Nvidia working on similar memory acess to AMD

[GhostRoadieBL]

Summary

 In a statement to Gamer's Nexus Nvidia is claiming to have a similar feature to AMD's new Big Navi cards with Smart Access Memory allowing better memory transfer between cpu and gpu. 

According to Nvidia they have had this feature working on Ampere cards in testing and can roll it out as a software update. 

Even going as far as to say it can even work with pcie3.0 and intel cpus. 

 

Quotes

Quote

 According to a statement Nvidia gave to Gamer's Nexus, the company says it will soon enable a feature similar to AMD's Smart Access Memory (SAM) tech, which boosts data transfer efficiency between the GPU and CPU. In fact, Nvidia already has the feature working on its Ampere graphics cards in its labs.

 

My thoughts

Nvidia acting like school kids on the playground "well my toy can do everything yours can" 

This feels like something Nvidia wasn't planning on announcing or even had much interest in until AMD kicked in the door and told Nvidia what for. If Nvidia had this available to them at launch why are they only mentioning it now? What about the Ampere cards is so special that a software update adds this kind of feature but no other cards in their lines are announced to have it? 

Holding back performance features is an intel move so I hope AMD can keep competing at a level which pushes others to release these features to keep up. 

 

Sources

 https://www.tomshardware.com/news/nvidia-amd-smart-access-memory-tech-ampere

[Skiiwee29]

Thing is.. its not an "AMD or NVidia" thing.. its a spec that was added into the PCI-E spec back in 2017. They just never used it because it just didn't give enough benefits. 

 

 

 

[CommanderAlex]

3 minutes ago, Skiiwee29 said:

Thing is.. its not an "AMD or NVidia" thing.. its a spec that was added into the PCI-E spec back in 2017. They just never used it because it just didn't give enough benefits. 

If that's the case, then would that feature be on Turing cards (RTX 20 series)? 

[Skiiwee29]

Just now, CommanderAlex said:

If that's the case, then would that feature be on Turing cards (RTX 20 series)? 

Possibly, but unlikely since that would have been added/ratified into the spec likely after Turing was in end stage development/production. 

[PopReference]

Just a theory but probably only implemented because consoles will be using similar tech so it's easy for AMD to transfer to PC.

[Random_Person1234]

Doesn't AMD Smart Access Memory require a Zen 3 CPU and a 500 series motherboard? How would Nvidia do it without cooperating with Intel or AMD, or did they find a way to do it without requiring specific CPUs and motherboards.

Edit: Nvm, just saw that it works with Intel CPUs.

[samcool55]

How about CPU support?

I mean, cool that Nvidia will support it too eventually but, what about the CPU?

Is anything holding AMD back to give nvidia the good old F-you and make it only supported on AMD gpu's if you have an AMD cpu?

[WereCat]

3 minutes ago, Random_Person1234 said:

Doesn't AMD Smart Access Memory require a Zen 3 CPU and a 500 series motherboard? How would Nvidia do it without cooperating with Intel or AMD, or did they find a way to do it without requiring specific CPUs and motherboards.

I'm still not sure about why it should only work with Zen3 and not Zen2 if both are on the same chipset. 

[Senzelian]

20 minutes ago, Skiiwee29 said:

Thing is.. its not an "AMD or NVidia" thing.. its a spec that was added into the PCI-E spec back in 2017. They just never used it because it just didn't give enough benefits. 

 

 

 

More like 2008.

 

 

Edit: Btw, important to not that this stuff doesn't work with MBR and CSM.

https://www.igorslab.de/en/b550-or-x570-new-ryzen-and-the-hope-on-the-radeon-6800-xt-why-many-your-windows-may-have-to-reinstall/

 

 

22 minutes ago, GhostRoadieBL said:

Nvidia acting like school kids on the playground "well my toy can do everything yours can" 

bs.

It's simply how things work. You hold features back to minimize cost.

[GhostRoadieBL]

15 minutes ago, Skiiwee29 said:

They just never used it because it just didn't give enough benefits. 

That sounds alot like "we have this feature which could boost performance but we just don't want to give users and game developers the option." 

[SFFDesigns]

This feature isn't restricted to AMD 500 series mobos and Zen 3, AMD is artificially limiting it.

 

I'm super happy that Nvidia is doing this, although it is pretty interesting that Nvidia is now the one who is being more "open" lol

[GhostRoadieBL]

5 minutes ago, Senzelian said:

It's simply how things work. You hold features back to minimize cost.

