AMD Radeon Pro W6600 Review: RDNA 2 Pro-Vis GPU For Less
How We Configured Our Test Systems: We tested the professional graphics cards represented in this article on a MSI X570 Godlike motherboard, equipped with a Ryzen 9 5950X and 16GB of G.SKILL DDR4 RAM clocked at 3,200MHz. The first thing we did when configuring the test system was enter the UEFI and set all values to their "high performance" defaults, then we disabled any integrated peripherals that wouldn't be put to use, and enabled Resizable BAR support. The memory's clock was dialed in to its optimal performance settings using its XMP profile and the solid state drive was then formatted and Windows 10 Professional x64 was installed and fully updated. When the Windows installation was complete, we installed all of the drivers, applications and benchmark tools necessary to complete our tests.
Out Test System Configuration:
| Hardware Used: AMD Ryzen 9 5950X (3.4GHz - 4.9GHz, 16-Core) MSI X570 Godlike (AMD X570 Chipset) 16GB G.SKILL DDR4-3200 Samsung SSD 970 EVO Integrated Audio Integrated Network NVIDIA Quadro RTX 4000 AMD Radeon Pro WX 8200 AMD Radeon Pro W5700 AMD Radeon Pro W6800 AMD Radeon Pro W6600 |
Relevant Software: Windows 10 Pro x64 AMD Radeon Pro v21.Q1 NVIDIA Quadro Drivers v466.11 Benchmarks Used: SPECviewperf 2020 LuxMark v4 Blender v2.92 SiSoft SANDRA 2021 VRMark Blackmagic RAW Speed Test IndigoBench 3DMark DXR Feature Test |
SiSoft SANDRA 20201 GPU Benchmarks
ANDRA's GPGPU Image Processing benchmark runs through an array of filters on its reference data and offers up an aggregate score, derived from a multitude of individual results. Its GPGPU Cryptography benchmark churns through an assortment of workloads, and presents individual results for overall bandwidth, AES256 encryption and decryption, and SHA2-256 hashing bandwidth. CUDA and OpenCL code paths are available in these tests, but we used OpenCL on all cards. Previously, using the CUDA path with NVIDIA GPUs resulted in better performance, but OpenCL actually outperforms CUDA in the latest versions of this test.
The new Radeon Pro W6600 put up relatively strong numbers in SANDRA's Image Processing benchmark, outrunning all of the previous-gen cards we had on-hand and available for testing. It nearly doubled the performance of the Quadro RTX 4000, but fell way short of the more powerful Radeon Pro W6800.
SANDRA's various Cryptography-related benchmarks had the Radeon Pro W6600 landing about in the middle of the pack, typically a few percentage points ahead of the previous-gen Radeon Pro W5700 and falling behind the Quadro RTX 4000 only in terms of hashing bandwidth.
LuxMark v4.0 Benchmarks
LuxMark is a cross-platform, OpenCL-accelerated 3D rendering benchmark. It's a tool based on the open source LuxRender physically-based spectral rendering engine, which accurately models the transportation of light and supports high dynamic range. LuxRender features a number of material types to allow rendering of photo-realistic and artistic scenes. LuxRender is free software, licensed under the GPL, that offers plugins for packages like Blender, Maya, Cinema 4D and 3DS Max.
The Radeon Pro W6600 manages to outpace only the Radeon Pro WX 8200 here, and even then only with the more-taxing 'Food' scene. When testing with the 'Hall Bench' scene, the Radeon Pro W6600 trails all of the other cards we tested.
IndigoBench Rendering Benchmarks
IndigoBench is based on Indigo 4's advanced rendering engine and offers both CPU and GPU rendering modes for its two built-in models. The standalone benchmark is available for Windows, MacOS, and Linux and outputs results in M/Samples per second.
The Radeon Pro W6600 finishes just behind the Quadro RTX 4000 in IndigoBench, but well ahead of the previous-gen Radeon Pro cards. Once again though, the high-end Radeon Pro W6800 holds onto a commanding lead.
Blender v2.92 Rendering Benchmarks
lender is a free and open source 3D creation suite that can handle everything from modeling, rigging, animation, simulation, rendering, compositing and motion tracking, even video editing and game creation. It has a built-in benchmarking tool that will track the time it takes to complete rendering a particular model. We used a GPU-focused BMW model for these tests here...
