AMD's Smart Access Memory: Nvidia equivalent possible?

In conventional Windows-based PC systems, processors can only access a fraction of graphics memory (VRAM) at once, limiting system performance.
With AMD Smart Access Memory, the data channel gets expanded to harness the full potential of GPU memory removing the bottleneck to increase performance.
System Requirements
Hardware
AMD 500 Series Motherboards
AMD Ryzen™ 5000 Series Processors
AMD Radeon™ RX 6000 Series Graphics*
Software
AMD Radeon Software Driver 20.11.2 or newer
Latest Motherboard BIOS (based on AMD AGESA 1.1.0.0)

Amd claims "Up To 11% Extra Performance Across Select Titles."


Is a response even possible from Nvidia ?

 

No but Intel can with their gpus.

 

how? nvidia hate intel and amd

 

I don't follow. He said Intel could offer a similar feature set (they are developing high performance mobile, soc & dedicated "Xe" offerings) and you replied NVIDIA hates Intel and AMD? What does that have to do with with an Intel-specific alternative to Smart Access Memory ?

 

Nvidia sure loves Ryzen though! - https://twitter.com/NVIDIAGeForce/status/1148242629940789248

I assume Undying meant Intel and their Xe dedicated GPU's, not Nvidia's.

 

https://docs.microsoft.com/en-us/windows-hardware/drivers/display/resizable-bar-support

 

I believe most Intel motherboards (Asus x99) have an bios option to enable access to 64-bit pcie devices (Above 4G decoding).

 

irrelevant, this is purely a setting for addressing multiple graphics cards at a time in a system where the resources are by default distributed elsewhere (IO resources)

 

The primary function is to allow the users with multiple cards to access memory mapped I/O for a 64-bit PCIe device to 4GB or greater address space, but can work with single cards (commonly Tesla).
Other names for "Above 4G Decoding" is ">4GB MMIO" , "64-bit MMIO", "Memory Hole for PCI MMIO".

https://www.reddit.com/r/Amd/comments/jk76u9/smart_access_memory_already_works_on_linux_and_is/

 

Yep at least *SPAM* far back as X58 possibly earlier than... that's just the first consumer chipset I recall seeing it on for some reason (not that it wasn't available on any non HEDT ones around that time)

As most users generally weren't using GRID cards in personal computers... PCs became more powerful with the dawn of personal workstations that levied robust resources inside a single system as opposed to what came before.

 

Except it breaks most graphics drivers because theres more to it than the description lets on, its not for this case period.

Its not really even intended to be used on a system unless its headless because the primary graphics card must always be mapped below 4GB. Most importantly, many bios's that implement this option do not even use 64bit pcie addressing.

 

Well that's good to know I was curious what that thing was in the bios.
Plus it'd have to be different from this since the new AMD AGESA introduced an additional bios setting.

Hardware restrictions aside it'd be interesting to see if AMD would unlock support for this and if it's going through a WDDM functionality then once supported whether AMD's 5000 processor series only or older eventually then other GPU vendors like Intel or NVIDIA could utilize it if their drivers supported this feature.
(Assuming there's nothing non-standard going on here but I'd assume not.)

EDIT:
https://www.reddit.com/r/Amd/comments/jmnlbc/hidden_resize_bar_support_option_in_latest_asus/



Though for now I don't expect this to have much of a effect unless the user has a 5000 series AMD processor and driver support which for AMD comes with the second November 2020 driver update.
https://www.amd.com/en/technologies/smart-access-memory

Still if that's what it is other than hardware restrictions and AMD locking this to X CPU's and Y GPU's and newer if it's a standard thing other vendors should be capable of utilizing support for this too.


EDIT: Game and game engine wise I would also expect a lot of them to be taking this what is it 256 MB data package size and working with that so there's gains to be had but nothing too major.
(Up to 5% is still free 5% performance though so it's something.)