If it's been bench tested on ampere the "minimize cost" doesn't make sense, the tech has been developed so the cost is already spent through r&d, further polish may take some cost but nowhere near the cost of initial development. 

[Blademaster91]

It seems like AMD may have something competitive, Nvidia wants to have the feature as well.

[.Apex.]

11 minutes ago, GhostRoadieBL said:

That sounds alot like "we have this feature which could boost performance but we just don't want to give users and game developers the option." 

As far as we've seen the feature only allows a ~6% performance boost which approaches to top NVIDIA's performance charts, it might have not been worthy to implement to begin with, but since the competition is this close it might be beneficial to increase sales, and if you are going to criticize NVIDIA for not supporting this feature on their previous gen GPUs then why not do the same for AMD as well? I'm not protecting either but it seems like an insignificant matter to get angry about.

 

AMD also copied dozens of features that NVIDIA has initiated first, it doesn't matter to me as long as competition is thriving and they're pushing standards and that benefits everyone.

[Senzelian]

13 minutes ago, GhostRoadieBL said:

further polish may take some cost but nowhere near the cost of initial development.

Oh yeah, we just need to invest a few [insert number I pulled out of my *ss] dollar to keep this feature running if we release it now. Yeah, that's nowhere near as much as before. We should totally spend that to gain... uh ... nothing. 

 

Edit:

Do you honestly think that there is any other reason to hold a feature back than to cut costs?

If you're not after long term profit as your only goal as a company, then you're either on drugs or already bankrupt.

[RejZoR]

35 minutes ago, Blademaster91 said:

It seems like AMD may have something competitive, Nvidia wants to have the feature as well.

While I can't shit enough on garbage NVIDIA Control Panel, I have to give them credit for adding new software features that work across their entire product ranges generations back, but not just for very latest series. Like Fast Sync/Adaptive Sync, Image Sharpening, Ultra Low Latency and hopefully this too.

 

This is what I really don't like with AMD where most new features only work with Navi. Everyone who bought RX Vega, well, tough luck, you don't even get freaking Image Sharpening feature... I get it RDNA is what they are going with now, but you can't just screw users over like this.

[BuckGup]

3 hours ago, samcool55 said:

How about CPU support?

I mean, cool that Nvidia will support it too eventually but, what about the CPU?

Is anything holding AMD back to give nvidia the good old F-you and make it only supported on AMD gpu's if you have an AMD cpu?

Nvidia probably didn't do it as the performance uptick was negligible at 1-3%. But since AMD has both CPU, Chipset, and GPU all accounted for they an get numbers as high as 8% like was was shown off in the presentation 

[Mark Kaine]

5 hours ago, Skiiwee29 said:

Possibly, but unlikely since that would have been added/ratified into the spec likely after Turing was in end stage development/production. 

Yeah they knew this is a thing, and despite this "news" I'm 100% certain they, or some other news talked about this before - though I'm a bit perplexed why it's not implemented yet or even more talked about... 

 

 

 

[StDragon]

5 hours ago, Hymenopus_Coronatus said:

This feature isn't restricted to AMD 500 series mobos and Zen 3, AMD is artificially limiting it.

That's a bold accusation! I'm not saying you're wrong. But such a statement is throwing down the gauntlet. Care to back that up with facts?

 

For anyone else, what exactly is implemented at the HW level (if any at all) on Zen3 that makes Smart Access Memory so special? Or in fact is AMD only enabling this at the device driver level by checking for Zen3?

[SFFDesigns]

5 minutes ago, StDragon said:

Care to back that up with facts?

Sure. Zen 2 and Zen 3 have the same IO die, so theoretically it should work with Zen 2.

 

I'm not an expert obviously, but I think it's likely it's an artificial limitation

[LAwLz]

I mentioned this in the Ryzen 5000 series review threads:

On 11/5/2020 at 5:10 PM, LAwLz said:

From what I've read, Smart Access Memory is just AMD's branding of the resizable PCIe BAR. I don't think there is any special feature the CPU must support in order to use that. That means we might get it on older CPUs too, but maybe AMD will artificially limit it so they can sell their new CPUs.

 

Some things to keep in mind here:

1) Smart Access Memory, which seems to just be AMD branding for resizable PCIe BAR, is not something AMD came up with. It's a PCIe specification.

