We have a SSD attached to a GPU now.

[sirmo]

No really, thats what the API is for. This is what i mean.

You will still have a little more latency because its a longer trip, but eletrically its the same thing.

Since when can a GPU autonomously talk to your computer's disk, without the drivers and CPU/kernel-filesystem involvement? Something has to negotiate and setup the DMA transfer and that's certainly the CPU.. GPU has no control over it.

 

[Paratus]

This may have already been posted but Ars had a little more info. The SSG gave the card an 80% boost in scrubbing though an 8K video.

http://arstechnica.com/gadgets/2016/07/amd-radeon-pro-ssg-graphics-card-specs-price-release-date/

Details on the rest of the card are thin, but AMD has said the SSG is based on a Polaris 10 chip, which is currently being put to work in the mainstream RX 480 graphics card. That might not seem that impressive on paper, but AMD claims that the SSG has a dramatic effect on the rendering speed of high-end workloads. To demonstrate, AMD showed a high-end system without SSG scrubbing through a RAW 8K video file at 17FPS. On the same system with the SSG it scrubbed through at a steady 30FPS. It even claims that a smooth 90FPS is possible.

Looks like for some professional applications using excessively large data sets this could be quite a speed up.

 

[Dresdenboy]

http://www.anandtech.com/show/10518/amd-announces-radeon-pro-ssg-polaris-with-m2-ssds-onboard

I told Ryan. Now check again.

 

[Shivansps]

Since when can a GPU autonomously talk to your computer's disk, without the drivers and CPU/kernel-filesystem involvement? Something has to negotiate and setup the DMA transfer and that's certainly the CPU.. GPU has no control over it.

Pcie devices can talk to each other directly, as someone already pointed out, thats how bridgeless cf work.

Im petty sure every gcn card can use this new amd api, thats probably the driver thay does everything, with any pcie ssd out there, as long is plugged on the same pci-e root as the gpu.

I see this as some kind of next gen virtual vram, where the gpu has direct access to a pcie ssd.

But ill stll be worried about the small latency added in access to system ram and other i/o by the pcie bridge on the card, also didnt 3dxpoint dimms render this obsolete already when they will start to show up? The good thing is this is also be able to used on 3dxpoint pcie ssd.

 

[Shivansps]

The hell?
https://developer.nvidia.com/gpudirect

http://docs.nvidia.com/cuda/gpudirect-rdma/index.html#axzz4FdWharf8

Just found that.

 

[wahdangun]

The hell?
https://developer.nvidia.com/gpudirect

http://docs.nvidia.com/cuda/gpudirect-rdma/index.html#axzz4FdWharf8

Just found that.

General remarks
Even though the only theoretical requirement for GPUDirect RDMA to work between a third-party device and an NVIDIA GPU is that they share the same root complex, there exist bugs (mostly in chipsets) causing it to perform badly, or not work at all in certain setups.

We can distinguish between three situations, depending on what is on the path between the GPU and the third-party device:
PCIe switches only
single CPU/IOH
CPU/IOH <-> QPI/HT <-> CPU/IOH
The first situation, where there are only PCIe switches on the path, is optimal and yields the best performance. The second one, where a single CPU/IOH is involved, works, but yields worse performance ( especially peer-to-peer read bandwidth has been shown to be severely limited on some processor architectures ). Finally, the third situation, where the path traverses a QPI/HT link, may be extremely performance-limited or even not work reliably.


Read more at: http://docs.nvidia.com/cuda/gpudirect-rdma/index.html#ixzz4FiMMaSy7
Follow us: @GPUComputing on Twitter | NVIDIA on Facebook

so it will work best if there are some kind of PLX chip in motherboard ?

 

[Shivansps]

so it will work best if there are some kind of PLX chip in motherboard ?

Certainly, thats the advantage in Radeon SSGs, the chip is already on the board, so there is no external hardware requirements.

But as far as technology goes, this is nothing new, it seems to have been possible for years now.