PC under performing multiple games (FF16/SH2R)

[Cornisgud]

Basically I've been troubleshooting my system not running games I think it's well above the recommended specs to play. After a day of troubleshooting and trying things I want to ask for advice on what I should try next, or if I'm missing something about my setup.

I recently had some trouble playing FF16 and the Silent Hill Remake on my PC and it looked like my bottleneck was the 8GB of VRAM (RTX 3070 TI). Both of these games ran, but stuttered in more open areas where a lot more was going on, even when lowering the settings. So I upgraded to the 4070 TI Super. The performance is even worse now.

I've reinstalled the graphics drivers, reinstalled the games, lowered the settings on both games, lower resolution of primary display and the game to 1080p, they're both installed on SSD's... And no matter what I do the games still perform poorly.

SH2R sits between 24-30 FPS and has massive stuttering spells and freezes that last 2-3 seconds. Outside the FPS sits around 15-20. About 9.5/16GB of VRAM is in use.

On FF16 FPS hovers between 12-18 at "mid" settings and goes into slow motion a ton. Uses about 9.0/16GB of VRAM consistently.

While running both games GPU sits at 97%+ usage. CPU sits at around 30-40% usage for both games but on FF16 it spikes to 100% for a few seconds every now and then, not sure why.

My setup:

Primary Display: 2560x1440p 144hz (I also have 2 1920x1080p monitors)
Processor: AMD Ryzen 7 5800X 8-Core Processor, 3801 Mhz, 16 Logical Processors
Video Card: NVIDIA GeForce RTX 4070 Ti SUPER (16GB VRAM)
Graphic Driver: 565.90 (Up to date in GeForce Experience)
Ram: 32 GB RAM
PSU: 800W Power Supply
Motherboard: Gigabyte B550 UD AC

I know that's long I just wanted to offer as much info as I can.

[Illusion of Progress]

r.linder の投稿を引用：

And this is why there needs to be more obvious pointers for first time builders to use the top slot as that's usually the only PCI-e x16 slot, the others have reduced bandwidth and will limit performance, to the untrained eye they look the same, but there's fewer pins.

What's amusing here is that most games don't need the amount of bandwidth that is available by the latest PCI Express versions (I mean, Cyberpunk 2077 was fine on PCI Express 1!?), which means it was even less obvious for any of us to guess as the possible cause. One of those threads you look back at and go "huh, I never would have guessed to think of that" even though it's pretty obvious it can cause it when you think of it.

[r.linder]

Illusion of Progress の投稿を引用：

r.linder の投稿を引用：

And this is why there needs to be more obvious pointers for first time builders to use the top slot as that's usually the only PCI-e x16 slot, the others have reduced bandwidth and will limit performance, to the untrained eye they look the same, but there's fewer pins.

What's amusing here is that most games don't need the amount of bandwidth that is available by the latest PCI Express versions (I mean, Cyberpunk 2077 was fine on PCI Express 1!?), which means it was even less obvious for any of us to guess as the possible cause. One of those threads you look back at and go "huh, I never would have guessed to think of that" even though it's pretty obvious it can cause it when you think of it.

It was PCI-e 3.0 1x, not PCI-e 1.0, there's a big difference, that x needs to be there to denote the lanes from the revision, causes confusion otherwise even though these motherboards shouldn't even support 1.0.

[Illusion of Progress]

r.linder の投稿を引用：

It was PCI-e 3.0 1x, not PCI-e 1.0, there's a big difference, that x needs to be there to denote the lanes from the revision, causes confusion otherwise even though these motherboards shouldn't even support 1.0.

Oh, I presumed that since they listed it as "link speed 2.5 GT/s" which is the amount of bandwidth PCI Express 1.0/1.1 has per single lane (3.0 would be 8 GT/s per lane).

[r.linder]

Illusion of Progress の投稿を引用：

r.linder の投稿を引用：

It was PCI-e 3.0 1x, not PCI-e 1.0, there's a big difference, that x needs to be there to denote the lanes from the revision, causes confusion otherwise even though these motherboards shouldn't even support 1.0.

