Hashrate for AMD Radeon Pro SSG
What will be the estimated hashrate for this when cracking an ethereum wallet (using the scrypt algorithm)? Even if its not a super accurate number I just need a baseline.
You will be sorely disappointed to know that it will likely be on par with the Vega64, but will cost an incredible amount more than a single Vega64. Depending on your scrypt settings(there are several scrypt wallet formats for ether with different settings) you will likely see anywhere from 0h/s to 6,000h/s. Assuming it even runs at all on your GPU.
Reason I'm spending so much on this is because I need it. It has 2TB of memory so should work with my scrypt settings.
https://www.sabrepc.com/amd-100-506014-r...-card.html Thats the GPU I'll be purchasing. So 2TB Of memory should be enough from my calculations https://paste.ofcode.org/s7L5Tnwd75Bg9MVj3z3P6H Correct me If i'm wrong at all.
The "2TB" of memory on the SSG, doesn't work like that. It's more of a local filesystem buffer to eliminate the bottleneck of going between system storage and the GPU over the PCI-e connection. You will NOT be able to utilize it as VRAM and it will NOT allow you to run high memory usage SCrypt settings any faster. This seems to be a very common misconception about how it works.

Yeah but, with my wallet KDFParameters I can't run it on a GPU whatsoever, however this should let me as it still has 2TB of memory. "applications must make specific changes to allow files to be stored locally on the SSG." does hashcat already have this implemented?
https://www.extremetech.com/extreme/2324...e-attached Also this post says that adding the SSD storage onto the GPU bypasses a common problem aka the GPU Memory limit
You will not be able to run it on an SSG any faster than on a VEGA64, as it simply does not work like that, at least to my knowledge. The SSD on the SSG is used as an extended buffer to replace system RAM/Storage and reduce latency when loading/unloading files. If you consider that currently you can not utilize System RAM/Storage as terabytes of VRAM, then an SSD designed to act in place of something you already can't utilize is clearly not going to be magically capable of acting as VRAM. The card has 16gb of HBM2 memory and that is ALL you will be able to use for the scrypt computations. If it was possible to utilize the SSD for Scrypt, then we could simply do that right now with your System RAM/Storage at the cost of slightly higher latency.

If you want to run Scrypt, get some fast RAM and fast CPUs and just run it on that. It will be far cheaper/faster/more efficient than trying to convince your GPU to use system resources as VRAM.
I may not to be run it faster but it should still run.... I've read multiple articles some referring to it as VRam and saying it is VRam and the only one you showed me saying it isn't. Running it on CPUs get about 4 -8 hashes/s, even if the GPU improves it to 5,000 hashes a second.
It is NOT VRAM. It is an extended frame buffer for loading/unloading large files when processing. This should be obvious by the 8x PCI-E lanes that the SSD literally hogs to itself to load/unload said large files from the system. And I have no idea what CPU you are running but if you are only seeing 4-8h/s on a modern, high performance CPU, you shouldn't expect anything even close to 5,000h/s from a GPU. Expect a maximum in the hundreds of hashes per second, across multiple(8+) GPUs in one system. A 1080Ti can do like 45h/s MAX per card on the higher default scrypt settings for ether wallets. And at least the 1080Ti can keep itself cool. Don't forget that the SSG is an AMD VEGA64 under the hood. As nice as they look on bursty benchmark, be prepared to lose like half that speed after it heats up and starts throttling.

I will say this again: Gather some CPUs into a cluster and you will come out ahead in damn near every way.
Okay from talking with someome who literally owns his own dedicated server hosting company, i.e he knows his shit, he agreed that the SSG on the GPU acts as memory for the GPU, i.e VRam. Do you have any other sources other than a blog post to prove it. (Sorry for sounding like a pain, but I need to know 100000%)