AMD Radeon RX 6700 XT 12 GB to $479 from March 18th.

render glitches on amd 5700.. had to switch over to 2060 super of Nvidia.. at least no glitches, but GPU memory full errors galore...

Did the prevalence of glitches increase with the complexity of the video, such as complex Fusion compositions - or did the glitches happen irrespective of the simplicity or complexity of the task?

irrespective. i never use fusion. just simple luts and colorgrading on one, 2 nodes

Source:

Too bad he used an Intel system. I'm curious to see how the smart memory access feature affects Resolve performance, and that requires a compatible AMD build.

Another Davinci Resolve bench here:



Source:

Too bad he used an Intel system. I'm curious to see how the smart memory access feature affects Resolve performance, and that requires a compatible AMD build.

I have a feeling that we'll be seeing more and more AMD systems now that Zen3 is out, and exceeds Intel's best processors' single-threaded performance, never mind the higher core counts.

Adobe just updated their suite to work better with the new AMD cards and their APU's.
Phoronix show that the compute performance in OpenCL is impressive compared to the Radeon VII, the VII has a lot more memory bandwith.
AMD also released new drivers today...

Some more benchmarks for Davinci Resolve:
https://www.computerbase.de/2020-11/amd ... xt-test/5/

Some more benchmarks for Davinci Resolve:
https://www.computerbase.de/2020-11/amd ... xt-test/5/

Hi.

We don't all read German, so here is a Google translate of the above link:

https://translate.google.com/translate? ... est%2F5%2F

Regards Carsten.

I still haven't found any information about supporting HEVC 10-Bit 4:2:2 footage yet. Sadly the new RTX 3000 series don't support the codec! Maybe it's time to switch to Apple M1!?

Too bad he used an Intel system. I'm curious to see how the smart memory access feature affects Resolve performance, and that requires a compatible AMD build.

There are reports that SMA is nothing but resizeable BAR - support for which is already available on Linux. Nvidia is working on bringing this to Windows platform for both Intel and AMD any generation Ryzens the last I heard:

https://www.techpowerup.com/274589/nvidia-is-working-on-technology-similar-to-amds-smart-access-memory

Some more benchmarks for Davinci Resolve:
https://www.computerbase.de/2020-11/amd ... xt-test/5/

Hi.

We don't all read German, so here is a Google translate of the above link:

https://translate.google.com/translate? ... est%2F5%2F

Regards Carsten.

So a pure and transparent GPU speed benchmark show that 6800XT mostly outperforms the RTX 3090 in Resolve.

Non-transparent Puget Benchmark Suite shows a very different picture.

Very interesting results. Seems that both nVidia cards and AMD ones have specific strengths and weaknesses that are expressed in different situations. AMD gets their ass handed to them in Blender but Resolve noise reduction is the other way around. Interesting.

Seems like tensor cores and OpenCL make the difference in Blender when Optix is off. Probably Resolve operations that use tensor cores are still going to be faster on RTX 3xxx. AMD just needs to re-consider the pricing and they have a winner. I wouldn't blame them for generally being the underdog in GPUs, they're takin on Intel and Nvidia all by themsleves and by being close they keep the prices lower.

Very interesting results. Seems that both nVidia cards and AMD ones have specific strengths and weaknesses that are expressed in different situations. AMD gets their ass handed to them in Blender but Resolve noise reduction is the other way around. Interesting.

BlackmagiceDesign really should launch it's own benchmark for Resolve like they did with blackmagic RAW speedtest. The candle benchmark is old and the puget benchmarks https://www.pugetsystems.com/labs/articles/Puget-Systems-DaVinci-Resolve-Studio-Benchmark-1523/#UpdateLog focuses to much on 60 fps(not really a thing in higher end productions), crappy hardware h265/264 encoding, no: BRAW, Sony X-OCN, ARRI-RAW, low compression REDRAW, it's a benchmark for the casual youtube content creator.

Optix and Cuda a way more mature than ROCm at the moment, AMD is catching up (together with Xilinx) https://www.phoronix.com/scan.php?page=news_item&px=AMD-ROCm-Xilinx

Some more benchmarks for Davinci Resolve:
https://www.computerbase.de/2020-11/amd ... xt-test/5/

Hi.

We don't all read German, so here is a Google translate of the above link:

https://translate.google.com/translate? ... est%2F5%2F

Regards Carsten.

Interesting that despite the low memory bandwidth on paper, the TNR performance is better than the 3090. Looks like I may have been right that SAM is good for this use case.

