Cinegy Cinescore - free benchmark for video production
|
Cinegy Cinescore - free benchmark for video production
|
Posts: 1,813
Threads: 143 Joined: 2015 Apr Thanks: 465
Given 4794 thank(s) in 1412 post(s) Country: 
2020-04-30, 08:37 PM
Not just me then. That's too bad. I was curious to see if my score improved.
The007Dossier
TheStarWarsTrilogy Donations always welcome: paypal.me/williarob4K77 / BTC: bc1qzr9ejyfpzm9ea2dglfegxzt59tys3uwmj26ytj 
Posts: 7,153
Threads: 601 Joined: 2015 Jan Thanks: 1081
Given 1460 thank(s) in 963 post(s) Country: 
2020-04-30, 10:17 PM
Yep, very curious too. But it was bad they used only CUDA, hence Nvidia, and not something that AMD could use.
Anybody knows some benchmark like this that we could use?!? 
Posts: 14
Threads: 1 Joined: 2020 May Thanks: 1
Given 3 thank(s) in 3 post(s) Country: 
2020-05-09, 10:50 PM
(2020-04-30, 10:17 PM)spoRv Wrote: Yep, very curious too. But it was bad they used only CUDA, hence Nvidia, and not something that AMD could use. Hy there. Just found this forum while working on some AI Upscaling stuff .... Maybe you guys already know (haven`t read a lot here, just starting)- in case not: Blackmagic RAW Speed Test has a similar benchmark which simulates Videoloads on CPU and CUDA. Benchmarking other cards and other frameworls (not only CUDA) strongly depends on the worflows - The Cinegy benchmark was aimed for theyr stuff, the blackmagic one for theyr software. What kind of benchmarks would be interesting in this forum? Is there a common ground of software or algo you guys are using? As far as I´ve read, it`s whatever tool works... In this case - generic benchmarks would work great. On OSX, Metal and OPENCL should be of interest (CUDA died a while ago on MAC but still can be used with older gear), OpenVino Benchmarks on Intel iGPU stuff, CUDA on Nvidia, Vulkan on all... If you give me some examples of what exactly you are looking for, I might be able to mention something.
Posts: 7,153
Threads: 601 Joined: 2015 Jan Thanks: 1081
Given 1460 thank(s) in 963 post(s) Country:
2020-05-10, 05:10 PM
(2020-05-09, 10:50 PM)Sabbelbacke Wrote: What kind of benchmarks would be interesting in this forum? Is there a common ground of software or algo you guys are using? As far as I´ve read, it`s whatever tool works... I guess most of us use Windows, but if there is a benchmark that will work on both Windows and Mac (or even Linux perhaps) it would be great! About GPU, I think there is not a majority of Nvidia or AMD, even if probably more Nvidia as it has bigger market share; again, a benchmark that would work with both would be perfect!
Posts: 14
Threads: 1 Joined: 2020 May Thanks: 1
Given 3 thank(s) in 3 post(s) Country:
2020-05-12, 01:49 PM
Ah ok...
The above mentioned benchmarks all are based on a specific Software/Workflow. So the generall ability of a setup is not taken into account but rather what the system can do with this specific peace of software or codec... This comes in handy if you use exactly this workflow/setup, but its hard to pinpoint generall capabilities. I am new to this forum, so I don´t know how much of the following is common knowledge here, but I´ll toss some stuff into this thread anyways: GPU Acceleration: Depending on what Software you use and how much it is using which part of your GPU with which framework (CUDA, OPENCL, Vulkan, Metal, etc..), different Cards or models better suit your needs... I´ll try to sum up: - GPU Decodiong or encoding of Video footage: Intel, AMD and NVIDIA nowadays all include some logic to decode or encode videostreams, mostly in h264 and h265. MPEG2 or MPEG4 Variants are also sometimes present, but since these are much eaiser to encode, you could simply do this on modern CPUs with very little stress on the CPU. VP9 and other codecs start to be a thing with GPU de/encoding, but are not very important for most workflows nowadays - they might become more interesting for social-media endpoints (Youtube, etc..). - GPU based encoders are in 98% of the cases not "assisting some software encoder" but a discrete logic on the GPU which takes care of the encoding. That`s why the quality and the features are fixed into the chip. No software update can change what the GPU is capable of. - Qualitywise, Software encoders are better at the moment than GPU variants. The latest Intel and NVIDIA h265 implementation are way better than previous versions of the GPUs and can in some cases be on par with x265 "fast". So Intel after "kaby Lake" will do better and RTX based Nvidia Cards are good (supporting B-Frames now). Only Intel offers 10Bit encoding right now. AMD is not really worth mentioning at the moment - quality wise... The main advantage is speed - even compared to very powerfull CPUs, a GPU often can do way faster than realtime (depending on the resolution).. One other advantage is "taking the workload of the CPU"... So having newest Intel iGPU or RTX Nvidia (on PC only, Mac ditched NVIDIA some time ago) can be an alternative, if speed is an issue and quality or efficency can be sacrfificed... - Specialized cards for spezific codecs in the workflow (recent