Blackmagic Forum

Hi editors,

I figured it'd be interesting if we all share our .BRAW Blackmagic RAW Speed Test results on our different workstations. This test software provided by Blackmagic Design is launching a lot of decoding of a sample .braw file that is the same for everyone, in different resolutions / compression ratio / CPU vs. GPU. Those benchmarks will help us :

• see differences of performance with a benchmark more adapted to our usual workloads (compared to benchmark online who are not made for video editing but more for gaming)
• identify if something is wrong with our current setup by comparing with others who have similar computers
• be able to know what to expect and this could influence their next computer purchase decision

I'll start with some of my workstations and then describe the methodology to follow so we all follow the same process to try to get more consistent results :

M1 Max, Macbook Pro 16 inch 2021 M1 Max 32 GB (Metal), macOS Sonoma 14.1, BRAW 3.4



Laptop nVidia, Windows 10, RTX 2060 Mobile (CUDA), Intel i7-9750H 2.6 Ghz, 32 GB RAM, NVME SSD, BRAW 3.4


Some benchmarks made by others are already available online, for example https://techgage.com/news/testing-out-8 ... teen-cpus/ they get on BRAW 8K 8:1 (note that the : 34 fps AMD Ryzen 3700X, 50 fps AMD Ryzen 3900X, 39 fps Intel i9-9900K.

Methodology to produce the screenshot to share here :

1. Install latest Blackmagic RAW 3.4 from here https://www.blackmagicdesign.com/suppor ... al-cameras and update your latest GPU Drivers if you can. If you're on a laptop, make sure your battery is full and you're plugged in
2. Restart computer, don't open your usual apps, disconnect internet, on Windows open Blackmagic RAW player, go in the About and make sure you are running version 3.4 then run BRAW Speed Test, on macOS you can directly run BRAW Speed Test and go in the About of the BRAW Speed Test and check it is version 3.4
3. Do not touch any setting on the middle gear above the Start button (keep it at BRAW 8K 12:1, the default value)
4. Click that big red button START !
5. Wait that the whole test suite gets properly benchmarked (we're actually more interested in the results inside the table below !) with all the resolutions
6. Once finished click on the gear icon above the START button to stop the tests and click Save Screenshot (it's better than using your own screenshot tool so we get consistent screenshot images, for example without macOS drop shadow)
7. Post your screenshot here with a description of your workstation (CPU, GPU, Disk Drive, RAM, OS version etc.), if you're using CUDA, Metal or OpenCL to decode on your GPU

Waiting for your results !

Antoine from Autokroma

(Edit : the computer below was not correctly configured, see below posts for a real benchmark)
[b]Desktop nVidia, Windows 11, RTX 4090 (CUDA), Intel i9-14900K 5.6 GHz, 32 GB RAM, NVME SSD, BRAW 3.4, Not Overclocked, I just configured this workstation and it might not run at max perf

Antoine_Autokroma_BlackmagicRAWSpeedTest_RTX 4090 (CUDA), Intel i9-14900K, 32 GB RAM, NVME SSD, BRAW 3.4.png (55.55 KiB) Viewed 20679 times

This is the result on my completely outdated Apple laptop (see sig):

Desktop - Windows 11, RTX 4090, AMD Threadripper 3960X, 128 GB RAM, MVME SSD, No Overclock


RazerBlade 16 Laptop, Windows 11, RTX 4090, Intel i9-13950HX, 32 GB RAM, NVME SSD, No Overclock


M3 Max Macbook Pro 16, 64 GB RAM, 4TB Hard Drive

Thanks for your tests Uli and Eric ! Something's def wrong with my latest Windows workstation, both my CPU and GPU underperform . M3 Max is insane !

