Monday, July 25th 2016

AMD Announces the Radeon Pro SSG

AMD today announced it is developing a new Radeon Pro solution for large dataset applications, which will be available initially as a developer kit. Starting at a full terabyte of Radeon Solid State Graphics (SSG), this innovative Radeon Pro technology will provide more than an order of magnitude greater memory capacity compared to existing GPU memory implementations, enabling vastly higher performance. The new solution is ideal for the next wave of demanding use-case scenarios including real-time post-production of 8K video, high-resolution rendering, VR content creation, oil and gas exploration, computational engineering, medical imaging and life sciences.

During the world's first demonstration of a Radeon Pro SSG solution, 8K raw video timeline scrubbing was accelerated from 17 frames per second to a stunning 90+ frames per second. As content creators rapidly adopt 8K resolution to future-proof their content, a 5X performance boost will improve quality of life, productivity and efficiency in the editing process.
State of the art content creation, scientific and engineering visualization applications require the processing of big datasets, far larger than can be contained within the capacities of existing GPU memory. Current limitations require slices of data to be processed individually and later merged by software, and often incur significant latency for fetching additional data from system memory. These big data problems discourage developers in these domains from leveraging the advantages of the GPU. Radeon Pro SSG memory addresses the big data problem for GPUs, paving the way for improved performance and dramatically increased user productivity.

"One of the most challenging constraints faced by GPU computing applications is the inability to access terabytes of data," said Raja Koduri, senior vice president and chief architect, Radeon Technologies Group, AMD. "Radeon Pro SSG is poised to not only speed-up processing for many applications with very large datasets, but also to enable new application experiences by utilizing data persistence of non-volatile memory. This will be a disruptive advancement for many graphics and compute applications."

"AMD has a long history of memory technology innovations, and Radeon Pro SSG is the latest example," said Patrick Moorhead, Founder and Principal Analyst, Moor Insights & Strategy. "Larger local memory on the graphics card can speed-up processing for many applications with very large datasets, and should also allow for results with finer granularity and resolution. This will be a notable advancement for many graphics and compute applications."

Memory architecture
When needed information is not available in GPU memory, typical memory architecture requires the GPU to send a request to the CPU. The CPU then retrieves the needed content from CPU memory, or if not there then from a hard drive. This entails considerable overhead that limits GPU performance. With this new GPU technology, a one terabyte extended frame buffer is dedicated to support the GPU. This enables much larger datasets to be loaded locally, connected over a dedicated PCIe bus. When the GPU requests content, it looks first in the local frame buffer and only needs to involve the CPU if the data is not already in the extended frame buffer.

Lineup
  • Radeon Pro WX 4100 is designed for small form factor (SFF) workstations and provides amazing performance in a half-height card.
  • Radeon Pro WX 5100 is ideal for real-time content engines and immersive real-time design and manufacturing, including CAD and CAM.
  • Radeon Pro WX 7100 is fully capable for design engineering and media and entertainment, whether video editing or image creation, and is AMD's workstation solution for professional VR content creation.
Availability
Applications for developer kits are now being accepted. These are available for $9,999. Full availability is planned for 2017.
Add your own comment

26 Comments on AMD Announces the Radeon Pro SSG

#1
Frick
Fishfaced Nincompoop
So if I understand this correctly it's a GPU with an SSD directly attached to it?
Posted on Reply
#2
Chaitanya
Frick
So if I understand this correctly it's a GPU with an SSD directly attached to it?
yes, there is a m.2 slot for ssd which can be used as local storage if needed. Interesting addition to GPU.
Posted on Reply
#3
D007
Pretty neat.
Posted on Reply
#4
lZKoce
I've always wondered if in the professional world they use SFF workstations with half-height cards ? Do they ? Can someone share an experience ?
Posted on Reply
#5
Devon68
Well the joke is on me for thinking the SSG meant Single slot graphics.
Posted on Reply
#6
TheGuruStud
Devon68
Well the joke is on me for thinking the SSG meant Single slot graphics.
Nah, it's just named after me.

I better put on my tinfoil hat, I've exposed myself.
Posted on Reply
#7
thevoiceofreason
Isn't the latency of accessing RAM over PCI-Express a thousand times better than that of M.2? Don't professional cards implement a DMA master or something? I am really confused by this product.
Posted on Reply
#8
the54thvoid
Anandtech says it's a Polaris 10 (RX480) card. If true that's quite a mark up, even for the professional end.
Posted on Reply
#9
R-T-B
thevoiceofreason
Isn't the latency of accessing RAM over PCI-Express a thousand times better than that of M.2? Don't professional cards implement a DMA master or something? I am really confused by this product.
All good points. I too am confused by this...
Posted on Reply
#10
Ferrum Master
ahh... it all comes back, like in the ol good days slapping few sips of RAM modules or IC's in the socket...