Oh, I presumed that since they listed it as "link speed 2.5 GT/s" which is the amount of bandwidth PCI Express 1.0/1.1 has per single lane (3.0 would be 8 GT/s per lane).

They said;

Cornisgud の投稿を引用：

I think the issue was: The Gigabyte B550 UD AC motherboard advertises PCIE 4.0 x16, but actually only the top one is while the bottom 3 are PCIE 3.0 x1 (Maybe this is normal? I'm not 100%).

If you go to GIGABYTE's site, it says this on their specifications page for the B550 UD AC:
"4 x PCI Express x1 slots, integrated in the Chipset:

- Supporting PCIe 3.0 x1 mode"

I didn't see where they said the link speed exactly but I saw this first.

(BTW OP, yes it's normal for this motherboard, because it has so many slots but the supported lanes do not change because the slot was designed that way and because the CPU and chipset only support so many lanes, regardless of how many slots you have, so more slots = less bandwidth to share between them. So they made them 3.0 x1.)

[Illusion of Progress]

Yeah, I'm not exactly arguing which mode it was or wasn't running in, or what the motherboard does and doesn't support. I was simply going off where they said they saw "2.5 GT/s" and that's exactly what the transfer rate of PCI Express 1 is per lane, so I presumed that might be what it was running in. It could have been what they saw was wrong.

Either way, it was definitely running in a very reduced capacity and that was causing the issues.

[C1REX]

Illusion of Progress の投稿を引用：

r.linder の投稿を引用：

And this is why there needs to be more obvious pointers for first time builders to use the top slot as that's usually the only PCI-e x16 slot, the others have reduced bandwidth and will limit performance, to the untrained eye they look the same, but there's fewer pins.

What's amusing here is that most games don't need the amount of bandwidth that is available by the latest PCI Express versions (I mean, Cyberpunk 2077 was fine on PCI Express 1!?), which means it was even less obvious for any of us to guess as the possible cause. One of those threads you look back at and go "huh, I never would have guessed to think of that" even though it's pretty obvious it can cause it when you think of it.

I think I’ve seen about three people discussing this on the FF16 forum. If this game is uniquely demanding for bandwidth, then it might be due to the implementation of DirectStorage.

I share the opinion that motherboards don’t make it clear enough what ports people should use for their GPU and RAM. Especially RAM as using slot 2 and 4 is not intuitive.

[Illusion of Progress]

That sounds like a good guess. Silent Hill 2 might stream a lot of assets too, since it does have a lot of traversal stutter in general.

Amusingly, you want to know what game I've noticed uses a lot of PCI Express bandwidth compared to most others? Minecraft. I'm not kidding. The game isn't fill rate limited, but it can push a lot of vertex data (especially at high render distances) and might do a lot of draw calls.

It's funny, but when the thread starter mentioned 20 FPS to 30 FPS, one of my thoughts was "that sounds the frame rate range you often get when VRAM is full and it starts having to go over the PCI Express link for shared VRAM". I dropped the thought when they said the 16 GB RTX 4070 Ti Super had the same thing going on because no way was it filling VRAM, but in a way, I was sort of thinking in the right direction. it was still the PCI Express bandwidth holding it up... but only because it was limited to a very low amount, and only in the rare games that need a lot.

At least in the case of RAM, if it works in slot 1 and 3, then you suffer nothing. The problem with using slots 1 and 3 to begin with is the interference from slots 2 and 4 being "open ends" can make it harder to stabilize, but if it stabilizes/passes POST (and if you otherwise have no stability issues), then it's fine. Like, you won't lose a massive amount of performance from it compared to like they did in this scenario. They also do indicate that you should use slots 2 and 4. Is it intuitive, no, but there's at least a physical reason for why it's that way.

[Indigo_Sun]

Assuming xmp/expo set in Bios and DRAM set to 6000-30 in BIOS?