tiborrr

5. Testing results 5.1. CPU core count impact on FPS The purpose of this test is to test out the impact of the CPU core count on the general FPS performance. The core count tested covers most of the currently available CPUs. Single core performance is added for reference. HW setup configuration for this test: - CPU: Intel Core i7 5820K @ 4GHz core / 3.3GHz uncore - MB: ASRock X99E-ITX/ac - RAM: 2x8GB DDR4-3200 C16 18-18-36 1T - GPU: AMD Radeon R9 Nano (+25% power target) - Drive: 2x 256GB Crucial BX100 @ RAID0 - OS: Windows 10 Pro x64 - Cooling: EKWB custom loop liquid cooling (GPU, CPU) Game settings: Results: CPU core count impact result analysis: - Please allow for +/- 2% result accuracy. The overlapping graphs peak/dips are always a good sign of a reliable data aquisition. - CPU core count effect lessens after 4 cores. It appears the performance peak is reached when utilizing 4 cores. Utilizing more than 4 cores gains to further performance boost - Average time for the simulation when FPS is lower than 60, varies greatly even though the average FPS appears to be high for smooth gameplay when running the game on a single thread (core). 5.2. CPU core frequency impact on FPS The purpose of this test is to test out the impact of the CPU core frequency on the general FPS performance. The frequencies tested range from 2.5- to 4.5GHz, which should cover the sprectrum from the higher-end notebooks to decently overclocked desktop CPU. HW setup configuration for this test: - CPU: Intel Core i7 5820K @ 2.5-4.5GHz core / 3.0-3.3GHz uncore - MB: ASRock X99E-ITX/ac - RAM: 2x8GB DDR4-3200 C16 18-18-36 1T - GPU: AMD Radeon R9 Nano (+25% power target) - Drive: 2x 256GB Crucial BX100 @ RAID0 - OS: Windows 10 Pro x64 - Cooling: EKWB custom loop liquid cooling (GPU, CPU) Game settings: Results: CPU Frequency Impact result analysis: - Please allow for +/- 2% result accuracy. The overlapping graphs peak/dips are always a good sign of a reliable data aquisition. - Performance of the DCS still greatly rely on the CPU frequency and scales almost linearly. - Overclocking proves beneficial and seems to be the great choice to enhance performance of the system w/o additional costs (providing that your cooling system can handle it) - 2.5GHz 'underclock' simulates a higher-end mobile CPU performance (i.e. Haswell or Skylake QH CPU). 5.3 CPU architecture impact on the FPS: The purpose of this test is to test out the impact of the different CPU architecture on the general FPS performance. I have selected the FHD resolution as a starting point, which is in my books considered a 'low' resolution. CPU tests are always done on 'low' resolutions to exclude the possible GPU bottleneck. This was one of the most time consuming tests which I don't want to repeat ever again. I hadn't had the chance to test FX 8350 yet, but hopefully I will be able to do so in the upcoming weeks. I guess it will do around ~ 65FPS on average. Test systems: - CPU: (as listed before) - MB: Gigabyte GA-990FXA-UD7 Gigabyte F2A88XN Wifi ASUS ROG Maximus IV Gene-Z ASRock Z77-Extreme4 ASUS ROG Maximus VII Gene MSI B150 Gaming Night Elf ASUS ROG Maximus VIII Extreme - RAM: 4x4GB DDR3-2133 C10 12-12-28 1T @ 1.35V 2x8GB DDR4-2800 CL15 16-16-34 1T 2x8GB DDR4-2133 CL15 15-15-31 2T - GPU: MSI GeForce GTX 970 Gaming 4G - Drive: 128GB Crucial BX100 - OS: Windows 10 Pro x64 - Cooling: factory cooling on CPU; factory cooling on GPUs - Monitor: Dell 2713HM - Drivers: Nvidia 361.43 Game settings (expect for the resolution which is 1920x1080 FHD): Final results are as following: GPU Architecture Impact result analysis: - Please allow for +/- 2% result accuracy. The overlapping graphs peak/dips are always a good sign of a reliable data aquisition. - Please note that Core i5 6600 setup was running on a non-overclockable B150 motherboard, thus memory ran at DDR4-2133 CL15 15-15-31 2T. Just by running it on a Z170 motherboard and cranking up the memory speed we would see up to 10FPS boost. - Please note that Core i7 6700K setup was running using DDR4-2800 CL15 16-16-34 1T memory settings. - Core i7 6700K stock clock is 4GHz, hence the same results for non-OC and OC test - Broadwell with it's 128MB L4 seem to offer nice boost compared to Haswell. When set to 4GHz (which this sample did with factory voltage) it even surpases Skylake. 5.4. DRAM frequency impact on FPS The purpose of this test is to test out the impact of the DRAM frequency on the general FPS performance. The frequencies tested ranged from DDR4-2133 to -3200, which should cover the majority of available memory kits. HW setup configuration for this test: - CPU: Intel Core i7 5820K @ 4.0GHz core / 3.3GHz uncore - MB: ASRock X99E-ITX/ac - RAM: 2x8GB DDR4 (2133 - 3200) - GPU: AMD Radeon R9 Nano (+25% power target) - Drive: 2x 256GB Crucial BX100 @ RAID0 - OS: Windows 10 Pro x64 - Cooling: EKWB custom loop liquid cooling (GPU, CPU) Game settings: Results: DRAM Frequency Impact result analysis: - Please allow for +/- 2% result accuracy. The overlapping graphs peak/dips are always a good sign of a reliable data aquisition. - DRAM frequency has a bigger impact on performance than anticipated but still far less than CPU core count or - more importantly - CPU frequency. - A somewhat big performance gain is obtained when choosing DDR4-2400 memory over DDR4-2133. - Faster memory (or memory overclocking) should always be considered as long as the price of the faster bin is reasonable. 5.5. Display resolution impact on FPS The purpose of this test is to test out the impact of the display resolution on the general FPS performance. The display resolutions tested ranged from FHD to 4K, which should cover the sprectrum from the entry-level PC screens to latest 4K displays. I did a brief study with Radeon R9 Nano on my setup to see if what is to be expected when pushing GPUs to draw 8+ megapixels. HW setup: - CPU: Intel Core i7 5820K @ 4.0GHz core / 3.3GHz uncore - MB: ASRock X99E-ITX/ac - RAM: 2x8GB DDR4-3200 C16 18-18-36 1T - GPU: AMD Radeon R9 Nano (+25% power target) - Drive: 2x 256GB Crucial BX100 @ RAID0 - OS: Windows 10 Pro x64 - Cooling: EKWB custom loop liquid cooling (GPU, CPU) HW setup configuration for this test: - Same as above, just tested out on three different resolutions: 1920x1080 (FHD); 2560x1440 (WQHD); 3840x2160 (UHD a.k.a. 4K). Results: Display Resolution Impact result analysis: - Please allow for +/- 2% result accuracy. The overlapping graphs peak/dips are always a good sign of a reliable data aquisition. - Somewhat expected result for a high-end GPU, the R9 Nano (and alikes) really start to shine only at higher resolution. It is an total overkill for the FHD. This GPU (and alikes) is strong enough to drive single 4K monitor resolution with ease.