Adding another benchmark of a small Windows laptop

Lightweight Laptop, ASUS Zenbook UX363EA 13 inch Laptop Windows 11, Intel i7-1165G7 2.8 GHz, GPU Iris Xe Graphics G7 96EUs (OpenCL), 16 GB RAM, NVME SSD, BRAW 3.4

This thread already is useful, it made me realize how seriously bad my computer was. I've changed my motherboard, now dual channel DDR5 2x32 GB is working, both CPU and GPU perform as they should and this BRAW Speed Test finally gives REAL results, instead of 72 / 103 fps I get 125-130 fps for the CPU and 132-141 fps with CUDA, on par with benchmark results from Eric Sorensen / MrRipfrog (Thanks Eric !) :

Desktop nVidia, Windows 11, RTX 4090 (CUDA), Intel i9-14900K 5.6 GHz, 2x32 GB RAM, NVME SSD, BRAW 3.4, Not Overclocked

Dissapointing result, but improved after enabling resizable BAR.
Intel 8600K, 16GB DDR4, Intel Arc A770 16GB

I’d assume that the ARC is not yet showing its full potential with BM.

Desktop Intel ARC, Windows 10, ARC A750 (OpenCL + Resize Bar), Intel i3-12100F 3.3Ghz, 2x16 GB RAM, NVME SSD, BRAW 3.6, Not Overclocked


Apparently the Ryzen does not agree with Intel Graphics:
Desktop Intel ARC, Windows 10, ARC A750 (OpenCL + Resize Bar), Ryzen 3 3200G 3.6Ghz, 2x16 GB RAM, NVME SSD, BRAW 3.4, Not Overclocked

Edward,

Have you enabled resizable BAR? You Motherboard may or may not support this, but I gained about a 60 to 70 percent jump in GPU performance across the board after I did. It might not help the CPU scores, but I bet it will drastically improve the GPU side of things.

M1 Ultra, MacOS 14.1.2

Monterey Hackintosh I5 9600k 48gb ram Radeon Vega 64 Metal

MSI GS75 Stealth 10SE - Win11 - Intel i7-10750H 2.60GHz - 16GB RAM - NVME SSD - RTX 2060 (CUDA) - BRAW 3.4, Not overclocked

Thanks Nico for the benchmark. Thanks Cpuraider / Paul Eaton for mentioning BAR, actually it's the opposite for me, disabling BAR in my BIOS gives me a boost, once I disable it go from 132-141 fps with CUDA to 162 - 171 fps (see results below). Note that others CPU and GPU benchmarks stay the same whether I have BAR or not. Maybe there's something special about how the .braw frame is uploaded to the GPU for further decoding.

It would be interesting to link results of those tests to performance of DaVinci Resolve / Adobe Premiere Pro with BMD Blackmagic RAW plugin and BRAW Studio

AMD Ryzen 9 5900X, Radeon RX6800 XT, 32 GB RAM, macOS 11.7.10 (hackintosh) & Windows 11

Apparently the Ryzen does not agree with Intel Graphics:
Desktop Intel ARC, Windows 10, ARC A750 (OpenCL + Resize Bar), Ryzen 3 3200G 3.6Ghz, 2x16 GB RAM, NVME SSD, BRAW 3.4, Not Overclocked
[/quote]

It seems that the BMRAW SPEED test is currently utilizing the Ryzen Integrated graphics instead of the Intel GPU. Nevertheless, the Ryzen 3200G is considered a relatively underpowered processor, largely due to its significantly limited memory bandwidth. Temporarily disable the R3 3200G GPU from the Devices List to ensure that the test runs using the correct GPU.

