Benchmarking Performance: CPU System Tests

Our first set of tests is our general system tests. These set of tests are meant to emulate more about what people usually do on a system, like opening large files or processing small stacks of data. This is a bit different to our office testing, which uses more industry standard benchmarks, and a few of the benchmarks here are relatively new and different.

All of our benchmark results can also be found in our benchmark engine, Bench.

Strategic AI

One of the hot button topics this year (and for the next few years, no doubt) is how technology is shifting to using artificial intelligence and purpose built AI hardware to perform better analysis in low power environments. AI is not relatively new as a concept, as we have had it for over 50 years. What is new is the movement to neural network based training and inference: moving from ‘if this then that’ sort of AI to convolutional networks that can perform fractional analysis of all the parameters.

Unfortunately the movement of the neural-network ecosystem is fast paced right now, especially in software. Every few months or so, announcements are made on new software frameworks, improvements in accuracy, or fundamental paradigm shifts in how these networks should be calculated for accuracy, power, performance, and what the underlying hardware should support in order to do so. There is no situational AI benchmarking tools using network topologies that will remain relevant in 2-4 months, let alone an 18-24 month processor benchmark cycle. So to that end our AI test becomes the best of the rest: strategic AI in the latest video games.

For our test we use the in-game Civilization 6 AI benchmark with a few custom modifications. Civilization is one of the most popular strategy video games on the market, heralded for its ability for extended gameplay and for users to suddenly lose 8 hours in a day because they want to play ‘one more turn’. A strenuous setting would involve a large map with 20 AI players on the most difficult settings, leading to a turn time (waiting for the AI players to all move in one turn) to exceed several minutes on a mid-range system. Note that a Civilization game can easily run for over 500 turns and be played over several months due to the level of engagement and complexity.

Before the benchmark is run, we change the game settings for medium visual complexity at a 1920x1080 resolution while using a GTX 1080 graphics card, such that any rendered graphics are not interfering with the benchmark measurements. Our benchmark run uses a command line method to call the built-in AI benchmark, which features 8 AI players on a medium size map but in a late game scenario with most of the map discovered, each civilization in the throes of modern warfare. We set the benchmark to play for 15 turns, and output the per-turn time, which is then read into the script with the geometric mean calculated. This benchmark is newer than most of the others, so we only have a few data points so far:

System: Civilization 6 AI (1080p Medium + GTX 1080)

Our Strategic AI test is new to the scene, and it looks like there is at least an asymptotic result wken you have a 'good enough' processor.

PDF Opening

First up is a self-penned test using a monstrous PDF we once received in advance of attending an event. While the PDF was only a single page, it had so many high-quality layers embedded it was taking north of 15 seconds to open and to gain control on the mid-range notebook I was using at the time. This put it as a great candidate for our 'let's open an obnoxious PDF' test. Here we use Adobe Reader DC, and disable all the update functionality within. The benchmark sets the screen to 1080p, opens the PDF to in fit-to-screen mode, and measures the time from sending the command to open the PDF until it is fully displayed and the user can take control of the software again. The test is repeated ten times, and the average time taken. Results are in milliseconds.

System: PDF Opening with Adobe Reader DC

Single thread frequency usualy works well for PDF Opening, although as we add on more high performance cores it becomes more difficult for the system to pin that individual thread to a single core and get the full turbo boost - if anything flares up on any other core then it brings the frequencies down. I suspect that is what is happening here and the next couple of thests where the i7-8700K sits behind the i7-7700K and i7-7740X.

FCAT Processing: link

One of the more interesting workloads that has crossed our desks in recent quarters is FCAT - the tool we use to measure stuttering in gaming due to dropped or runt frames. The FCAT process requires enabling a color-based overlay onto a game, recording the gameplay, and then parsing the video file through the analysis software. The software is mostly single-threaded, however because the video is basically in a raw format, the file size is large and requires moving a lot of data around. For our test, we take a 90-second clip of the Rise of the Tomb Raider benchmark running on a GTX 980 Ti at 1440p, which comes in around 21 GB, and measure the time it takes to process through the visual analysis tool.

System: FCAT Processing ROTR 1440p GTX980Ti Data

Dolphin Benchmark: link

Many emulators are often bound by single thread CPU performance, and general reports tended to suggest that Haswell provided a significant boost to emulator performance. This benchmark runs a Wii program that ray traces a complex 3D scene inside the Dolphin Wii emulator. Performance on this benchmark is a good proxy of the speed of Dolphin CPU emulation, which is an intensive single core task using most aspects of a CPU. Results are given in minutes, where the Wii itself scores 17.53 minutes.

System: Dolphin 5.0 Render Test

3D Movement Algorithm Test v2.1: link

This is the latest version of the self-penned 3DPM benchmark. The goal of 3DPM is to simulate semi-optimized scientific algorithms taken directly from my doctorate thesis. Version 2.1 improves over 2.0 by passing the main particle structs by reference rather than by value, and decreasing the amount of double->float->double recasts the compiler was adding in. It affords a ~25% speed-up over v2.0, which means new data.

