AMD Radeon Pro SSG flash-based memory to boost graphics performance

Published by

Click here to post a comment for AMD Radeon Pro SSG flash-based memory to boost graphics performance on our message forum
https://forums.guru3d.com/data/avatars/m/263/263841.jpg
I'd love to know if it requires special firmware to talk to the GPU over some variation of the X-fire protocol (maybe?) or if any PCIE SSD would work. And while I'm speculating wildly, what are the chances of this tech coming to consumers e.g. partitioning a section of your SSD to act as additional VRAM , allowing nightmare mode on Doom for example?
https://forums.guru3d.com/data/avatars/m/209/209146.jpg
I'd love to know if it requires special firmware to talk to the GPU over some variation of the X-fire protocol (maybe?) or if any PCIE SSD would work. And while I'm speculating wildly, what are the chances of this tech coming to consumers e.g. partitioning a section of your SSD to act as additional VRAM , allowing nightmare mode on Doom for example?
I think you can do "Nightmare" settings on Doom reliably with "just" a 8 GB 300 or 400 series AMD GPU, 1 TB might be overkill. 😀 (Not that I would mind if this tech were to ever get to desktop models but that's a ways off if it's even planned at all. :P ) (Being able to put an entire game into VRAM would be kinda fun too...)
https://forums.guru3d.com/data/avatars/m/238/238369.jpg
Blue Theme hah? Interesting... About the card, well nice technology from amd once again.
https://forums.guru3d.com/data/avatars/m/198/198862.jpg
For those who did not get it as it seems bit confusing for many. This is Graphics card with additional M.2 slot on board to increase available data storage for purpose of caching/editing.
Thanks for explaining it, Fox.
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
Ah, now I get it, I thought it was indeed running via PCIe lanes, not onboard with the GPU itself.
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
Good idea for 4K/8K editing alright. So the GPU will have access to a large buffer with copy of the timeline, but the thing I can't understand is, won't the CPU also need the timeline buffer too most of the time? So will there be a big overhead constantly keeping both the main memory (CPU) and this graphics card buffer (GPU) in sync with each other? I mean if the CPU does some sort of operation on the frames, then the entire SSD on this card will have to be rewritten with updated data from main memory? Or can the software be written so there is only a single copy of the timeline framebuffer in the GPU/SSD card framebuffer only, and the CPU can modify this also?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Anandtech is reporting these as Polaris based, for $10,000? The M.2 slot idea is kind of cool but it's not much different then just having your working files on a M.2 drive on the motherboard.
https://forums.guru3d.com/data/avatars/m/260/260828.jpg
"Next gen memory"?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
not even close
Damn, I wish I had thought of it that way. You're right.
https://forums.guru3d.com/data/avatars/m/263/263841.jpg
Proof of Concept waiting on xpoint memory to become available?
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
Proof of Concept waiting on xpoint memory to become available?
Maybe. Currently it's limited to 8x PCI-E -- so technically you could achieve better throughput by just raiding a bunch of SSD's together. AMD's system should technically have a latency advantage, which I guess helps with things like the video scrubbing they were showing off. Although it's unknown how much it helps because who knows what system they were comparing it too.
https://forums.guru3d.com/data/avatars/m/164/164033.jpg
Maybe. Currently it's limited to 8x PCI-E -- so technically you could achieve better throughput by just raiding a bunch of SSD's together. AMD's system should technically have a latency advantage, which I guess helps with things like the video scrubbing they were showing off. Although it's unknown how much it helps because who knows what system they were comparing it too.
The chip on the ssg might be fiji. Well there is a picture of the board from the backside and people are thinking that it is a fiji chip. https://twitter.com/FudzillaNews/status/757774661081899008 Sure it is Fudzilla. But Fiji would make more sense. 10 000$ well it ain't only for the hardware so understandable.
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
I would think Fiji too, but both Anandtech and PC Perspective say it's Polaris.
This new professional graphics card pairs a Polaris GPU with up ot a terabyte of on board solid state storage and seeks to solve one of the biggest hurdles in GP GPU performance when dealing with extremely large datasets which is latency.
http://www.pcper.com/news/Graphics-Cards/AMD-Introduces-Radeon-Pro-SSG-Professional-GPU-Paired-Low-Latency-Flash-Storage http://www.anandtech.com/show/10521/amd-announces-radeon-pro-wx-series-wx-4100-wx-5100-wx-7100-bring-polaris-to-pros Title says it all on the second one.
https://forums.guru3d.com/data/avatars/m/259/259654.jpg
Maybe. Currently it's limited to 8x PCI-E -- so technically you could achieve better throughput by just raiding a bunch of SSD's together. AMD's system should technically have a latency advantage, which I guess helps with things like the video scrubbing they were showing off. Although it's unknown how much it helps because who knows what system they were comparing it too.
Latency wise the differences are probably orders of magnitude.
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
well its connected directly to the gpu, latency wont be the big issue
That's his point. Otherwise there would be no benefit over just having a raided M.2 setup. I'm sure there is a difference, I'd like to see it compared with a real system though and not AMD's black box. ~800MB/s is way to low for there to be raided SSD's in the system like there is on the SSG card. I definitely like the idea -- I'd just like to see it compared to a similar setup.