Posted on Reply
#11
truth teller
thevoiceofreason
Isn't the latency of accessing RAM over PCI-Express a thousand times better than that of M.2?
thousands? no, but the dram throughput is higher only if there is nothing using the bus.

i can see the need for a local cache/buffer so the main pci bus doesnt get saturated while moving/playing data, but 10k american zlots for it is a complete joke, even for the professional market.
also why couldnt they release the cards used to make these like this, in single slot form factor, with a regular bios and sans all the additional professional grade mambo jambo?
Posted on Reply
#12
Recus
No HBM2 - $9,999. Ethical milking. =)
Posted on Reply
#13
Nergal
Well done by AMD.
Will perhaps bring another evolution in GPU´s.
Posted on Reply
#14
iO
thevoiceofreason
Isn't the latency of accessing RAM over PCI-Express a thousand times better than that of M.2? ...
16 Pcie lanes offer much more bandwith but completly eliminating the CPU overhead will lower the latency a lot.
Also a TB of M.2 storage is super cheap compared to the same amount of RAM and its needed platform.

Edit: It has two x4 M.2 slots which will be connected to 8 lanes of the GPU with a PLX chip or something like that..

That thing might be a small revolution for the pro market...
Posted on Reply
#15
Frick
Fishfaced Nincompoop
truth teller
i can see the need for a local cache/buffer so the main pci bus doesnt get saturated while moving/playing data, but 10k american zlots for it is a complete joke, even for the professional market.
Pennies for the target audience, I expect. And developers kits are always expensive. They're not even taking orders, just applications.
Posted on Reply
#17
deu
truth teller
thousands? no, but the dram throughput is higher only if there is nothing using the bus.

i can see the need for a local cache/buffer so the main pci bus doesnt get saturated while moving/playing data, but 10k american zlots for it is a complete joke, even for the professional market.
also why couldnt they release the cards used to make these like this, in single slot form factor, with a regular bios and sans all the additional professional grade mambo jambo?
10k for a developer-kit. (Call it a firstmoverprice) but to be honest i dont see what justify this price other than that some people have too much money anyways
Posted on Reply
#19
bug
Won't these be superseded by Vega in 2017?
Posted on Reply
#20
dyonoctis
deu
10k for a developer-kit. (Call it a firstmoverprice) but to be honest i dont see what justify this price other than that some people have too much money anyways
Professional gpu always had their price getting multiplicated by a huge factor: the firepro w7000 cost 700, even when the R9 270x only costed 199.

Add to that the manufacturing cost of a gpu+ RAID 0 of two samsung 512 Gb 950 pro directly accessible by the gpu on the same board, +the ultra premium rma you get from a professional product.

Appenrently this will be actually faster than trying to get the gpu to access memory from another pci-e storage. From what I've read only really heavy rendering (like state of the art Hollywood VFX) really need something like this.
Posted on Reply
#21
Kaotik
The developer kit might actually be full computer, too, with Radeon Pro SSG inside
Posted on Reply
#22
Jism
I've read that this sollution can be used to speed up the reading and loading proces of 8K projects / video's, for example everytime you open up a project it does'nt take an half hour to an hour to fully cache the video, but straight load it from the GPU's 'SSD' memory.

For big data, weather and all this is very interesting as well, and this is exactly where the big money is for AMD. They are doing a very good job with new and interesting things.
Posted on Reply
#23
Basard
Nergal
Well done by AMD.
Will perhaps bring another evolution in GPU´s.
Having HBM on them, they should just have a socket on the mobo for them!
Posted on Reply
#24
truth teller
relevant (and probably one of the few usage scenarios for this), skip to the 1:06:00 mark (the youtube bb code tag doesnt parse time params)
Posted on Reply
#25
$ReaPeR$
truth teller
relevant (and probably one of the few usage scenarios for this), skip to the 1:06:00 mark (the youtube bb code tag doesnt parse time params)

that was also what i read about it and it seem quite "groundbreaking". i hope it turns out good for them since they are in dire need for cash.
Posted on Reply
Add your own comment