Click here to Skip to main content
65,938 articles
CodeProject is changing. Read more.
Articles / nvidia

Serenity - New Build Time - Part 2 - The Benchmarks

5.00/5 (4 votes)
28 Aug 2018CPOL5 min read 7.8K  
The pursuit of Serenity, it's new build time!

Introduction

This article is Part 2 of the Serenity build. In this part we will take a look at benchmarks for the system.

If you missed the first part you can find it at Serenity - New Build Time - Part 1 - The Build

Benchmarks Used

For the benchmarks I will only be using the free versions of any benchmarks. This allows anyone else to grab a copy and bench their own systems to see how they compare without having to fork out money to do so.

The operating system will be a fully up to date (all windows updates) for Windows 10 Pro, Version 1803.

The system display resolution settings are 2560x1600.

The benchmarks used are as follows;

CrystalDiskMark, Version 6.0.1 x64, Available from Windows Store

PCMark 10 Basic Edition, Available from UL Benchmarks

3D Mark Basic Edition, Available from UL Benchmarks

Cinebench R15, Version R15.038_RC184115, Available from Maxon

FurMark, Version 1.20.1.0, Available from Geeks3D

RealBench, Version 2.56 , Available from Asus ROG

CrystalDiskMark

This benchmark was run against the only storage drive in the system, the primary OS drive, C: and was on the 1TB Samsung 970 Evo M.2 NVMe SSD.

As you can see in the image below peak speeds obtained where 3284.7 MB/s Read and 2519.6 MB/s Write.

Image 1

PCMark 10 Basic Edition

PCMark 10 is a general benchmark, covering video conferencing, office applications, web browsing as well as some rendering and visualisation.

PCMark 10 was the first benchmark I ran on the system, but I also had some issues with it. When Windows was first installed, I installed the latest Nvidia drivers 398.82, and managed to get 3 good runs of the benchmark. I then updated all the system drivers etc. to see what difference that would make. After doing this however, PCMark would not longer run fully. It crashed the system 4 seconds into the rendering benchamrk, no blue screen or errors just straight to power cycle. No errors in the event log or anything. This made it very hard to diagnose what was going on. I tried various uninstalls, reinstalls, display driver removal using DDU. Evenutally after doing a DDU in safe mode, then installing the driver NVidia 398.11 drivers (which are the official supported version), the benchmark would now complete. I also noticed a signifiant improvement in the scores.

The score went from 5324 to 6433, which is a 20% increase.

Image 2

Image 3

Full result details for these runs are at:

https://www.3dmark.com/pcm10b/328057 and https://www.3dmark.com/pcm10b/338662

3DMark Basic Edition

3DMark is targetted towards the gaming side of things. It actually has a number of different benchmark routines that are run individually.

Time Spy

Time Spy targets DirectX12 for gaming, in this test it gave me a score of 16,818, better than 99% of all results.

Image 4

Full result details at https://www.3dmark.com/spy/4294594

Fire Strike

Fire Strike targets high performance gaming PCs. In this test it gave me a score of 28,827 better than 99% of all results. 

Image 5

Full result details at https://www.3dmark.com/fs/16222394

Sky Diver

Sky Diver targets gaming laptops and mid-range PCs. In this test it gave me a score of 63,016 better than 99% of all results.

Image 6

Full result details at https://www.3dmark.com/sd/5211694

Cloud Gate

Cloud Gate targets basic notebooks and home PCs. In this test it gave me a score of 50,546 better than 99% of all results.

Image 7

Full result details at https://www.3dmark.com/cg/4344298

Cinebench R15

Cinebench is an intenseive CPU and GPU test. The CPU test is a render of approx 2000 objects and the GPU test is a car chase scene rendering complex scene and textures.

Image 8

Image 9

The results for these tests were CPU: 2524cb and GPU: 151.57 fps.

Image 10

FurMark

FurMark is an intestive GPU/OpenGL stress test/bench. Running at the different preset resolutions;

1280x720: Score 25904, 431 FPS avg.

1920x1080: Score 17964, 300 FPS avg.

2560x1440: Score 12580, 210 FPS avg.

To test the effectiveness of the cooling system, I also ran a 10 minute stress test at native 2560x1600 and achieved an average 192 FPS and the max GPU temperature was steady at 55'C for both GPUs. This was with an ambient room termperature of 27'C. I can probably drop the temperature by adjusting the fan curves in the Bios, but this is still very good considering the original air coolers would have meant the GPUs would have been temperature limiting around the 84'C.

Image 11

RealBench

RealBench is a multi-targeting benchmark, it includes GIMP Image Editing, Handbrake video encoding, Luxmark rendering and heavy multi-tasking. The results from the benchmark are shown below with an overal system score of 199,506

Image Editing: 169061; Time: 31.5152

Encoding: 240692 ; Time: 22.1362

OpenCL: 233987 ; KSamples/sec:  43068

Heavy Multitasking: 154287; Time: 49.466

System Score: 199506

Image 12

I'm sure even with a small GPU overclock, this would breakthrough the 200k mark. Comparing to similar CPU/GPU combos the base results are in line with expectations.

Points of Interest

Overall, the performance is very good for the machine. I haven't even looked at any overclocking potenital or fully optimizing the fan profiles to see if there is any improvement to be had there. Although, given that the temperature are all within the thermal limits of the devices, there isn't really much to worry about.

One last thing....power usage. The highest power use seen during all the testing was 691.4W at the wall socket, this was during the FurMark stress test.

I then pushed the system to the limit using FurMark. Running a 24 Thread CPU Stress and a GPU Stress at native 21560x1600 and although it run for a while, although a bit jerky at around 100FPS, peak power was 792w and eventually the system power-cycled. I dare say at this sustained load, the PSU is probably at its limits for stability.

 

Hope you enjoyed this 2 part build log and benchmarking, I certainly enjoyed the build. Although the early issues with PCMark and fact the system was failing towards the end of the benchmark made it quite tedious to try and fix. Having to go through the whole benchmark to see if that one change had resolved things or not, then rinse and repeat!

Well, that's all for the time being. If I do any more tweaks or any other benchmarks I'll try and remember to update here.

History

  • 28th August 2018 - Added RealBench scores.
  • 27th August 2018 - First published.

License

This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)