Dead Space Performance Jumps 46% with Resizable BAR Enabled

[btarunr]

The PCI Resizable BAR feature can have a major impact on the performance of "Dead Space" (the 2023 reboot). sxKYLE on Reddit discovered that his GeForce RTX 4080 "Ada" experienced a major performance jump from 76 FPS up to 111 FPS (+46%) with resizable BAR enabled. NVIDIA Profile Inspector was used to enable resizable BAR for the game. The feature allows the CPU to see the graphics card's video memory as a single, continuously-addressable block, rather than through 256 MB apertures. We can imagine how "Dead Space" in particular could benefit from the feature, as the game's campaign is a single continuous action sequence with assets being constantly streamed to the GPU (and no level load screens).



View at TechPowerUp Main Site | Source

 

[Space Lynx]

I thought resize bar was on by default? So all benches should already have that included in their benching.

I know W1zz told me one time that all modern systems have resize bar on by default, so this should be a null find.

 

[lemonadesoda]

Very significant improvement. I am surprised this makes such a large difference. Especially when most assets are already loaded into GPU memory. I would have thought the rebar helps more in the situation of making a smaller GB card perform a little closer to the performance of a larger GB card, ie. close the gap between an 8GB and a 16GB card.

But to increase performance by such a large percentage? There must be other driver optimisations added or bugs removed. This can't be just the rebar.

 

[nguyen]

I thought resize bar was on by default? So all benches should already have that included in their benching.

I know W1zz told me one time that all modern systems have resize bar on by default, so this should be a null find.

Nvidia drivers have a whitelist of Rebar games, currently Dead Space is not in the whitelist, so doing this trick can improve FPS on Ada massively

 

[R00kie]

I thought resize bar was on by default? So all benches should already have that included in their benching.

I know W1zz told me one time that all modern systems have resize bar on by default, so this should be a null find.

On system level yes, but it also has to be enabled for each and every game, which is what the dude in the article did

 

[Assimilator]

Incredibly useful, but omitted from news post, context as to why they were playing with ReBAR disabled (spoiler, they weren't):

is there any reason Nvidia barely ever updates the whitelist for reBAR? At this point, why not turn it on by default for all games for 40 series, and utilize a blacklist instead?

It is not so straightforward. We test games across multiple levels and across each new build of a game before release. So while some users may see a bump in performance in one level/map, if you try a different level/map, the game assets might be quite different and users may not benefit as much or worse experience a regression in FPS/frame times.

As for Dead Space remake, we are still testing ReBAR performance so there is a chance we may enable it in a future driver.

TL;DR while you may have ReBAR enabled in your BIOS, and GPU-Z may show it as Enabled, that doesn't mean your NVIDIA GPU will actually use it! It seems like ReBAR support is something that is baked into the drivers on a per-game basis, and NVIDIA chooses to blacklist all games by default from using ReBAR, while whitelisting only titles they have tested and found it works well with. This is in direct contrast to AMD who whitelist by default and blacklist non-working games; presumably Intel does the same as AMD.

 

[Space Lynx]

On system level yes, but it also has to be enabled for each and every game, which is what the dude in the article did

I had no idea... so are you saying all this time I have been gaming with rebar enabled in BIOS, my 6800 xt was actually not utilizing it? Where the hell do I even enable it for each game individually... I am so confused. wtf.

edit: nm just read @Assimilator post

 

[lemonadesoda]

Someone needs to re-test performance with the rebar on/off with these new drivers. There's no way 46% improvement comes from whitelisting the rebar. Or I'll eat my hat.

 

[chrcoluk]

Yeah its whitelist only, there is a way to do it via nvidia inspector as well as what is stated in this thread, which I think even quoted me in a old TPU article.

 

[R00kie]

Someone needs to re-test performance with the rebar on/off with these new drivers. There's no way 46% improvement comes from whitelisting the rebar. Or I'll eat my hat.

I'm gonna do that when I get home and post it here

 

[opinali]

The fact that ReBAR isn't yet enabled for all games on NVidia GPUs, or at least enabled by default (denylist instead of allowlist), is a sad joke. NVidia fans are quick to jump on Radeon driver issues but it's time to own up to this.

ReBAR is a strict improvement in the PCIe protocol with absolutely zero tradeoffs. Any scenario that regresses performance with ReBAR-on can only be caused by a deficiency somewhere (game engine, GPU driver, or GPU architecture - more likely the latter two, especially for recent releases). Intel's Arc GPUs have an even bigger dependency on ReBAR because it's a new arch so they didn't bother to optimize the ReBAR-off case, and that's the right thing to do since every minimally current CPU/chipset/motherboard also supports ReBAR. I suspect RTX GPUs are the opposite, they probably still have drivers and maybe even memory controllers over-tuned for the ReBAR-off case, causing more regressions than any competing GPU if you turn it on for all games.

 

[nguyen]

The fact that ReBAR isn't yet enabled for all games on NVidia GPUs, or at least enabled by default (denylist instead of allowlist), is a sad joke. NVidia fans are quick to jump on Radeon driver issues but it's time to own up to this.

ReBAR is a strict improvement in the PCIe protocol with absolutely zero tradeoffs. Any scenario that regresses performance with ReBAR-on can only be caused by a deficiency somewhere (game engine, GPU driver, or GPU architecture - more likely the latter two, especially for recent releases). Intel's Arc GPUs have an even bigger dependency on ReBAR because it's a new arch so they didn't bother to optimize the ReBAR-off case, and that's the right thing to do since every minimally current CPU/chipset/motherboard also supports ReBAR. I suspect RTX GPUs are the opposite, they probably still have drivers and maybe even memory controllers over-tuned for the ReBAR-off case, causing more regressions than any competing GPU if you turn it on for all games.