TomsheenPL wrote:AMD Ryzen 9 5900X, Radeon RX6800 XT, 32 GB RAM, macOS 11.7.10 (hackintosh)

Wow it's insane that a RX6800 XT beats my 4090 on a hackintosh ! Would be interesting to know what results you get on Linux or Windows using the same hardware. Maybe the Metal parts of the BRAW SDK are very well optimized also

Hi

My contribution, running Blackmagic RAW Speed Test Ver 3.6.1

Workstation APY: Intel i9 14900K / 64 GB RAM - DDR5 / Nvidia RTX A4000 (Ampere Architecture) / Windows 11 Pro

Just for fun

Of course, we like to see the numbers on the "tachometers"
So there are the results, for the 4 tests respectively: 8K 3:1 / 6K 3:1 / 4K 3:1 / HD 3:1

Workstation APY: Intel i9 14900K / 64 GB RAM - DDR5 / Nvidia RTX A4000 (Ampere Architecture) / Windows 11 Pro

the test #4: HD 3:1

Old Mac Studio M1 Ultra 64GB ram

I dont know what this means, but here it is. I only use this for color grading. My footage is 4k and deliver HD @24/30fps
Specs:
Z390 Desingware CF
i9 9900K@3.6 Ghz
128 GB RAM
Nvidia Quadro p4000. I am seeing those RTX are beasts!!! However, I am really fine with my Quadro for 4k.

So if any of you can give me an insight...but anyways, my workflow is smooth and my rendering times are really acceptable(for me)

Hey ! Thanks for the last batch of screenshots. Helge don't forget to post the full window, otherwise we only get the results for the item you selected in the list (here HD) and we don't know what compression (by default for me it is 8K 8:1 compression). In general the top results cannot be compared between us unless we're sure to all use the same resolution and compression ratio. Point 6) in the guide for the benchmarks : "Once finished click on the gear icon above the START button to stop the tests and click Save Screenshot (it's better than using your own screenshot tool so we get consistent screenshot images"

360VVSpain wrote:I dont know what this means, but here it is. I only use this for color grading. My footage is 4k and deliver HD @24/30fps
Specs:
Z390 Desingware CF
i9 9900K@3.6 Ghz
128 GB RAM
Nvidia Quadro p4000. I am seeing those RTX are beasts!!! However, I am really fine with my Quadro for 4k.

So if any of you can give me an insight...but anyways, my workflow is smooth and my rendering times are really acceptable(for me)

Your benchmarks seems a bit faster than the ones from Paul Eaton (Nov 29, 2023 5:53 am) - Intel 8600K, 16GB DDR4, Intel Arc A770 16GB, I'd say it's quite good given your GPU is 5 years older even taking into account the price was double. Maybe you can try to turn on or off the BAR setting in your BIOS to see if you can get some speed up like me.

Tested on the three I have. Old editor now used for email WEB etc , Threadripper and Studio Max.

3770K +1080Ti



Threadripper + 4070Ti



Studio Max M1




Have all the other resolution if anyone is interested.

First time chiming in here.
Core i9-13900F, RTX 4070 Super, 128GB DDR5 5200 GT/s, Samsung 990 Pro SSD

Windows 11 pro Version 23h2.
Rog Maximus z90 Hero
Intel 14700K
128GM RAM
RTX 4070 TI Super 16GB

Lenovo Legion 5 15ACH6H
AMD Ryzen 5 5600H
RTX 3070 Laptop
2x8GB DDR4 RAM 3600 MHz
512GB NVMe SKHynix
Windows 11

Im struggeling with the Multicam Viewer, it gets stuttery when only 4x 4K Braw Videos are visible, even though CPU, GPU, RAM, Power-draw and SSD-Drive are not really fully utilized. Its also not a thermal issue. HWInfo blames it for the GPU on reliable voltage limit, even though on stress tests of both cpu and gpu at the same time (with FurMark), it shows power limit instead (and has much higher power draw). Maybe someone has a guess where the bottleneck might be?

Here's mine !

AMD Threadripper TRX50 - CPU 7965WX - 64GB RAM - GPU RTX 3060 - No Overclocking
BlackmagicRAWSpeedTest.png[/attachment]

Disk Bench - Samsung SSD 990 Pro - No RAID
[attachment=1]

2017 Apple iMac Pro Xeon W1 10 core 64gb RAM
DaVinci Resolve Studio 18.6.6 and MacOS Sonoma
2x AMD 8gb vram Vega 56 gpu (1 internal and 1 external)
1 Tb Internal SSD
RAID 0 quad NVMe 12Tb External project and work drive (connected via TB3)
2 Tb NVMe External Cache drive (connected via TB3)
24 Tb data drive (backup DAS connected via USB 3.2 Gen 2)

Disk Speed (external drive):


RAW Speed (8K):


RAW Speed (4K):


The largest file that I’ve handled so far is 6K BRAW and delivery are usually in 4K DCI or 2K DCI ProRes so this has been a pretty good workhorse for my needs.

