I'm missing the plot.
With announced decompressed speed, it takes 20 seconds to get to 100GB figure, how on Earth is that "instantly"?
You seem to be treating this as if all of this is being loaded into system ram or something, and not simply just a case of the CPU being bypassed altogether as opposed to the typical way where the GPU makes a request to the CPU for more data, they go to the hard drive, and then contents from the hard drive start getting copied into GPU memory.
On the contrary, it seems the scenario we're dealing with is one where the SSD is just being used directly like a slower, much larger amount of system RAM. In such a scenario it isn't necessarily about 100GB being accessed instantly all at once, like it needing to be loaded in somewhere like system RAM.
It's more that anything across a 100GB stretch of a game's stored data on the SSD can be instantly called upon and used for the game without copying into system RAM.
I believe this is the way it's actually meant. How often has Microsoft said they can use the hard drive as virtual ram. So forget the 20 seconds. That's unnecessary because they were already able to hit anything within that 100GB pretty much whenever they felt like it, and it wouldn't take them 20 seconds to do so either. It'd be much, much faster than that. The SSD can instantly provide information to the GPU without having to go through system RAM.
This link explains it somewhat. Digital Foundry's Richard Leadbetter even did a whole video on o it. Clearly he knows what's going on.
AMD says its Solid State Graphics Technology will revolutionize professional GPU performance by adding NAND memory to its upcoming pro graphics cards.
www.pcworld.com
AMD said that given the relatively small amount of memory in a graphics card today, the GPU must slice and dice workloads and then manage the merging of those disparate parts. When the GPU wants more work, it has to signal the CPU, which then fetches data from the larger pool of local system RAM or primary storage. This incurs latency.
With SSG, the CPU is bypassed entirely, which greatly reduces latency, AMD said. The company didn’t disclose too many details, but SSG appears to be based on a standard M.2 interface using PCIe.
SSG would not replace graphics RAM itself, such as GDDR5+ or HBM or HBM2, as it would be too slow. The company’s new Polaris-based Radeon RX 480, for example, has about 256GBps of memory bandwidth on tap. The best you can get out of a single M.2 interface today is 1.5GBps to 2GBps.
...
The SSG memory would be treated as part of a huge pool of RAM. If the GPU can’t find its data in the local GDDR5+ or HBM RAM, it would then search the SSG. Only after that would it have to ask the CPU for what it wants.
Here's is Richard's video on it.
At 4:30 you see a large dataset using the SSD as RAM on the left side, and on the right side you see them attempting to load all that content in and out of RAM. The right side where they're trying to use system RAM is hitching and slowing up, the left side with the SSD is smooth. Xbox Series X's SSD is faster than the one they were using for this, especially when the content is compressed.
www.resetera.com
github.com
