Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Radeon Pro SSG

DJ Meyer

DJ Meyer

Well-known member
Joined
Dec 22, 2008
Messages
1,070
Reaction score
47
Points
48
The unicorn has landed. Tests will ensue.

mpxlbz756ysves1yb2y4jds7u34iorji.jpg
 
Looking forward to see the added value compared to the VEGA FE + 4 x 960Pro in raid-0.
 
Šabović Adis;1787049 said:
8K R3D @full debayer?
Click to expand...

The claim made at Siggraph was that the Radeon Pro SSG can do this on a lowly CPU spec system - presumably due to some sort of Vega + SSG caching secret sauce. Hard to believe on paper, but that is why we test.
 
DJ Meyer said:
The claim made at Siggraph was that the Radeon Pro SSG can do this on a lowly CPU spec system - presumably due to some sort of Vega + SSG caching secret sauce. Hard to believe on paper, but that is why we test.
Click to expand...

Yeah, Siggraph...that's where I saw "the other brand" crash while trying to load a project.
 
Šabović Adis;1787057 said:
That's cool!
Um, is 4x512GB on PCIe comparable to 2TB "on board"? With Highpoint You could go as high as 4TB, or 8TB, if I'm not mistaken.
Click to expand...

We ran out of money and the SSD's are formated at 400 GB each so only 1.6 TB left.
 
Šabović Adis;1787068 said:
Yeah, Siggraph...that's where I saw "the other brand" crash while trying to load a project.
Click to expand...

Ha yeah i remember that. The whole presentation was pretty cringeworthy. Lots of people not used to being on stage.
 
Hey guys, quick sort-of update.

I have been holding off on updating this thread as the SSG has been more or less useless since I received one in February. Applications must explicitly support the SSG API to be able to leverage the NVME cache, and there are currently zero publicly available applications that I am aware of that do so. Adobe announced that the latest premiere version does, however in my testing it doesn't work. I've been told that yet another version of Premiere is imminent that will fix things, but I guess we'll have to see.

So at the current time, without any vendor support, you basically have a wx9100 workstation card with a 2TB NVME volume bolted on and no synergy between the two to speak of. .r3d performance is as you'd expect - roughly in line with a GeForce 1070.
 
DJ Meyer said:
So at the current time, without any vendor support, you basically have a wx9100 workstation card with a 2TB NVME volume bolted on and no synergy between the two to speak of. .r3d performance is as you'd expect - roughly in line with a GeForce 1070.
Click to expand...

So it is a lot slower than a VEGA 56 with VEGA 64 firmware because that one is on GTX1080ti speed, our VEGA FE's run at Titan V speeds in Resolve.
 
We did manage to get it to at least do something.

In Premiere, persistent GPU Preview Cache (SSG volume) is only utilized when you render the clips. So import, create a sequence with clip, hit render, wait, then you can play back. If you make any changes to the clip source settings or effects, you get to render and wait again. No automatic .r3d caching, etc. It also throws errors pretty consistently.

So I have no idea what they were showing on stage at Siggraph last August, but it wasn't this.
 
From what I remember from Siggraph is that it is more focussed on CGI.
 
Misha Engel said:
So it is a lot slower than a VEGA 56 with VEGA 64 firmware because that one is on GTX1080ti speed, our VEGA FE's run at Titan V speeds in Resolve.
Click to expand...

Vega FE is not Titan V speeds. More like 1070 at best. I have a 1080Ti and tested a Vega FE just last week. The 1800Ti blew away the Vega FE in Resolve, and was also faster in Premiere and FCPX. There's a few things the Vega FE was almost as fast as a 1080Ti at, but most things I threw at it the 1080Ti was 50-100% faster.
 
Mike Halper said:
Vega FE is not Titan V speeds.
Click to expand...

Our's are (water cooled, undervolted, overclocked 1600 MHz).
The good thing about Davinci Resolve is, that it is not a game, it just needs RAW compute power (FP32) and only that.

Happy gaming with the Geforce cards, I just love the OpenGL overlay at 10-bit color depth and the 16 GB of HBM2.

Vega FE does 11.321 TFlops at 1382 MHz(Base Clock) and 13.1 TFlops at 1600 MHz(ours).
GTX1080ti does 10.6 TFlops at 1480 MHz(Base Clock) 50% faster than Vega FE base clock makes 11.321 x 1.5 ~ 17 TFlops ~ 2370 MHz (That is a hell of an over clock).

Titan V does 12.288 Tflops at 1200 MHz(Base Clock).

https://www.slashcam.de/artikel/Test/Auf-Augenhoehe-mit--Ti-tanen---AMD-Vega-56-unter-DaVinci-Resolve-14--Was-bringts--63--Vega56-unter-Resolve-14.html#Was_b
 
Misha Engel said:
Our's are (water cooled, undervolted, overclocked 1600 MHz).
The good thing about Davinci Resolve is, that it is not a game, it just needs RAW compute power (FP32) and only that.

Happy gaming with the Geforce cards, I just love the OpenGL overlay at 10-bit color depth and the 16 GB of HBM2.

Vega FE does 11.321 TFlops at 1382 MHz(Base Clock) and 13.1 TFlops at 1600 MHz(ours).
GTX1080ti does 10.6 TFlops at 1480 MHz(Base Clock) 50% faster than Vega FE base clock makes 11.321 x 1.5 ~ 17 TFlops ~ 2370 MHz (That is a hell of an over clock).

Titan V does 12.288 Tflops at 1200 MHz(Base Clock).

https://www.slashcam.de/artikel/Test/Auf-Augenhoehe-mit--Ti-tanen---AMD-Vega-56-unter-DaVinci-Resolve-14--Was-bringts--63--Vega56-unter-Resolve-14.html#Was_b
Click to expand...

I don’t see any Titan V on that chart. The two V’s I have here are substantially faster than Titan Xp in most applications despite having similar tflops ratings on paper. Not all tflops are created equal.

Perhaps Resolve just loves Vega. Unfortunately for AMD, most other software does not seem to.
 
DJ Meyer said:
So I have no idea what they were showing on stage at Siggraph last August, but it wasn't this.
Click to expand...

Unfortunately that is what they were showing. Playback of lots of pre-rendered stuff. Because that’s all this is... Instead of utilizing your system’s storage for large scale cache and sending data off the GPU onto the PCIe bus, they strapped an SSD to the GPU itself on an
X4 PCIe bridge/ NVMe link. There’s absolutely no advantage here vs. having a PCIe v3 X4 NVMe or M.2 SSD in your system in terms of speed. The only thing this does is let the GPU be in complete control here with specific access to the SSD on it’s own private bus and via its firmware/software combo. It was a marketing gimmick then and it still is now. OTOH, cacheing large amounts of computed GPU data and preemptively computing ahead is the next logical evolution of where the whole GPU is headed. And quite frankly, we should already be there... Some of us have been asking for this for YEARS. It’s been 9 years since Adobe rolled out Premiere with GPU acceleration and we have spent those 9 years watching our GPUs/CPUs endlessly compute the same clips/data over and over and over again within our projects. Especially now that high-speed and high capacity storage have become affordable and ubiquitous within our industry, it’s well past time we have these abilities.
 
You must log in or register to reply here.
Back
Top