System: 3D Particle Movement v2.1

DigiCortex v1.20: link

Despite being a couple of years old, the DigiCortex software is a pet project for the visualization of neuron and synapse activity in the brain. The software comes with a variety of benchmark modes, and we take the small benchmark which runs a 32k neuron/1.8B synapse simulation. The results on the output are given as a fraction of whether the system can simulate in real-time, so anything above a value of one is suitable for real-time work. The benchmark offers a 'no firing synapse' mode, which in essence detects DRAM and bus speed, however we take the firing mode which adds CPU work with every firing.

System: DigiCortex 1.20 (32k Neuron, 1.8B Synapse)

DigiCortex can take advantage of the extra cores, paired with the faster DDR4-2666 memory. The Ryzen 7 chips still sit at the top here however.

Agisoft Photoscan 1.3.3: link

Photoscan stays in our benchmark suite from the previous version, however now we are running on Windows 10 so features such as Speed Shift on the latest processors come into play. The concept of Photoscan is translating many 2D images into a 3D model - so the more detailed the images, and the more you have, the better the model. The algorithm has four stages, some single threaded and some multi-threaded, along with some cache/memory dependency in there as well. For some of the more variable threaded workload, features such as Speed Shift and XFR will be able to take advantage of CPU stalls or downtime, giving sizeable speedups on newer microarchitectures. The 1.3.3 test is relatively new, so has only been run on a few parts so far.

System: Agisoft Photoscan 1.3.3 (Large) Total Time

Benchmark Overview Benchmarking Performance: CPU Rendering Tests
Comments Locked


View All Comments

  • FireSnake - Thursday, October 5, 2017 - link

    Awesome revies! Let's read...
  • prisonerX - Thursday, October 5, 2017 - link

    No need, here is a quick summary: "Intel blind panic."
  • StevoLincolnite - Thursday, October 5, 2017 - link

    At-least they have finally soundly beat my 3930K in the mainstream after 6 years.

    Still. No point me upgrading just yet.
  • mapesdhs - Friday, October 6, 2017 - link

    Even then there's an interesting option if you want threaded performance; I just upgraded to a XEON E5-2680 v2 (IB-EP) for 165 UKP. Lower 1T speed for sure, but MT should be the same or better as a 3930K @ 4.8. No oc means more stable, less heat/noise/power, and being IB-based means it ups the slots to PCIe 3.0. Not a relevant choice for gaming, but a possibility for those doing VMs, rendering, etc., and just want to get by for a little while longer.
  • Breit - Friday, October 6, 2017 - link

    OR search for an XEON E5-1680v2... :)
    It's an Ivy Bridge-E 8c/16t chip that will fit in Sandy Bridge-E mainboards (x79) and has an unlocked multiplier opposed to this E5-2680v2. So with this you won't lose your overclocking ability.

    But in the end, I guess that the greatly reduced power draw and the more "modern" platform from an i7-8700K system compared to the x79 platform will give it the edge here.
  • mapesdhs - Monday, October 9, 2017 - link

    Very interesting that the 1680 v2 is unlocked, I didn't know that.

    Alas though, availability of the 1680 v2 is basically zero, whereas the 2680 v2 is very easy to find, and the cost of 1680 v2s which are available (outside the UK) is extremely high (typical BIN of 600 UKP, normal auction start price of 350 UKP, completed listings only shown for BIN items which were purchased for between 500 and 600 UKP). By contrast, I bought several 2680 v2s for 165 UKP each. Testing on a P9X79 WS (all-core turbo of 3.1) gives a very impressive 15.44 for CB 11.5, and 1381 for CB R15 which is faster than a stock 8700K (for reference, the 1680 v2 scores 1230 for CB R15). Note the following page on AT has a very handy summary of all the turbo bin levels:

    So, I'm very pleased with the 2680 v2 purchase, it's faster than my 3930K @ 4.8, runs with very low temps, much lower power draw, hence less heat, less fan noise and since it's not oc'd it'll be solidly reliable (this particular test system will eventually be in an office in India, so power/heat/reliability is critical). For the target systems in question, it's a great solution. Only thing I noticed so far is it didn't like me trying to set a 2133 RAM speed, but it worked ok at 1866; I can probably tighten the timings instead, currently just at 9/11/10/28/2T (GSkill 32GB kit, 8x4GB).

    The 4930K I have though will go into my gaming system (R4E), since I don't mind the oc'ing fun, higher noise, etc., but it's not a system I'll use for converting video, for that I have a 6850K.

  • MrSpadge - Friday, October 6, 2017 - link

    Full throttle: yes. Panic: no. Blind: no.
  • Zingam - Saturday, October 7, 2017 - link

    Can you buy it? No? Paper launch of Unobtanium 8000? -> panic, PR propaganda bullshit and dirty Intel marketing tactics as usual targeted at lamer fanboys.

    This comment is written by an Intel user! ;)
  • prisonerX - Saturday, October 7, 2017 - link

    We've got enough dumb Intel apologists here already, thanks.
  • coolhardware - Sunday, October 8, 2017 - link

    The i7-8700 is *finally* going to replace my trusty i5-2500K.

    Ordered my 8700 on Amazon ($319) and looking forward to a nice upgrade :-) That is a lot of CPU for the money IMHO.

    Kudos to AMD for bringing competition back to the CPU market!

Log in

Don't have an account? Sign up now