2) Resizable PCIe BAR is a standard in the PCIe specs. Any member of PCI-SIG can implement it if they want, including AMD, Intel, Nvidia, ARM, etc.

3) Resizable BAR has not been supported in Windows until fairly recently. That's one of the reasons why we haven't seen it before.

4) AMD seems to be artificially limiting it to their newest CPUs, motherboards and graphics card in order to sell more products. I can't think of any reason why they couldn't enable it on older products if they wanted to. It is fairly telling that AMD even has a requirement of a 500 series motherboard, despite the motherboard barely having anything to do with PCIe since that is controlled by the CPU.

 

 

Let's hope Nvidia enables it for all CPUs.

Would be a bit funny if Nvidia cards had better feature compatibility with AMD CPUs than AMD's own GPUs has.

[RejZoR]

AMD tends to do this a lot. Even feature sets on Radeon cards are mostly limited to RX 5000 series. If you have Vega, you just don't get it. Or if you have Polaris... Where NVIDIA tends to add features that go quite few series back. Currently I believe they go as far as Maxwell 2 (GTX 900 series) for the most part and the rest is hardware limited (like variable shaders that are only supported on Turing or newer because you just can't do it on older cards).

[Poison4K]

28 minutes ago, LAwLz said:

I mentioned this in the Ryzen 5000 series review threads:

 

Some things to keep in mind here:

1) Smart Access Memory, which seems to just be AMD branding for resizable PCIe BAR, is not something AMD came up with. It's a PCIe specification.

2) Resizable PCIe BAR is a standard in the PCIe specs. Any member of PCI-SIG can implement it if they want, including AMD, Intel, Nvidia, ARM, etc.

3) Resizable BAR has not been supported in Windows until fairly recently. That's one of the reasons why we haven't seen it before.

4) AMD seems to be artificially limiting it to their newest CPUs, motherboards and graphics card in order to sell more products. I can't think of any reason why they couldn't enable it on older products if they wanted to. It is fairly telling that AMD even has a requirement of a 500 series motherboard, despite the motherboard barely having anything to do with PCIe since that is controlled by the CPU.

 

 

Let's hope Nvidia enables it for all CPUs.

Would be a bit funny if Nvidia cards had better feature compatibility with AMD CPUs than AMD's own GPUs has.

Yeah i really have no fucking clue why you'd need a Zen 3 CPU, i thought that was really weird for AMD when it was first revealed, can't wait to get SAM on a 3080 and get up to an almost ~20% boost in performance in certain scenarios lmao

[porina]

54 minutes ago, LAwLz said:

4) AMD seems to be artificially limiting it to their newest CPUs, motherboards and graphics card in order to sell more products. I can't think of any reason why they couldn't enable it on older products if they wanted to. It is fairly telling that AMD even has a requirement of a 500 series motherboard, despite the motherboard barely having anything to do with PCIe since that is controlled by the CPU.

I'd cut AMD some slack for this. After all, it is a common complaint not just on AMD, but Intel and nvidia about requiring at least a certain if not latest gen product to get a feature.

 

I think overall it is a compatibility verification thing. It is one thing to say you're implementing an industry spec, but another for it to inter-operate well. For now, if you limit the variables to the latest AMD CPU, latest AMD GPU, latest AMD chipset series, then you have something easier to deal with. Let the dust settle, then let's see if they will expand support to older generations of CPU, GPU, or whatever. By that time, maybe we'll also know more about what Intel and nvidia are doing also, and have a better picture overall.

[Mark Kaine]

5 minutes ago, porina said:

I'd cut AMD some slack for this. After all, it is a common complaint not just on AMD, but Intel and nvidia about requiring at least a certain if not latest gen product to get a feature.

 

I think overall it is a compatibility verification thing. It is one thing to say you're implementing an industry spec, but another for it to inter-operate well. For now, if you limit the variables to the latest AMD CPU, latest AMD GPU, latest AMD chipset series, then you have something easier to deal with. Let the dust settle, then let's see if they will expand support to older generations of CPU, GPU, or whatever. By that time, maybe we'll also know more about what Intel and nvidia are doing also, and have a better picture overall.

Aggeed it's kinda like they're now having "RTX" for GTX 1000 series, for pretty much no reason than to say "we have it" I mean idk if there's any practical use for it on 1000 series but even if I'd imagine it to be really limited...