Seems like tensor cores and OpenCL make the difference in Blender when Optix is off. Probably Resolve operations that use tensor cores are still going to be faster on RTX 3xxx. AMD just needs to re-consider the pricing and they have a winner. I wouldn't blame them for generally being the underdog in GPUs, they're takin on Intel and Nvidia all by themsleves and by being close they keep the prices lower.

Afaik Tensor cores don’t do anything in Cycles render engine itself which most of the renders test, they are, if at all, used for AI denoiser.

Seems like tensor cores and OpenCL make the difference in Blender when Optix is off. Probably Resolve operations that use tensor cores are still going to be faster on RTX 3xxx. AMD just needs to re-consider the pricing and they have a winner. I wouldn't blame them for generally being the underdog in GPUs, they're takin on Intel and Nvidia all by themsleves and by being close they keep the prices lower.

Afaik Tensor cores don’t do anything in Cycles render engine itself which most of the renders test, they are, if at all, used for AI denoiser.

Yes, just found out about it. So it's either just lack of grunt or OpenCL vs CUDA.

According to Nvidia blurb for Studio drivers https://www.geforce.com/en_US/gfecnt/whats-new/articles/cinema-4d-nvidia-studio-driver

Featuring new AI-powered features to make video editing and color grading easier and more powerful, Blackmagic Design’s DaVinci Resolve Studio 16 makes use of several NVIDIA AI libraries, and the Tensor Cores found on the latest NVIDIA RTX GPUs, to accelerate inferencing. RTX-powered AI features include:

Speed Warp, which interpolates frames when changing the playback speed of videos resulting in smooth slow-motion with fewer artifacts
Super Scale, to increase footage resolution by up to 4x
Auto color and shotmatching to accelerate the color grading workflow
Facial recognition for automatic tagging of clips for smart bins and tracking of facial features easy feature refinement
Stylize for creating artistic looks on video footage
“DaVinci Resolve 16 Studio uses the latest multiple GPU innovations for AI and deep learning,” said Dan May, Blackmagic Design’s USA president. “With the new DaVinci Neural Engine using NVIDIA CUDA 10 and TensorFlow acceleration, our tests show a better than 2x performance increase over previous NVIDIA GPUs. These same GPUs are also used for decoding and debayering Blackmagic RAW images which makes them an attractive investment.”

It will be interesting to benchmark AMD vs Nvidia for these specific tasks like speed-warp, superscale and stylize (not sure if Auto-Color/shot-matching can be benchmarked inside Resolve). I suppose the new magic mask feature in DVR17 will also use Tensor cores and AI. Auto-Color/shot-matching/Facial recognition would be an edit/color tab compute load not export because these activities would be tweaked and fine-tuned at the edit stage so dunno how one can benchmark these with an export render... But then Facial recognition works well with OpenCL/AMD so I wouldnt worry about that...

Hi.

Normally is Wccftech a site to read early rumors. But today they publish a nice 'AMD Radeon RX 6800 XT & RX 6800 Custom Model Roundup, ASUS, MSI, Gigabyte, PowerColor, Sapphire, XFX, ASRock & More'
Here is a link:

https://wccftech.com/amd-radeon-rx-6800 ... fx-asrock/

Regards Carsten.

Any comments on these charts with Resolve Standard Candle Benchmark, courtesy of folks from LiftGammaGain:

https://datastudio.google.com/u/0/repor ... page/v8kCB

RX6800/6800XT seem to do veeery well with TNR, beating even the RTX3090?

Has any one actually been able to test an RDNA2 card with resolve? I am wondering if BMD still need to add PCIE IDs or any thing else in before they will work.

I wonder how much that huge cache (or any cache for that matter) actually affects Resolve type of processing where most of the "action" is related to loading image data into gpu and then throwing it away. I think all the cuda/opencl kernels are loaded into gpu memory and not discarded, so what kind of data would actually benefit in this context from the 128MB cache? Noise reduction and other temporal effects maybe...?

Where I think the cache might be useful is some thing like denoise where it needs to examine more than one frame at a time. There is also the fact that for many codecs a frame doesn't hold all the infomation for a frame rather just the changes from the last one.

I guess the developers at BMD will need to move from their CUDA skillset and expand it to the AMD set of technologies in order to optimize it for the RDNA GPUs.

Funny how the system requirements for Resolve include cross platform openCL or Cuda, not just Cuda don't you think? With Apples new platform and Intel finally getting serious about compute Cuda will continue to become less relevant.

