Hashrate for AMD Radeon Pro SSG
#11
http://www.tomshardware.co.uk/amd-radeon...53609.html

"AMD indicated that it is not using the NAND pool as a memory-mapped space currently..."

"The current implementation merely uses the SSD as a storage volume for frequently-accessed data..."

As I have already said a number of times, as it stands right now, you CAN NOT use the SSD as VRAM. It is simply an extended frame buffer.
Reply
#12
(01-05-2018, 09:42 PM)DonManD Wrote: Okay from talking with someome who literally owns his own dedicated server hosting company, i.e he knows his shit,

Chick3nman literally works as an engineer for an HPC company that specializes in GPGPU -- i.e., he also knows his shit, and is far more qualified to speak on this subject than someone who merely owns a hosting company.

(01-05-2018, 09:42 PM)DonManD Wrote: he agreed that the SSG on the GPU acts as memory for the GPU, i.e VRam.

Then he does not, in fact, know his shit. For the SSG, AMD simply added a PLX chip and hung two x4 PCI-e M.2 slots off of it for on-board storage. That's all it is. The purpose is to minimize host-device transfer latency by moving the device storage to local storage. It is not VRAM. It does not behave like VRAM. And it cannot be used as VRAM.

(01-05-2018, 09:42 PM)DonManD Wrote: Do you have any other sources other than a blog post to prove it.

RTFM? http://pro.radeon.com/_downloads/SSG_API...0v1.01.pdf

Pretty clear from the API documentation that the OpenCL extensions added are solely for file operations. And even if the API did have functions to use the onboard storage as VRAM (which, again, it does not), those functions would have to actually be implemented within Hashcat to make use of them, which would be a significant undertaking for a device that surely no one would use for password cracking. It wouldn't just magically work, and we would not be willing to invest the level of effort required to make it work. But again, it does not exist, so this is a non-issue.
Reply