Whitelisting is the safer default if you think about it, Nvidia is willing to sacrifice performance so that every Geforce owner can have a stable gaming experience, tech enthusiasts can of course override the defaults and try Rebar on their own.

AMD and Intel are taking riskier approach, in order to compete on price to perf, at the cost of possible system instabilities that will turn away people who just want to play game and not tinkering with their system all the time.

Kinda funny that you think leaving out performance for the sake of stability is a problem

 

[Legacy-ZA]

Someone needs to re-test performance with the rebar on/off with these new drivers. There's no way 46% improvement comes from whitelisting the rebar. Or I'll eat my hat.

Time to gather some spices & drinks, I think it's gonna be a hard swallow. :/

I can't wait for DirectStorage to come into play as well, it's about damn time.

 

[Sithaer]

TL;DR while you may have ReBAR enabled in your BIOS, and GPU-Z may show it as Enabled, that doesn't mean your NVIDIA GPU will actually use it! It seems like ReBAR support is something that is baked into the drivers on a per-game basis, and NVIDIA chooses to blacklist all games by default from using ReBAR, while whitelisting only titles they have tested and found it works well with. This is in direct contrast to AMD who whitelist by default and blacklist non-working games; presumably Intel does the same as AMD.

Well thats good to know, honestly I've never really read after Rebar in detail and ever since I have a build that supports it I had it enabled most of the time but I did notice that its actually causing some issues in certain games so I turned it off after that and just turned it back on yesterday to check in Cyberpunk. 'only game I'm playing currently'

 

[Garrus]

Does this mean the Dead Space performance review that was just published at TechPowerUp needs to be redone?

Anyone?

 

[roberto888]

Does this mean the Dead Space performance review that was just published at TechPowerUp needs to be redone?

Anyone?

Since it's not the out-of-box experience, I don't think it would make sense.

 

[Vya Domus]

The faster storage on consoles is finally starting to make a difference, they can load in and out of memory data way faster and the memory is unified, these performance penalties from disabling resizable bar could become much more frequent.

AMD and Intel are taking riskier approach, in order to compete on price to perf, at the cost of possible system instabilities that will turn away people who just want to play game and not tinkering with their system all the time.

Kinda funny that you think leaving out performance for the sake of stability is a problem

Hilarious comment, the only issues I've ever heard with respects to resizable bar have been on Nvidia cards, you know, the ones that take the "safer" approach.

It's a problem because they're leaving out performance and it's not any more stable. AMD has resizable bar working with no problems and Nvidia doesn't, which is why they whitelist everything by default, it's as simple as that.

 

[Legacy-ZA]

Interesting to note; the power delivery to the CPU/GPU is vastly different between the two screenshots.

 

[W1zzard]

Interesting to note; the power delivery to the CPU/GPU is vastly different between the two screenshots.

Yeah benchmarking doesn't work like that .. has to be the same scene .. looking at a wall will always give you higher FPS.

Not saying that this claim is invalid, but these two screenshots aren't a good comparison

Don't have any data right now, finishing my rebench of all cards all games on 13900K

 

[Legacy-ZA]

Yeah benchmarking doesn't work like that .. has to be the same scene .. looking at a wall will always give you higher FPS.

Not saying that this claim is invalid, but these two screenshots aren't a good comparison

Don't have any data right now, finishing my rebench of all cards all games on 13900K

Awesome, can't wait to see the results.

 

[swirl09]

A bit surprised to see this story here. I saw that post yesterday on reddit and my first thoughts were that the shots are different so the uplift couldnt be properly measured from that, also the power difference is noticeable (which yes could be because its able to better utilize under ReBar - but it does need to be there at all to be tapped in the first place, which wont always be the case.) An additional 100w is real.

Hopefully there is a performance bump and more games make use of ReBar, but I will wait for proper testing before buying the %.

 

[Assimilator]

It's a problem because they [NVIDIA] are leaving out performance and it's not any more stable.

but then you say:

AMD has resizable bar working with no problems and Nvidia doesn't

Is NVIDIA disabling ReBAR by default for no good reason; or is it because their implementation isn't perfect, they know that, and they're being cautious about enabling it to prevent users experiencing instability? Because it can't be both.

 

[TheDeeGee]

And that is why i love NVIDIA.

NVIDIA Inspector is an invaluable tool.

 

[Vya Domus]

Is NVIDIA disabling ReBAR by default for no good reason; or is it because their implementation isn't perfect, they know that, and they're being cautious about enabling it to prevent users experiencing instability? Because it can't be both.

It can, because issues with it still exist, if you google "resizable bar crash" you're mostly gonna get results from people with Nvidia cards. So despite their whitelisting efforts the instability is still there on top of the potentially lost performance.

Having to disable a feature by default but then you still get both instability and potentially lost performance is "no good reason" in my book, they should just get on it with it and fix it.

A bit surprised to see this story here. I saw that post yesterday on reddit and my first thoughts were that the shots are different so the uplift couldnt be properly measured from that, also the power difference is noticeable (which yes could be because its able to better utilize under ReBar - but it does need to be there at all to be tapped in the first place, which wont always be the case.)

Interesting to note; the power delivery to the CPU/GPU is vastly different between the two screenshots.

Slow memory transfers are causing more stalls in the rendering pipeline, the lower power usage is exactly what you'd expect.

 

[ExplodingCaps]

Just tried it and see substantial improvement, but nowhere near 40%. Mine went from 75 fps @4k with 100% usage and 330-360 watt power usage to 90-95 fps with 100% usage to 390-410 watt power usage. Using 4090 @4k resolution.