Besides the shortage of units the biggest let down so far is that LTT is reporting that RDNA2 didn't bring any improvements to AMD's hardware encoding. I was really expecting them to up thier game on that front.

I still haven't found any information about supporting HEVC 10-Bit 4:2:2 footage yet. Sadly the new RTX 3000 series don't support the codec! Maybe it's time to switch to Apple M1!?

I think - Just because there is no h/w support does not mean it wont run - maybe 10% extra cpu load? Unless your system is already weak I don't think we should worry...

I wonder how much that huge cache (or any cache for that matter) actually affects Resolve type of processing where most of the "action" is related to loading image data into gpu and then throwing it away. I think all the cuda/opencl kernels are loaded into gpu memory and not discarded, so what kind of data would actually benefit in this context from the 128MB cache? Noise reduction and other temporal effects maybe...?

Where I think the cache might be useful is some thing like denoise where it needs to examine more than one frame at a time. There is also the fact that for many codecs a frame doesn't hold all the infomation for a frame rather just the changes from the last one.

This kind of partial frames are not something Resolve engine itself operates on. Once frame exits decoder, what the source format was becomes irrelevant, all sources are expanded to full frame full float RGBA data.

Puget Systems - Davinci Resolve: https://www.pugetsystems.com/labs/artic ... ance-1990/

Hi.

Here are links to the first AMD Radeon RX 6900 XT Reviews:

https://www.tomshardware.com/news/amd-r ... -xt-review

https://www.anandtech.com/show/16303/la ... rx-6900-xt

Regards Carsten.

Hi.

Quote: 'AMD will reveal a new Radeon RX 6000 graphics card during Episode 3 of its "Where Gaming Begins" event on March 3 on 11 AM US Eastern.'

Here is a link with pictures:

https://www.tomshardware.com/news/rx-67 ... ts-3690746

Regards Carsten.

If you are running Linux AMD has been making some very troubling announcements this week about supporting graphics with ROCm. Be aware.

https://www.phoronix.com/scan.php?page= ... Cm-Non-GUI

Kind of makes me wish Resolve would use OpenGL compute shaders instead of OpenCL.

For image processing you can pretty much do the same. All the hw video decoder can directly export to OpenGL texture too. And you can sometime even get better performance than CL/GL interop because when staying in the OpenGL domain, you can synchonize operation much better. (with CL/GL interop when handing texture from one domain to the other, you need to flush the video command pipeline).

Oh and also obviously removes the need for RoCM / AMDGPU-PRO and all that stuff ...

If you are running Linux AMD has been making some very troubling announcements this week about supporting graphics with ROCm. Be aware.

Hi.

One year ago AMD announced their new dedicated GPU architecture for data center compute, called CDNA. So CDNA and RDNA was becoming their own independent architectures. Here is a link:

https://www.anandtech.com/show/15593/am ... ta-centers

What AMD have done is updating some ROCm Documentation covering the the AMD Instinct MI100 accelerator for Datacenters.

I can't se why it can be a 'very troubling announcements' or why Resolve users should 'Be aware', as Resolve is designed to run on RDNA graphics cards, and not headless on the CDNA platform, according to this from AMD. Even on Linux:

Quote: 'Note: The AMD ROCm™ open software platform is a compute stack for headless system deployments. GUI-based software applications are currently not supported.'

From: https://rocmdocs.amd.com/en/latest/

Yes, AMD updated some documentation, but for an another line of cards, designed to run headless in Datacenters.

Regards Carsten.

I guess you missed this part :

"""
The Radeon Software for Linux packaged driver currently ships a ROCm-based OpenCL compute driver for Navi 2x graphics processors
"""

So the new Radeon RX 6000 series cards driver, from the amdgpu-pro package have an OpenCL driver based on ROCm ... and if that doesn't support CL/GL interop for instance (being a "GUI" feature), then resolve won't run.

Hi.

AMD Radeon RX 6700 XT 12 GB vRam to $479 from March 18th.

https://www.tomshardware.com/news/amd-r ... e-revealed

I watched the online launch. You can buy the AMD branded card from Amd.com to $479 from March 18th.
And the Partner Cards from the normal channels.

Regards Carsten.

Hi.

Today is time for the AMD Radeon RX 6700 XT Reviews. Here are some links:

'AMD Radeon RX 6700 XT Review: Big Navi Goes on a Diet:'

https://www.tomshardware.com/reviews/am ... -xt-review

https://wccftech.com/review/amd-radeon- ... on-review/

And

https://wccftech.com/review/msi-radeon- ... rd-review/

Regards Carsten.