ASUS, Asrock, and Biostar Have Enabled Resizable BAR On Their Z490 Motherboards

[SFFDesigns]

6 minutes ago, GabenJr said:

In the end, it'll have been a short-lived gimmick that gave them an advantage over Nvidia, similar to how "RTX" gave Nvidia an early advantage in ray tracing despite it also being a standard thing in both DirectX and Vulkan.

I agree. I do think raytracing is a bigger advantage than this though, because it can't just be enabled through the driver like Nvidia is doing with "SAM" (I mean it can, it just won't be hardware accelerated). Nvidia getting raytracing earlier gives them a "headstart" that will likely help them in the long run as raytracing becomes more and more useful. 

 

This does give AMD one selling point they can use to try to regain a spot in the high-end tho

[Kisai]

On 12/4/2020 at 10:12 AM, Hymenopus_Coronatus said:

I agree. I do think raytracing is a bigger advantage than this though, because it can't just be enabled through the driver like Nvidia is doing with "SAM" (I mean it can, it just won't be hardware accelerated). Nvidia getting raytracing earlier gives them a "headstart" that will likely help them in the long run as raytracing becomes more and more useful. 

 

This does give AMD one selling point they can use to try to regain a spot in the high-end tho

It would make far more of a difference in neural net work, since the bottleneck is ram size. Though it's a bit speculative since most NN stuff is done on Linux machines because of how much of a pain Windows is.

 

 

[Mark Kaine]

On 12/4/2020 at 3:29 PM, Hymenopus_Coronatus said:

Asrock has released a BIOS update for the Z490 Taichi.

 

Here's the headline feature: "add C.A.M. (Clever Access Memory) function.". The naming is just pure gold 

 

ASRock > Z490 Taichi

shoulda been S. C. A. M. (super clever access memory) but I get it, sniding nzxt (and their "cam" / spy app) gets my approval at any point also! 

 

[Random_Person1234]

Biostar has also enabled SAM on Intel 400 series motherboards.

https://videocardz.com/press-release/biostar-enables-amd-smart-access-memory-on-intel-400-series-motherboards

[Mark Kaine]

btw what's stopping mobo manufacturers from doing the same for older Ryzen CPUs? 

[porina]

8 hours ago, Mark Kaine said:

btw what's stopping mobo manufacturers from doing the same for older Ryzen CPUs? 

Historically AMD have been a bit more controlling via AGESA updates what is or isn't allowed. Look at PCIe 4.0 support on 400 chipset for example. I'd expect Zen 3 support on 300 chipset to go the same way soon. So if a mobo manufacturer were to do this somehow without AMD's blessing, let's just say it wont help their relationship.

 

There was a rumour floating around that older AMD CPUs lacked some instruction or feature required for its implementation, but without confirming it I saw something more recently suggesting that wasn't the case. I think in short we just have to wait for AMD to give their official response, and it is safest to assume older CPUs wont get it until we know otherwise.

[igormp]

On 12/6/2020 at 2:55 AM, Kisai said:

It would make far more of a difference in neural net work, since the bottleneck is ram size. Though it's a bit speculative since most NN stuff is done on Linux machines because of how much of a pain Windows is.

 

 

No, it doesn't. Above 4g mmio decoding has been available on linux since 2017, and batch sizes are usually allocated in many small blocks, not huge blocks at once (where such feature makes sense).

This is a feature that's really useful for NICs and FPGAs that move tons of big chunks of data (4gb+ as a single piece of data) at once.

 

On 12/11/2020 at 11:07 PM, Mark Kaine said:

btw what's stopping mobo manufacturers from doing the same for older Ryzen CPUs? 

Older Ryzen CPUs lack an actual hardware implementation to some BMI2 instructions (PEXT and PDEP), which are "emulated" through microcode. This makes it so that the latency required to perform some PCIe manipulations required with reBAR are really high (10~20x higher than a hw implementation), which makes the benefit of having rebar almost null for games. iirc, there's some limitation on windows that makes it not work that way, but I'm not so sure about that.

 

Anyway, Intel CPUs have had this since their 4th core µarch, iirc, and AMD only made it a hardware implementation on Zen 3.

 

As to your actual question, nothing, it has already worked since 2017 on linux, all you needed was a mobo with the option for "Above 4g decoding" (which they now renamed to rebar) or something like that, enable it and you're good to go. It also works with older Zen CPUs.

[Falkentyne]

6 hours ago, igormp said:

No, it doesn't. Above 4g mmio decoding has been available on linux since 2017, and batch sizes are usually allocated in many small blocks, not huge blocks at once (where such feature makes sense).

This is a feature that's really useful for NICs and FPGAs that move tons of big chunks of data (4gb+ as a single piece of data) at once.

 

Older Ryzen CPUs lack an actual hardware implementation to some BMI2 instructions (PEXT and PDEP), which are "emulated" through microcode. This makes it so that the latency required to perform some PCIe manipulations required with reBAR are really high (10~20x higher than a hw implementation), which makes the benefit of having rebar almost null for games. iirc, there's some limitation on windows that makes it not work that way, but I'm not so sure about that.

 

Anyway, Intel CPUs have had this since their 4th core µarch, iirc, and AMD only made it a hardware implementation on Zen 3.

 

As to your actual question, nothing, it has already worked since 2017 on linux, all you needed was a mobo with the option for "Above 4g decoding" (which they now renamed to rebar) or something like that, enable it and you're good to go. It also works with older Zen CPUs.

Then how come there is a option for BOTH 4G decoding (which motherboards have had for a long time) and now "resizebar BAR" if they are the exact same thing?

If they were identical, they wouldn't have an option for both, with BAR being added in addition to the existing "above 4G decoding."

[igormp]

4 hours ago, Falkentyne said:

Then how come there is a option for BOTH 4G decoding (which motherboards have had for a long time) and now "resizebar BAR" if they are the exact same thing?

If they were identical, they wouldn't have an option for both, with BAR being added in addition to the existing "above 4G decoding."

Probably something specific to Windows in order to report BMI2 or something like that. As I said before, it already worked on linux with "above 4g decoding" only.