NVIDIA PCI-Express Resizable BAR Performance Test

[chrcoluk]

Similar to my own testing, I have been testing mainly dx9/dx11 stuff since all the reviewers mostly focus on dx12.

I also tested 3dmark timespy.

My conclusion is the majority of stuff gets 0-2% benefit.

Also I have discovered having REBAR enabled on the system, even if its disabled in the game, there is a very small overhead, its small maybe about 0.5 to 1%. Usually the REBAR gain (if enabled in inspector) will cancel it out. So the net gain over a system with REBAR disabled is actually closer to 0-0.5%.

I will speculate since this feature was planned by AMD, they likely contacted some game dev's had some optimisation done and hence we have a bunch of new titles with decent gains (that can showcased by reviewers) whilst most other titles is barely anything at all. In that respect its kind of like DLSS, benefits newer stuff but not older stuff.

 

[efikkan]

Also I have discovered having REBAR enabled on the system, even if its disabled in the game, there is a very small overhead, its small maybe about 0.5 to 1%. Usually the REBAR gain (if enabled in inspector) will cancel it out. So the net gain over a system with REBAR disabled is actually closer to 0-0.5%.

I will speculate since this feature was planned by AMD, they likely contacted some game dev's had some optimisation done and hence we have a bunch of new titles with decent gains (that can showcased by reviewers) whilst most other titles is barely anything at all. In that respect its kind of like DLSS, benefits newer stuff but not older stuff.

Anything this low level isn't controlled directly by the game engines, this is all done by the graphics drivers.

I am worried though, if any of them (AMD or Nvidia) added workarounds in their drivers for specific games. Such "optimizations" are never a good idea, they add complexity to drivers, potential bugs, and generally adds to driver overhead.

 

[voltage]

"but most motherboard vendors for whatever reason restricted their Resizable BAR enabling BIOS updates to the 300-series chipset, or 8th Gen "Coffee Lake" (and later)"

maybe these motherboard vendors bought amd shares at the time. after all, that seems like intentional performance restriction, why else would they choose to restrict Intel's performance?

 

[toilet pepper]

Enabling ReBAR on my system brought out some instability I didn't know I had when playing Cyberpunk. I guess thanks ReBAR for helping me realize my infinity fabric overclock wasn't stable. Now that everything is working it does feel a little bit better to play, but that could also be the Cyberpunk 1.2 patch that came out. Oh well, happy to get any little bit of performance I can.

I've been seeing posts in reddit about this. Basically, if you have your OC pushed to the limit and stable and you enable ReBar - you have to dial your OC a notch to make it stable.

 

[cellar door]

Enabling ReBAR on my system brought out some instability I didn't know I had when playing Cyberpunk. I guess thanks ReBAR for helping me realize my infinity fabric overclock wasn't stable. Now that everything is working it does feel a little bit better to play, but that could also be the Cyberpunk 1.2 patch that came out. Oh well, happy to get any little bit of performance I can.

Same exact experience here as well - I went from 1900 to 1866mhz to correct it.

 

[so11ex]

Great review, thanks!

 

[lightning70]

Every performance increase from free is good. But it has increased a little.

 

[Fierce Guppy]

Well, it was interesting to learn that even the Haswell CPU supports ReBAR in hardware. So it's a matter of firmware and microcode support limiting the availability of ReBAR.

 

[Xenom]

A tech that will be useable to the general person in 2-5 years when the cards are available

 

[qubit]

Wow, makes a real difference at 1080p, not so much at 4K where hard performance limitations of the graphics card, like fill rate, have a bigger effect. Worth having if you can get it, though. My next rig will be built with this in mind. Should be standard by then, anyway.

Great review W1z, as always.

 

[Yopis]

This test was a fail. at least for Red Dead if you actually play the game then you'll see that the performance increases a lot larger this is one of the things that people have figured out that reviewers that didn't take any time except running The in-game benchmark didn't realize try it again actually playing the game in the same spot and you'll see market improvement and 1% lows and regular frame rate I'm seeing five to six frame increase a lot larger than the in-game benchmark shows me don't be lazy and actually try the game as well. I'm referring to 4k.

I dont think 2% will make people wanting it that badly, if it was 10% then it would be a different story. Thanks for the tests @wizard.

The test were flawed.

 

[R0H1T]

I will speculate since this feature was planned by AMD

Or maybe it has something to do with that software scheduler thing & actually better planning from AMD in the past?

why else would they choose to restrict Intel's performance?

You're kidding right? You remember the last BIOS updates for Haswell/Broadwell gen boards? No(?) ~ it was for smeltdown & probably only because Intel forced them to! You think board makers will update BIOSes for boards released 5-7 years back? It's not like they can be used for CPUs released even 2 years back, unlike AMD & AM4. You're almost funny, if not for the pathetic thing Intel does forcing users to update mobos nearly each year!

