Nvidia Resizable BAR Tested, Benchmarked

Irata

Posts: 1,434   +2,315
So, I guess the results put the „SAM is just turning on a standard PCIe feature“ to rest as obviously there is a very noticeable performance difference between AMD‘s and nVidia‘s implementation or

Quick question Steve: which Radeon drivers were used for the RX6800 results ?
 

Kosmoz

Posts: 191   +293
Nobody is buying a 3080 or a 6800XT and plugging it into a 1080p monitor and turning off RT.
Hey watch you tongue! I would totally buy a 3080/6800XT for my 1080p monitor, although I would use it maxed out in games, so RT ON.

I would do that if I could buy one (at MSRP that is) because I like to buy a good (overpowered GPU) to last me 4-5 years at 1080p Ultra. I don't like to upgrade every 1-2 years, nor do I like to play 1st year at 4k, next year at 1440p and 3rd year at 1080p because that's how fast tech gets old when stressed vs the latest AAA games...

But seeing how upset I am at nvidia for their past and recent lies, I would buy the RX 6800/XT before giving the nvidia cards a 2nd chance, though...
 

zamroni111

Posts: 182   +129
We need to see this retested on rtx 3060.
3070, 3080 & 3090 are not made from the ground up to support resizable bar
 

Steve

Posts: 2,728   +2,802
Staff member
I prefer to simply go for the most VRAM.
My 3090 FTW3 is 24GB.
Easier to just "brute force" my way forward.
I ain't got time for overclocking and memory gimmicks.
Buying a 3090 is 100% the smart way to go #EmbarrassingFlex

Should be obvious, anyone playing games.
Wasting time in the BIOS is probably more constructive ;)
 

DrSuess

Posts: 98   +68
"It’s worth noting that in order to enable/disable resizable BAR with a Radeon or GeForce graphics card, you need to reboot the system, enter the BIOS, and toggle it on or off there. So that’s not exactly a practical solution and we’d argue doing so means the performance gains are no longer free, they come at the expense of your time and energy"

This is the best reason not to enable it.
Who has +-2 minute these days?
Is there a compelling reason to disable it after you enabled it? Particularly if you are not experiencing any incompatibility issues.
 

Mr Majestyk

Posts: 726   +616
So Nvidia trash talks and AMD just delivers. This was embarrassing result for Nvidia. I notice the article which started off gushing about Nvidia's claims didn't slam them when the results were collated. If AMD had made the same grandiose claims and delivered almost sweet fa in most cases they would have been slammed. You went very easy on them in the summary
 
Guys @ Techspot, ReBar is a feature if you have lots of VRAM to spare. A 3080 is not the right card to test this on. It makes way more sense to test the 3090 here or even the upcoming 3080 Ti with 12GB VRAM. All Radeons benefit the most from this feature because they have 16GB VRAM etc.
 

Steve

Posts: 2,728   +2,802
Staff member
Guys @ Techspot, ReBar is a feature if you have lots of VRAM to spare. A 3080 is not the right card to test this on. It makes way more sense to test the 3090 here or even the upcoming 3080 Ti with 12GB VRAM. All Radeons benefit the most from this feature because they have 16GB VRAM etc.
I'm at a lost here, why exactly do you think having more VRAM makes Resizable BAR work better? What is your understanding of how this feature works?
 

jpuroila

Posts: 332   +183
Is there a compelling reason to disable it after you enabled it? Particularly if you are not experiencing any incompatibility issues.
10% performance regression seems to be a pretty compelling reason if you found it worth enabling in the first place.
 

cliffordcooley

Posts: 12,739   +6,106
Hmm, guess that is where nVidia drivers doing so automatically is more than just a convenience. That is if there is no way to easily toggle manually.
 

quadibloc

Posts: 281   +170
I remember looking up information about the Xeon Phi card, and learned that you need to enable Resizable BAR to make use of it. So it's something Intel has had available, at least with its Xeon chips, for quite some time. Understandably, since most gamers don't use Xeon systems, it hadn't been worthwhile for Nvidia to include this on their graphics cards until AMD started including it on their Ryzen 5000 series.
But this makes the news story about AMD graciously providing technical assistance to Nvidia - and Intel (!) - to achieve compatibility with Smart Access Memory a bit puzzling. Of course, no doubt there's a simple explanation, such as SAM being a partial implementation of Resizable BAR, so there are some parts that have to be avoided for compatibility.
 

neeyik

Posts: 1,876   +2,190
Staff member
Theoretically, it's possible to change BAR size, without recourse to fiddling about with the BIOS. It all depends on the graphics card and the system. Take Nvidia's Tesla models, as an example: these can be switched between graphics and compute mode, provided the hardware and software environment permits it. In graphics mode, BAR size defaults to 256 MB, whereas in compute mode it's 8 GB. It's by no means a simple toggle in a GUI, prevents the GPU from acting as a primary display device, and requires a full system reboot, but it ultimately implies that a BAR size reconfiguration via software running in the OS is possible.
 

Markoni35

Posts: 1,074   +440
How come a 3080 was getting ~100fps at 4K and not its getting nearly 150fps with suposedly the same settings ?? Almost 50% more..
And all other outlets also get around ~100fps at 4K high settings . Its looks like you are running at 4K med , not high settings.
https://www.techspot.com/review/2099-geforce-rtx-3080/

It's even weirder that nobody except me took your comment seriously. And it IS a serious matter. I've seen this kind of thing before. Same card, same game, sudden change of performance. Newer drivers cannot explain the 50% change in performance. CPU can't explain 50% change in performance at 4K, where GPU is the bottleneck. It has to be something else.
 

Steve

Posts: 2,728   +2,802
Staff member
It's even weirder that nobody except me took your comment seriously. And it IS a serious matter. I've seen this kind of thing before. Same card, same game, sudden change of performance. Newer drivers cannot explain the 50% change in performance. CPU can't explain 50% change in performance at 4K, where GPU is the bottleneck. It has to be something else.
What game are we talking about here?
 

Markoni35

Posts: 1,074   +440
What game are we talking about here?
Death Stranding in 4K. In the latest article the GPU had around 150 fps, while in this older one:
...it had around 100 fps.
 

Steve

Posts: 2,728   +2,802
Staff member
Death Stranding in 4K. In the latest article the GPU had around 150 fps, while in this older one:
...it had around 100 fps.
Did you notice that the 1440p results changed as well? Did you also notice that the test system also completely changed?

Hint: compare the 5950X and 3950X below.

1618892148923.png

And finally, yes in addition to all that the game has seen performance updates as well.
 

Steve

Posts: 2,728   +2,802
Staff member
"The way AMD promoted and implemented Resizable BAR on Radeon 6000 GPUs, it required to be paired with a Ryzen 5000 series CPU and newer 500 series motherboard"
It's false. Resizable BAR works with Ryzen 3000 and chipset 400.
https://www.amd.com/en/technologies/smart-access-memory
That's not what the sentance says. AMD only supported SAM on Ryzen 5000 series CPUs with a 500 series chipset with a Radeon 6000 GPU when it was first implemented. The Ryzen 3000/400 chipset support came later.