Macbook Pro M4 Max - 64 Gig Memory

Insane results and on battery.

KRDodds wrote:Macbook Pro M4 Max - 64 Gig Memory

Insane results and on battery.

I wonder if the Mac Mini M4 Pro with 64gb memory will yield the same results.

Johannes Jonsson wrote:Windows 11 pro Version 23h2.
Rog Maximus z90 Hero
Intel 14700K
128GM RAM
RTX 4070 TI Super 16GB

I had to do a double take but noticed they were HD numbers. How about 8K numbers?

Ellory Yu wrote:

Johannes Jonsson wrote:Windows 11 pro Version 23h2.
Rog Maximus z90 Hero
Intel 14700K
128GM RAM
RTX 4070 TI Super 16GB

I had to do a double take but noticed they were HD numbers. How about 8K numbers?

Yeah sorry, I used what were there before, all the numbers for other resolutions are shown in the list below

Its amazing how things have changed sins not so many years ago.
This is single Samsung 990 PRO 4TB drive used as project drive, but still backups are into mechanical mirrored backup drives thought and second backup on external mechanical drives.

MacBook Pro 14" M4 Pro (12 core/16 core) 24GB Memory
Mac OS Sequoia

Win11 version
Hardware in the sig
No resizable bar

Mac Sonoma version
Same harware as in signature
No resizeable bar

And here is linux Fedora
same hardware but linux is not so great with AMD gpus.

driedeker wrote:And here is linux Fedora
same hardware but linux is not so great with AMD gpus.

Thank you for this I was going crazy trying different things thinking it was something wrong with my install/config, I checked Raw Speed in Linux and it showed 70fps... not great for 7900xtx I guess but should be fine right?? I saw your comment about Linux being slow and thought I would investigate further.

Magic mask is very slow in Linux using 7900xtx on ROCM (tried different versions on different OS), best was ROCM 6.3.2 getting 5fps on a simple mask. Scrubbing 4k@120fps footage with any LUT/effect/color required proxy no matter the effect otherwise FPS would go unstable like (31/30) and would get sound distortion/lag.

This I know is terrible because I had a 3090 previously and it was way faster/smoother than this and the 7900xtx should be at least as good according to Puget. So for sanity I tried Windows 11 and suddenly got 27fps for same project in magic mask!! Scrubbing timeline is smooth with any effect/text etc. Ok great... check Raw Speed test... 70fps!! So I guess this Raw Speed test is useless at least for AMD.

I am guessing this is a problem with Davinci since everything else (AI/Games etc run great).



TL;DR: Linux + AMD GPU is very slow for Davinci Resolve.

Linux1: Ubuntu 24.04: Gnome 47, ROCM (6.2.3 & 6.3.2 tested) Official AMD driver
Linux2: Gentoo: Gnome47/KDE6.3, Kernel 6.13.2, Mesa 24.3.4, ROCM (6.2.3 & 6.3.2 tested) LLVM19
Windows: 11 24H2 Amd Adren 24.12.1

CPU: 9950X Delid / Tuned PBO&CO
GPU: Sapphire Nitro+ 7900XTX PTM/UTP-8
RAM: GSkill Royal DDR5 @ 6000MT 30-36-36 manually tuned
Motherboard: ASUS X870E ProArt Creator Chipset/VRM pad replaced with UTP-8
SSD: 4 x Kingston KC3000 4TB

ProArt Px 13
AMD Ryzen AI 9 370 HX
Nvidia Geforce RTX 4060