However if Direct Storage could eventually start being utilized then sending even more data might really be advantageous but that's just me theorizing we don't really know how this is going to turn out and what the initial state of support on PC for this will even look like.

 

I'm not sure i understand what's preventing this from working on Nvidia+intel or intel+6000 series video cards for that matter.
How about a AMD 5000 cpu +500 series mobo and 5700 xt or a Nvidia RTX combo?
Is AMD just locking this feature on purpose to the 5000 series cpus +500 series mobos + 6000 series video cards combo just to boost sales ?
Is it the cpu that needs to support it , the mobo chipset, bios, drivers ?

 

AFAIK the only requirement is MMIO and a PCIE4.0+ controller

almost no pcie 3 controllers support 64bit tagging or addressing to my knowledge, though Resizable bar was first promoted for Gen 2.0 mostly for server and workstation packages.

 

From the link in my post above:

 

Yeah. It will be interesting to see the enable/disable performance variances under different scenarios once the RX 60xx reviews are out.

 

See, what I don't understand is why the base premise is even a problem to begin with -- so, why is it that in conventional Windows-based PC systems, processors can only access a fraction of graphics memory (VRAM) at once, limiting system performance?

Why is that even a problem to begin with? Sounds like something Windows should fix, no? Of course I may be misunderstanding the problem/how it works, but I would think all CPUs should get full data channel access.

It does kinda seem to me that a full AMD rig for 4K is the way to go right now -- Freesync is a bit cheaper than G-Sync, AMD's GPUs have a lot more VRAM for 4K gaming, and their CPUs (5600X for example) are very fast for gaming now. Now, Nvidia I bet will have the faster raytracing performance and we haven't seen tests yet for 6800 / XT and all that, but my bet is an all AMD rig will be a decent buy soon. Smart access memory is something I'm honestly surprised hasn't been addressed yet by others in the space -- it just seems like a problem that we would've worked around a long time ago, no?

 

Classic Y2K, aka "eternal legacy". Somebody, somewhere, sometimes either pulled a number out of their @ss radomly or made an educated guess about the foreseeable future but there came a time when it proved indecent to carry any further. Often times these "legacy" things are buried so deep into "industry standards" or "best practices" (and people's memory from books, practice, etc) that they are carried over far far FAR longer than it could be considered a sane practice from "cleaner" perspectives (but people inside the industry treat these as axioms, they don't happen to question them until something seriously looks broken and they have to bisect everything in the chain with an open perspective).

My favorite example is how chroma subsampling was a sheer necessity for making analogue television broadcast colorful while maintaining backwards compatibility. Now some 50+ years later we are still stuck with this crude data butchering workaround for UHD HDR formats (which are in no way directly backwards compatible anymore and much better alternatives could have been put into place for bandwidth efficient compression and encoding). Yet it will stick around for at least 10 more years.

 

It seems Nvidia already has something similar in the works, and will be releasing it for Ampere soon.
https://twitter.com/Disney/status/1327006795253084161

I agree with BlindBison, seems like a really odd problem given we've been 3 generations above 8GB Video memory. Maxwell I could understand, but atleast Pascal onward it should've been a given. So if I get this right, windows can only access 256MB out of my 11GB at a given time? Why is this even a problem to begin with? Either way, at the end of the day, competition is good.

Seems Nvidia will enable it for both Intel and AMD, so the consumer wins.

 

From the tweet I see it does not require pcie 4 this means they could enable it for all cards not just the 3k series.
I just hope they don't make it a 3k series exclusive to push people into upgrading using marketing bullshit.
Up to 11 % is a lot of performance that can be had without it depending on the video card since this just gives the cpu access to full vram.
It's like adding up to 500 mhz extra overclock on a 4500mhz cpu in terms of fps performance.
I can't help but wonder how much extra performance is left on the table because of legacy stuff like this on modern platforms.
Msi mode for lowering input latency and improving stuttering...