 

[Night]

If this was supposed to be marketing for the GPU, why even release it now? By the time GPUs get available again, no one will remember Resizable BAR. Especially with such small improvements on select titles, it's important as much as Raytracing is.

 

[W1zzard]

The in-game benchmark

I'm not using the integrated benchmark

 

[stimpy88]

Thanks w1zzard.

 

[TheoneandonlyMrK]

If this was supposed to be marketing for the GPU, why even release it now? By the time GPUs get available again, no one will remember Resizable BAR. Especially with such small improvements on select titles, it's important as much as Raytracing is.

I get your point, but definitely disagree on how important it is, Raytracing is over hyped at the moment(IMHO) and this isn't anywhere near as important to performance or the end game Feel or look.

 

[Ravenas]

MSI hasn't offered a BIOS update to their X570 MSI ACE MEG board yet. Only for SAM.

 

[Sora]

You can use NVIDIA Profile Inspector to modify Resizable BAR game support settings. Credit goes to @chrcoluk.

yeah because i didn't already post this info to TPU 4 weeks ago or anything.

Or maybe it has something to do with that software scheduler thing & actually better planning from AMD in the past?

Oh, ffs, please don't cite anyone about the bloody scheduler, the only thing nvidia removed was the data hazard block leaving instruction dispatch up th the compiler, the rest of the bloody scheduler is still hardware based.

I would love to punch Ian in the face for misrepresenting what he was told in the kepler review all those years ago.

Enabling ReBAR on my system brought out some instability I didn't know I had when playing Cyberpunk. I guess thanks ReBAR for helping me realize my infinity fabric overclock wasn't stable. Now that everything is working it does feel a little bit better to play, but that could also be the Cyberpunk 1.2 patch that came out. Oh well, happy to get any little bit of performance I can.

If you find any games get stuttery, you can thank Ryzens poor grading of IO dies and the numerous cpu's that spam the pcie controller with errors.

 

[blu3dragon]

Anyone seeing

Similar to my own testing, I have been testing mainly dx9/dx11 stuff since all the reviewers mostly focus on dx12.

I also tested 3dmark timespy.

My conclusion is the majority of stuff gets 0-2% benefit.

Also I have discovered having REBAR enabled on the system, even if its disabled in the game, there is a very small overhead, its small maybe about 0.5 to 1%. Usually the REBAR gain (if enabled in inspector) will cancel it out. So the net gain over a system with REBAR disabled is actually closer to 0-0.5%.

I will speculate since this feature was planned by AMD, they likely contacted some game dev's had some optimisation done and hence we have a bunch of new titles with decent gains (that can showcased by reviewers) whilst most other titles is barely anything at all. In that respect its kind of like DLSS, benefits newer stuff but not older stuff.

Did you see an improvement in timespy?

I tried it today, and the GPU portion was basically within margin of error, but my cpu score dropped a bit when I used profile inspector to enable it.

The one game I tried forcing it on for showed zero difference. It's an older dx11 architecture and doesn't use a huge amount of memory.

 

[nguyen]

Anyone seeing
Did you see an improvement in timespy?

I tried it today, and the GPU portion was basically within margin of error, but my cpu score dropped a bit when I used profile inspector to enable it.

The one game I tried forcing it on for showed zero difference. It's an older dx11 architecture and doesn't use a huge amount of memory.

Try console ported games, the idea behind Resizeable Bar or SAM is because console CPU have direct access to VRAM after all.
Games that are made on PC originally are unlikely to gain much from Re-Bar or SAM

 

[Sora]

Try console ported games, the idea behind Resizeable Bar or SAM is because console CPU have direct access to VRAM after all.

no it isn't.

 

[Metroid]

MSI hasn't offered a BIOS update to their X570 MSI ACE MEG board yet. Only for SAM.

I have that msi b450 gaming plus and I can confirm it got a bios for it and in the bios there is an option to enable rebar, however I have not done the test yet to confirm if it works with the 3xxx series, i have a 3080 but not using as main gpu right now, I'm using a gtx 1070 as main and rebar does not work with pascal yet, yeah only 3xxx series but like i said i have to confirm it.

 

[chrcoluk]

Anyone seeing
Did you see an improvement in timespy?

I tried it today, and the GPU portion was basically within margin of error, but my cpu score dropped a bit when I used profile inspector to enable it.

The one game I tried forcing it on for showed zero difference. It's an older dx11 architecture and doesn't use a huge amount of memory.

yes but its tiny, basically it cancelled out the bios overheads.

 

[Prima.Vera]

Wait, except for the driver, you also have to re-flash your GPU, not only your Mobo????

 

[W1zzard]

Wait, except for the driver, you also have to re-flash your GPU, not only your Mobo????

Correct, and you might have to reinstall Windows or MBR2GPT