MAC PRO, professional encoding cards) are a different story and might blow up this topic too much  A simple speed test of the GPU encoding speed and quality can be done with handbrake, it´s available for MAC and WIN. You can get a rough estimate on how fast your GPU is doing encoding compared to the CPU you have built in. http://handbrake.fr/ With software encoding, you nowadays run into the bottleneck of not saturating all the cores you have. Threadripper Owners know what I am talking about. In case of testing with handbrake, you can easily spot this behaviour on high-end CPUs when encoding lower resolutions: Not all cores are used, it doesn`t "max out".. The higher resolution you encode, the more cores are used (there are some other tricks, but let´s keep it simple for now). So encoding 8K with x265 better utilitzes all cores than going for SD. In case you would want to know what could be done at max, look up this benchmark for parallel x265 encoding: https://forum.pclab.pl/topic/1184884-x26...e__st__100 In real world applications, you always have not only encoding, but filtering etc... So if you want to know how much your software can benefit from a GPU to take over some of the calculations, you will first want to check out, what is supported by your software. These are most common: OPENCL (GPU or CPU, MAC/WIN/Linux, INTEL IGPU, Xeon PHI, NVIDIA, AMD) CUDA (MAC only up to high sierra, mostyl dead on OSX nowadays, WIN, Linux - Only for NVIDIA GPUs) Metal (MAC Only, NVidia, Intel, AMD Gpus) Vulkan (all GPUs, all OS) Most calculation types here are FP32, so that´s the majority... Some speedups can be acheived with FP16 calculations, which become a thing since the introduction of Tensor Cores in the RTX Cards (actually Volta Nvidia Series introduced them, but not too many people will spend this amont of money in prosumer...). Intel has FP16 capability, but no cards with many cores.. FP64 are something AMD is very good in, enterprise NVIDIA Cards you will find this, too, but most consumer cards from Nvidia after Kepler generation are very weak with this. Most Video realted stuff is dealing with FP32 and FP16 nowadays, so bad FP64 performance is not "bad"... Most of the times it´s hard to pinpoint actuall speed from this rather technical information, so using a gaming benchmark can at least give a rough estimate on the compute power fo the card, giving a hint on what to gain with a certain setup: geekbench ist able to test vulkan, CUDA, OpenCL and is available on several plattforms: https://www.geekbench.com/ they also have a browser with results to look up how other cards are doing: Metal (MAC) Example: https://browser.geekbench.com/metal-benchmarks More for gaming, but giving a rough estimate: https://gfxbench.com/benchmark.jsp Also x-plattform and capable of Vulkan or Metal... I suppose harddisk benchmarks in generall are "easy" and known to everyone, so I´ll leave this out for the moment - i case you want some stuff here, too, I´ll be happy to add some... Just a few general thoughts on storage: - SSDs: These have become favourable for Videostuff due to theyr high transfer speeds and low access times. Not all SSDs are equal, though: SATA are most common and often give the most storage for the money. But they are limited to SATA Interface speed. So in case you are dealing with heavy workflows, you might want to consider a PCIE-version or NVME - watch out which NVME Version you actually have.. If you look around a little, NVME - there are some nvme cards at around the price range of SATA models. - If possible, get one SSD for the system and the software - and one for video editing. - for endstorage, get a big HD - much cheaper and for the final encoding step fast enough. Offloading not used projects is also a usecase for traditional HDs. - With SSDs, check on the sustained write performance - multiple layers in the memnory chips have cut down production costs but slowed down speed quite a lot. Most drives compensate with larger caches and most benchmarks only take typical office workloads into account, which can easily be done within the cache range. So stay away from cheap QLC type drives and check if the drive can hold up high write speeds over a long time (many Gigabytes)... To check real_life workflow with Premiere, this could come in handy: https://www.pugetsystems.com/labs/articl...-Pro-1519/ Blackmagic has a speed test: https://apps.apple.com/de/app/blackmagic...4550?mt=12 For Win, I don´t have a seperate download, but it´s included in the free davinci resolve download. I´ll stop for now... Not to forget to once more stretch: It heavily depends on your workflow, footage and Software, which speeds are relevant (that´s why there is no general benchmark for video editing, it simply makes no sense).... On some systems, choosing the right codec for intermediate files or changing the colour space you work in can make the difference from "unusable" to "rocket speed". It´s always worth spending some time to investigate, whats working "under the hood" instead of simply throwing money at the problem.. |
|
« Next Oldest | Next Newest »
|
