupid kits "shine" (if you can call 12% advantage for a 300% price premium "shining").

Log in

Don't have an account? Sign up now