changelog: 9. Feb 6th 2016: Added CPU architecture results and findings. 8. Jan 16th 2016: Added simulated VR resolution results and findings. 7. Jan 10th 2016: Added GTX 960 and R9 295X2 results. 6. Jan 7th 2016: Added GTX 780 and Fury X GPU results. 5. Jan 4th 2016: Added multi-monitor high-FOV results and findings. 4. Jan 3rd 2016: Added most of the FHD, WQHD and 4K resolution GPU results 3. Jan 2nd 2016: Added first FHD resolution GPU results 2. Jan 1st 2016: Added display resolution impact on the FPS 1. Dec 31st 2015: Initial publish 1. Introduction Hello everybody, my name is Niko and I'm an avid DCS fan since the first LOMAC release. Professionally I am an R&D Chief over at EK Water Blocks (EKWB), a Slovenian liquid cooling gear manufacturer. My current hobbies are mountaineering, alpine climbing and still - to some extent - extreme overclocking and benchmarking. In my best days (early 2010s) I used to be top 20 worldwide in the Hwbot extreme overclocking league and top 10 in the hardware master league. I'm not participating much in the extreme OC anymore but the benchmarking and tweaking spirit is still well present. Instead of extreme overclocking I am now enjoying exploring, learning and building 24/7 stable price/performance and performance/watt setups. Dry and warm winter this year allows me to spend more time data mining as the winter climbing season is still not here. 2. The purpose of this thread I would like to present you some data mining I have been doing whilst preparing my computer for the simpit I am building. I have an access to plethora of GPUs and system platforms as well and I intend to exploit this for the benefit of the DCS community. My goal is to show the bottleneck effect of components (GPU, CPU, RAM) on the performance of the DCS 1.5.x (currently 1.5.2). At the end of the day I hope this thread will be a good buying/upgrading guide for the DCS community. Nothing is more frustrating than buying HW upgrades which yield no performance gain... 3. What is to be tested? This test will focus on the following: 1. CPU core count impact on FPS 2. CPU core frequency impact on FPS (using Intel Haswell-E) 3. CPU core architecture impact on FPS 4. DRAM frequency impact on FPS (using Intel Haswell-E and DDR4) 5. Display resolution impact on FPS (using Radeon R9 Nano) 6. Multi-monitor gaming performance analysis (using GeForce GTX Titan X) 7. VR resolution gaming performance analysis (using GeForce GTX 960 and -980 Ti) 8. GPU architecture impact on FPS Open the 'spoiler' below for explanation: :prop: Hint: If you do not know exactly what kind of hardware is inside your computer I recommend you to get CPUID CPU-Z and TechPowerUp GPU-Z - a must-have tools for any PC user. The following CPU platforms will be tested, preferably at various clockspeeds: - AMD Phenom II X4 955 (Deneb 45nm - 4c/4t) - AMD FX-4100 (Zambezi 32nm - 2M/4t) - AMD FX-8350 (Vishera 32nm - 4M/8t) - AMD FM2+ A10-7700K (Steamroller 28nm - 2M/4t) - Intel Core i5 750 (Lynnfield 40nm - 4c/4t) - Intel Core i5 3570K (Ivy Bridge 22nm - 4c/4t) - Intel Core i5 4670K (Haswell 22nm - 4c/4t) - Intel Core i7 5775C (Broadwell 14nm - 4c/8t) - Intel Core i7 5820K (Haswell-E 22nm - 6c/12t) - Intel Core i5 6600K (Skylake 14nm - 4c/4t) The following DRAM settings will be tested: - DDR4-2133 CL15 15-15-36 1T (representing low cost memory kit) - DDR4-2400 CL16 17-17-40 1T (representing most common value memory kit) - DDR4-2800 CL16 18-18-39 1T (representing gaming oriented memory kit) - DDR4-3200 CL16-18-18-36 1T (reprensenting enthusiast oriented memory kit) The following GPUs will be tested: - Nvidia GeForce GTX 580 - Nvidia GeForce GTX 780 - Nvidia GeForce GTX 780 Ti - Nvidia GeForce GTX 960 - Nvidia GeForce GTX 970 - Nvidia GeForce GTX 980 - Nvidia GeForce GTX 980 Ti - AMD Radeon R9 270X - AMD Radeon R9 280X - AMD Radeon R9 290X/390X - AMD Radeon R9 295X2 - AMD Radeon R9 Nano - AMD Radeon R9 Fury X 4. Testing methodology explained - The performance is tested using using DCS: Flaming Cliffs 3 by running a pre-recorded F-15C flight (replay) on a fixed graphics settings - Pre-recorded F-15C flight is a F-15C quick mission - it consists of BVR with 4 enemy airplanes, merge with remaining MiG-23 and close-combat dogfight (guns and missiles) with MiG-23 until the end of the track. - Different GPUs are tested using the same pre-recorded flight at three different resolutions: FHD, WQHD and 4K - Fraps software will log a session FPS for a total duration of the benchmark run (finite time) - Every test setup is equipped with adequate (16GB) quantity of system DRAM to prevent disk swapping - Each system runs Windows 10 Pro x64 and the latest graphics card drivers - We are not solely focusing on minimum-, maximum- and average FPS but also the percentage of time FPS dipped under certain threshold Image quality as tested (4K): - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - What have we learned so far (tl;dr): Compiled list of lessions learned: