Baldur's Gate 3 Performance Stress Test - Can Your CPU Hold Up?
HTML-код
- Опубликовано: 17 авг 2023
- In the wake of our initial Baldur's Gate 3 review coverage, questions were asked about performance in the game's most stressful area - the city of Baldur's Gate itself, as found in Act 3. In this video, Alex tackles this head-on, finding that the more dense environment is particularly challenging to your CPU, with older chips unlikely to maintain a decent level of performance. By the way, as our initial optimised settings were GPU-based, they remain the same here - and there's not much scalability in settings on the CPU side, as this content shall reveal.
To get access to the free-camera photo mode mod used in this video, visit Frans Bouma's Patreon: / otis_inf
Subscribe for more Digital Foundry: bit.ly/DFSubscribe
Join the DF Patreon to support the team more directly and to get access to everything we do via pristine quality downloads: bit.ly/3jEGjvx
Want some DF-branded tee-shirts, mugs, hoodies or pullovers? Check out our store: bit.ly/2BqRTt0
For commercial enquiries, please contact business@digitalfoundry.net - Игры
Respect to Digital Foundry for going back. Most wouldn't put in the effort.
Well, it isnt a waterblock 😂
Because most people have REAL jobs....
But adding some Chapters into the video is a bridge too far apparently :P
@@prehanramsamy6728 Wait you are saying this isn't a 'real' job?
It would cost them a lot of resources and having a person sit there
The game is constantly tracking sight cones for all the characters in the game, but it's limited to a circular kernel around the active player character. When you're spinning the player character around even in that little circle, it's enough to move one tile off from where it previously was on the internal grid, so the kernel gets updated and all the sight cones need recalculated. Obviously it's pretty easy to think of "optimizations" to this implementation, but they all come with their own trade-offs. I'll be eager to see what Larian decides to do.
If this is what is actually causing all the bad performance it should be a pretty easy fix for them to stop this tracking for non essential NPCs... or to start it when you go in stealth mode.
It seems really foolish of them to do this considering how much of a pain it would be to CPUs. (While I'm at it might as shamelessly plug my own video on the subject, if you want to check it let me know what you think).
underrated comment right here
@@KeyholeDweller The NPC can see me stealing things behind walls if I'm not in stealth mode. Guess they have already "optimized" it?
that would be "they heard it" simulation
@@Teiby3
Kudos to you for making a follow up video! I am running the game on a Ryzen 5 2600 and I am getting an average of 27 FPS in act 3 compared to the 50+ in acts 1 and 2.
What res are you at if i may ask
Are you playing on fullscreen ? I have R5 3600 and it actually helped me to get more FPS in game.
People will literally tell me its my fault i have a 5600x amd Same issue bg3 Fans are crazy in the head
CPU usage in this game is weird but if you don't move you have more fps kinda weird though thanks god it's turn based so it doesn't bother me that much what gpu are you using btw ?
I went for locked 30 fps because I was getting drops from 50-60 to 30ish
30 FPS is just a homage to Baldur's Gate 1 framerate. ;)
Why, it didn't run at 30fps
Larian take staying true to the material to a whole new level :p
@@simon_198730 FPS was the default max. Furthermore game speed was tied to framerate, so increasing it just made things speedy instead of smooth.
Animations in BG were with less than 30 fps.
@@nifftbatuff676 yep, probably 24 or 25
Going from a consistent 120fps from Act 1 to Act 2 then suddenly dropping to 40-60fps in Act 3 is a very jarring experience.
Concordo, parei de jogar por enquanto porque está muito ruim o desempenho no final do ato 2 e no ato 3 está uma droga.
I did not play the game yet but it would be enough to make me quit playing entirely. I don't own a high refresh monitor to look at a fucking slideshow.
@@michaelzomsuv3631 while the performance drop is embarassing and should be fixed, calling 40-60 fps a slideshow is completely out of touch.
@@vanjazed7021 Then enjoy your slideshow, I would just uninstall.
@@michaelzomsuv3631 kid, it's bedtime
The quirks with hyperthreading is strange indeed.
Honestly you should make a video about hyperthreading and multi-processing in video games. I'd love to learn about the challenges and possible solutions for better CPU performance. CPU performance has been rough for many recent releases.
oh those 4x games could really use some more cores/threads utilization i'm sure. tho i know it's akin to making a deal with a hag.
My speculation: this is old school multithreading where each core is pinned to a specific logic (sound, ai, physic, etc). So the game is developed with a set number of core (6 is the number of cores accessible on the ps5). Pros: easier to code, can be more performant for a given number of cores. Cons: doesn't scale automatically to a higher number of cores.
@@taaazzzzzzz well, could be the case here. But this is more like the exception to the rule. Many engines like Frostbyte, ID Tech,Unreal,etc scale well beyond 6 cores, up to 16 at least in the case of Frostbyte. The fact the consoles have max 6 cores doesnt help much either
Speculation: but I can imagine hyper threading hurting when very work heavy threads end up on the same physical CPU core.
So if you have the main game thread, a rendering thread, and a networking thread + a CPU with 2 physical cores that becomes 4 logical cores (from hyperthreading), the CPU scheduler might accidentally put the main thread and rendering thread on the same physical core (because hyper threading makes it look like two separate ones) while the networking thread gets a core to itself and spends most of its time doing nothing.
@@TheCrewExpendable only specific workloads not typical to video games would have a problem with two on a physical core, and Windows/Linux task management gets around that easily, unless you are saturating all cores with similar workloads of course.
There's an option called "Dynamic Crowds" in the menu. I assumed it automatically reduced clutter NPCs to preserve performance if needed, but the tooltip says it makes NPCs more dynamic. I don't know what exactly that entails, maybe you can reach out to the devs. At any rate, I would have liked to see if this option has an impact on performance particularly in the areas you highlighted in the video!
It will stop pathfinding on NPCs and they will go through characters now instead of avoiding them
@@Hree Ah! Yeah, that should lighten the CPU load quite a bit, one would think.
xD Can't really feel difference between 40-45 fps I have with it on or off. Smoothness feels like im sliding of sand instead of butter
@@xPvtMurphyxNo, just a liiitle tiny bit. I can't tell what's different when you turn it off, so...
@@Hreeit would affect frame-times not framerate
The performance issues hit my enjoyment of the game like a truck when I got to Act 3. This is the game I've enjoyed the most in many many years, but now I can't really enjoy it as much because my 3900X PC can't run the game above 30fps, and what's more frustrating is that since it's a CPU issue, lowering settings doesn't do anything, you're stuck in poor performance hell with nothing to do about it regardless. I really hope a big performance patch is coming soon, because right now I feel like the game is borderline unplayable for me.
sell your 3900x and buy 7600 ,13600k or just a am4 5700
@@AK-tf3fc3900x is a gpod cpu, sounds more like a problem with the game itself
Wish you could have tested with a 5800X3d or 7800X3d to see if the extra cache can alleviate the spikes.
7800X3D is by far the fastest CPU for this game. Minimum fps: 140 😊
Source: pc gamers hardware de
I,m running a 5800x3d and it does not help the stuttering in my case.
Reportedly the 5800x3d smokes the Intel 13900.
it doesn't@@raggedcritical
No point, they already have high end cpu test (i9 12900k), and r7 5800x3D is in many cases faster than this I9, so it's kinda pointless test. More cache can't really help with these scenarios.
I mean, it makes sense. It's 100% a pathfinding thing, btw. Notice no characters EVER walk through each other. They always, always ALWAYS walk around other characters and obstacles, and this needs to be calculated and updated every time any character (including summoned ones) moves, for every AI character nearby, even those not in the scene. It's really noticeable after he summons the zombies.
It doen't make sense, the game could utilize the Ryzen 5 3600 at 100% cpu utilization and avoid the low gpu usage and bad fps. This is bad optimization.
@@ji-tansome of the developers better have something like that
Have you looked at the CPU utilization? 1 thread gets hammered, the rest of the CPU is just chilling at a low usage - the usual sign of sloppy work. Multithreading is just plain bad in this game when a 100%+ increase in core count only results in a 30% increase in frames (and only because it went from 4 cores to 8) and the nothing past that. This game heavily relies on single-core IPC, which isn't the right way to distribute the load.
That "ji-tan" guy (comment deleted during me writing, lmfao! ) should maybe not call ppl's opinions "random bull" if he isn't even willing to look at the video screen (afterburner overlay), let alone examine the game on his own system...
Fuck, there was error and my multi paragraf answer is gone. Fuck it. Here is tldr if it.
Its mainly problem with DX11 as before DX12 cpu utilization of cores was bad in basically all games. Tgis makes no sense for Vulcan, that is, if they use updated version and use new api methods.
Pathfinding isn't supposed to be that complicated, we have had it in games since forever ago
Hopefully by the time most people get to Act 3 they've addressed these issues. Maybe its in their Patch 1 that they've been talking about? 🤔Either way, hopefully they at least improve the Vulkan implementation.
Code optimization is often one of the last things that people do. "Get if working first". So there is hope, but honestly from a developer's perspective, this is the not-so-fun thankless work, so keep that in mind when calculating when it might be addressed. Frame lag may be annoying, but as long as it is not "unplayable" it won't be a priority.
@@sofaking4779Yeah it's not as if the game needs twitch responses. Jittery framerates are not the highest priority
@@timk6181I am pretty sure it is, especially if you can't have a stable 30 fps on a high end system in the end that don't take advantage of your cpu performance more often.
vulkan crash for me
@@timk6181doesn't matter even turn based games with poor performance are still an issue old game but the last remnant by example
I noticed serious cpu thread limitation when just playing in act 2 so I was hoping there would be a explanation and follow up video. unfortunately there's no amount of optimization that fixes this without developer pushing updates.
Yeah, being CPU limited is hell
you can't ootimize that really, we just need faster cpus
I just bought a ryzen 9 7950x3d because of games not being optimized. Hopefully this fixes things
@@markojovanovic9651CPUs are fast enough, what we need is better and less lazy programmers.
Yes, the hell starts in Moonrise towers with a massive battle at entrance.
Thanks for testing this, expecting late game coverage on release was unreasonable! I wrote about this in a not so nice manner, unfortunately
My biggest gripe with the whole CPU situation, both here and in some other games, is that when the game is CLEARLY CPU-limited - often only ONE thread is being hammered, with the rest of the CPU just chilling.
Surely it's possible to distribute the load more evenly so that we aren't gimped by single-threaded performance every time???
There are operations that CANNOT be distributed among many cores. This is particular present in games where a lot of workloads require the output of a previous step to continue meaning you can't run them EVER in parallel. This shows up heavily in games and why intel trashed amd for so long with their better per core performance.
Multithreading video games definitely is doable (lots of games do it) it is not always straight forward, unfortunately. Its hard to program the game logic so that it can be broken up amongst multiple threads. And even when you do, communication between threads can be slow, so its easy to do it wrong and have the threads spend more time trying to sync with eachother than actually doing useful work.
I've got to admit, it surprises me that threading is still a major issue with games considering how long multicore cpu's have been out, and with how quickly 8 cores could become mainstream, developers really need to thread their games better over more cores.
A good start would be to abandon the older api's, I mean seriously, they are not good at multi threading, Vulkan and DX12 can do it far better with a good implementation over lots of cores, it also reduces cpu usage a lot and with a solid job at using the newer api's, it can offer a lot of performance gains on both cpu and gpu.
I find it remarkable in this day and age, why do they bother to offer a Vulkan, DX12 option if they are not going to take advantage of them? Clearly in this case, the Vulkan version is just a wrapper of the DX11 version, hence the worse performance.
Publishers and developers really need to start learning from companies like I.D. software which puts far more focus on Vulkan and the performance really shows what the newer api's can offer in the right hands, I mean seriously, Vulkan and DX12 are not new any more, they've been out for years and games should be transitioning to them exclusively without the need for DX11 or OpenGL versions, the ones that are doing that seem to be getting far better results out of the newer api's.
@@TheCrewExpendable it is doable, but depends on what you are codding, depending on the stufff you are doing you can't run at same time, its just how it works, some stuff needs to wait the task before it to finish before you can code the next step and theres nothing that can be done, now i do belive theres things that could be made use more cores and are not being used for sure i do belive this can be fixed yes, but what im saying is that in all games alot of things will have to run in a single core but this don't mean it must be like this.
Order of operations says other wise. And you likely might open up even worse bottlenecks. And that's ignoring the significant increase in bugs that end up being far worse than micro stutter a whole 10,000 people can even see.
Games that distribute things across those threads are much simpler than anything here.
Fun fact, when I tried the game on my old 4core/4thread cpu main character moving animations frequently stoped.
Meaning that sometimes the main characters moved around like pawns
almost sounds like a feature
You were just playing chess xD
SO... a really low-end PC could play the game like a table-top D&D game! Improvement?
@@photonboy999 it's kinda cool tbh
if you watch vinesauce vinny stream, its also clear his characters are freezing and skipping animations ALL the time. He does not have a dedicated streaming rig so its HAMMERING his CPU.
Hello DF! Thank you for going back to show us the full scope of the game! However, I think that the peroformance analysis is missing one giant part of the game - namely the local spli screen CO-OP. This is an absolutely integral part of the game and I am sure that many people were waiting for this feature especially, so they can have this adventure, at home, with a friend or a partner. The performance penalty that I observe when playing in local split-screen CO-OP is ~100% ( I assume it is CPU limited , due to running all game logic twice) in Act 1. I think this is a major point and it is the reason why the console releases are being delayed. I think it is an integral part of the complete review and it is an interesting topic to explore. Thank you!
This is probably the main reason for the delay of the console releases
Not on PS5. Just on XBOX. The PC version was actually brought forward (probably to avoid competition with starfield) while the PS5 version is releasing on the original date. The PS5 version reportedly had no issues with the split-screen while they were unable to bring it to the Xbox series S version. It may still be added post release, though.@@CaptainMisery86
CPU limited stuff is this gen's Achilles' heel apparently.
As a wizard running on a Ryzen 5 3600, currently in Act 2 and already noticing a steady frame decrease, this video scares me. But I appreciate the heads up.
If it helps, I'm on a 2600x and while there are noticeable performance issues at times its still very playable for me. Hope your experience is the same!
Im on R5 3600 too and Act 3 is not great to run but you can get use to it
I run a 3800X so I'm slightly better off but 30 FPS in this one specific town won't really bother me. Its a turn based RPG. better performance is always nice but not mandatory.
Yep, if you noticed worse performance close to the tawern in 2nd act, that's just a small sample of much bigger problem in act 3.
I hate to be that guy, but you actually never entered the city. All the footage you've shown is of Rivington, a small town that's like one day's journey away from Baldur's Gate, you can see the city itself if you look westwards from the bridge, just to give a scale of how far away it is.
Rivington area is tiny compared to the city, so you completely missed the Act 3 perfomance issues.
I used the rivington road as a Benchmark area since it is easily saved from, traversed and shows the issues
@@DigitalFoundry fair enough. In my experience, the city was far worse than Rivington, but I'll take your word for it.
@@DigitalFoundry Rivington is not even the worst case scenario in act 3, you really need to travel to the lower city, go up on the roof of elfsong tavern etc to see how bad it is. Paladin auras seem to cause more performance drops and entering lower city causes some people to have severe input delay where doing anything takes seconds like talking or jumping, loading takes over a minute and so on. It's really bad for some.
> but you actually never entered the city.
@8:30 he is running around in front of Lower City Waypoint.
@@abhishekvalsangkar From my experience, the Rivington bridge is already as bad as the inner city.
I think that they should have added a slider to control amount of npc's around, cause 99% of the npcs in the town are useless.
I expect a call mod to be in the works
And they are all clothed the same 😂
"omg this world is so empty and dead"
This would definitely be useful - I assume reducing draw distance still has to perform the pathfinding for non-visible NPCs but it could at least reduce number of draw calls so there is a bit more headroom for pathfinding logic, unless pathfinding is (hopefully) running separately on its own cores.
The npc's are not the problem tho.. Witcher 3 had big cities aswell and performance was fine.
My CPU: 🥶
My GPU: 🥵
It would be interesting to standardize the provision of more late-game access to technical outlets like DF.
Great work as always, Alex. But I'd like to see more Zen 3 and 4 testing in your videos, especially the X3D variants.
AMD's 3D vcache parts are murdering this game, even in these sections.
Murdering meaning? 5800x3d still stutters.
I'll definitely test it out on a 7800X 3D and a few GPU's I got access to but am not there yet. Not even close lol. I love to take my time, especially with this game.
correct@@LaisvasisRadikalas
I am getting 15 to 35% higher FPS here in this area,with HT off, then Alex,but I am on a 7800X3D/2x32Gb/OPTANE 905P 960gb/ DDR5 6400 and 4090.@@LaisvasisRadikalas
When I got to baldurs gate I was surprised by the large number of npc and I asked myself would there be any performance issues? , to my surprise I didn't notice any but I'm not one to notice tearing or stutters either despite having played countless UE4 games but after this video I can definitely see it, I guess when a game is good I get easily distracted 😅😅 and I imagine that the perspective of the camera is also a factor
I'd like to note that the "pathfinding" seems to account for a lot of frame time loss and frame rate loss. In act 3 I tried to jump through a small "window" but my character bugged during the jump and I ended up in the wall. The area I was in was an instanced area and I was receiving roughly 100 - 130 fps with only 4 enemy entities that were not attacking (disabled), my two player friends, and 1 npc ally - (Lorroakans tower). Once I landed inside of the wall, and each time it was my turn in the "fight" my frame rate would drop to a crawl of around 11 - 16 fps. Immediately when ending my turn and moving to a character that was not 'stuck', the frame rate would return to 100+, this was easily repeated. My inclination is that the game is CONSTANTLY refreshing what pathing is available to the controlled character and when more clutter and ai actors start getting in the way of that, the pathfinding code is updating to account for it and that adds strain to the workload. (Hardware: AMD 5900x / Strix 3080Ti / 32gb DDR4 3000 / EVO 970 nvme - Vulkan API / 2k Screen Res)
I would love a revisit to act 3 with the new update released yesterday. It would appear they made changes to the issues you pointed out.
it's still trash unfort.... I've been trying to get through act 3 but going from 120fps in act 1/2 to dips as low as 30 is miserable.
Can't wait to see how you use the PresentMon Betas gpu_wait values to get you more insight. I can't download it today but I want to try it with a GPU to see if I can better understand how the smaller PCIe pipe affects texture and mesh loading times and the effect on an idle GPU in these cases.
I was curious about wether they would use that tool or not? Did DF ever mention it?
Smaller pcie lane count available should not have an effect unless the bandwidth requirements are exceeded. If they are, the frame time spikes you see are simply due to the GPU cores waiting on data. And sometimes the game will rather load "worse" textures or let them be queued (pop in) than let the framerate drop. And that results, off course, in visual artifacts/poor visuals.
@@fVNzO I can definitely say the 40g tb bandwidth has a huge impact on some games. For example if I play a game the loads progressively some games drop to 3 fps until everything is loaded, other games just load earlier. Looking at GPU busy I can actually see when it happens, so for me it's been great insight
@@Hossimo I think it's a really cool metric and it shouldn't really be a limiting factor for a GPU. I think that if there is a PCIE bottleneck you can consider it poor design. It has only really been an issue lately on some low end cards because the chips are essentially made for laptop first where form factor and power limits are more important. And the chips will be running slower anyway so the 4x/8x restriction wont be as noticeable.
As far as GPU busy goes i think it's a very cool metric.
I absolutely knew it was the NPCs. In Act 1, on my crap GTX 1080, it was almost butter smooth. But in Act 2 and Act 3 as the numbers of NPCs increased, my frames dropped like 20% like Digital Foundry said. HOWEVER, I am not complaining because Baldurs Gate 3 has some of the most fantastic visuals and NPC path finding that I’ve seen. Characters don’t just magically float or clip through objects. If you Minecrafted a bunch of boxes, NpCs will find new pathing around those objects. It’s an evolution of the same boring AI pathing that games like Assassins Creed or even The Witcher 3 implemented. Hopefully, next patch will clean up some of these cpu issues.
I’m really looking forward to the first big patch for this game. It runs pretty well for me, but there are definitely bugs and performance issues. Great game though! I’m loving my time with it.
It's already in my top 10. Might even get into top 5, I'm only in second act too :D
Love DF content!
FYI, it's data streaming in when you're walking in a circle. It tries to predict the way you're going and loading everything up.
@4:48 - this is exactly what I have been trying to highlight for game performance analysis in the past - "the per frame frametime differential" can be an important indicator of poor performance. However, one of the things I realised after the fact was they it cannot capture sequences where the frametime differential is small but results in improved or worse performance over time if the per-frame difference is small during this transition. It's good at capturing stutters and large changes in performance but not in capturing 'leanings' in frame delivery over time.
@4:49 specifically, later on you call out Vulkan as performing worse but in this test, the dark green graph shows that DX 11 is worse?
I would also argue about your choice of comparison at 5:38 where you say Vulkan is 10% worse. 10% worse in what, exactly? The frametime dips and lengths are very comparable between the two APIs and I doubt the majority of users would even notice a difference between the two!
Without the raw data it is difficult for me to provide accurate feedback but I think your averaging system might be too broad here? It could also be because your default display is counter-intuitive to normal expectation (i.e. 0 at the bottom of the graph) and you do not zoom in during periods where the frametime differential is small.
@7:47 I would love to see a frame rate locked analysis - showing us a 60 fps limit performance - does this improve the situation?
My biggest issue with the third act is I was hitting VRAM limits on my 3080, so my game would frequently slow to like 15 FPS while in the lower city. I lowered my textures to high instead of ultra and that fixed it.
Really enjoying the game. Can confirm. Thank you for the video
SO HAPPY TO SEE THIS!!!!!
Dang, I'm rocking an R5-3600 w/ 3060Ti in my desk machine and am really bummed to hear how much worse the performance gets. Act 1 has been buttery smooth so far, but this is quite a concerning difference. Here's to hoping Larian gets a performance patch out before I reach Act 3 (which will be several weeks at this point 😅).
Jus bad optimization enjoy your PC
@@subjectivereviews yup, so much for "omg this game is prefect!11!" lmao as if
Messed up thing is you're at/above the official recommended spec, yet still below what you'll need to even sustain a stable 30fps. imo "recommended" specs listed on Steam etc. should always be enough to maintain 10% lows above 60fps at 1080p as a standard in 2023. Though it's obvious why Larian haven't done that here, because the percentage of people with a CPU capable of that in this game is tiny.
I guess a lowish framerate cap is going to be needed to have a good experience. Thanks for the update!
I just set mine to 30 fps and felt so much better as it feels more stable without constant frame dips.
Yup, setting to 30fps is the way to go
As a hardcore MMORPG player, I really wish you had tested with even the 5800X3D. The extremely common occurrence of NPC's or large player counts dropping performance due to cache misses is alleviated by a great degree 99% of the time. Its the main reason I upgraded my rig from my 5900x to the 5800X3D when it released.
The extra cache of these chips in open world games with lots of AI helps a ton by turning those cache misses into hits. Meaning less times your system has to fetch from system RAM. If you had time, would love to see it tested using those chips as too many people think CPU bottlenecks are only up to total usage of the CPU or clock speeds.
The reason you may be seeing more frametime spikes with more summons may be because of puddles.Besides losing/gaining auras causing frametime spikes I noticed even walking on puddles of water (which act 3 is full off) and puddles of blood after killing a group of enemies becomes a hitch fest as you walk your party over the blood.I'm assuming it's your party losing and gaining the "wet" condition.
Some other user in the steam forums theorized gale and shadowhearth possibly being the cause of the puddle micro stutter.
Today's patch massively improved performance in act 3. On my 5900X it more than doubled. And looking at the graphs, they did it by having much better multi-threaded load now.
Great work as always. I can't wait for the performance review of the PS5 version.
I’m worried about the performance, since PS5 is what I’m gonna be playing this on…
@@304enjoyer3same here, this could be a big issue on console… maybe this is why it got delayed?
@@304enjoyer3 I think I saw a comment where they said they're aiming for 60 fps on PS5, so I wonder what types of compromises they'll have to make to achieve that. Based on what we've seen, it seems like a stretch.
The PresentMon stuff that was covered by GN today, seems like it could help get to the bottom of what specifically the issues are that cause the frametime issues.
Yea thats a interesting app from Intel.
i quite the game at act 3 because of this. truly sad moment as I was really enjoying this game.
This game simply doesn't have the visual fidelity to be so taxing, the devs need to do better at optimizations.
I'm confused...wasnt Baldurs Gate 3 getting praise for being such a leightweight for its presentation?? Like not re-inventing the wheel but running nicely even on old hardware and having this "HDD loading mode" ?? Maybe I'm tripping
extremely helpful thank you!
As a player with only i3 8100 and GTX 1050ti 6gb VRAM, what i did was to limit fps to 30 so it can at least be playable for me. Planning on upgrading to i5 9400 f.
6gb vram wtf!?
Im on i3 6100 with 1050ti 4gb, can i run it? Cos im in though for buying the game but its pricey for me and dont know how it will run for me
Thank you for listening the community
Thanks for this.
It's funny non of the reviews actually mentioned these performance issues. I guess the length and front ended nature of this game played great favor when it comes to the first impression and the review scores.
elden ring on release ran like shit aswell on alot of pcs , reviewers usually talk about core game and only in some cases of absolute unplayability like cyberpunk its going to be mentioneed.
The reviews were based on completed runs. They saw the performance issues. Something people just like to forget is that performance is not as important an aspect in a games quality to the majority. Most people can never run things at max to begin with. If the performance is notably bad. Then you will see it affect reviews. But it’s not. The game just struggles to hold higher frame rates in the act 3 city zone. That’s worth a single sentence worth of mentioning in a review and I would question the review’s integrity if it actually affected the score by more then half a point.
@@lynxyu11which is fair imo. Since 90% of players aren't going to notice the issues or care about them. You need to understand that digital foundry mostly has an audience of enthusiasts, which view these topics differently than your average gamer that just watches a review to get an overview of the game and see whether it is suited for him.
This isnt really a game that NEEDS High performance so
@@lynxyu11Because CP2077's reviews were exclusive to the PC port.
Most early reviewers had a rig that could brute force the game's bad performance and in turn get less bugs because a lot of them were caused by low framerates.
Technical state of the game isn't very good, but it absolutely can be improved
well taking early access into account, i can understand the performance issues with act 3. It just sounds like an optimization issue, one that also sounds like it can be rectified. Hopefully they address this. Also, running an i7 9700k and a 3070, and 32 gb of ddr4 ram, my vulkan performance is 10-15% better than dx11, but ive only played the first 2 acts. not sure if im tripping ( i probably am if everyone else is having worse performance on vulkan) or maybe theres some weird bug with the api's. also triple buffering does not work as intended on vulkan for me.
Huh now I’m really wondering how well my rig will hold up in act 3 playing split screen… been enjoying the game a lot, and only ran into a handful of minor issues so far, but performance is definitely worse with 2 players, I can only imagine what it will be like in act 3 😢
Thanks Alex ! You are truly a gem in this sector. In the initial analysis, I was one of those who commented that ACT 3 is really rough in terms of bugs and performance issues. Thank you for taking comments seriously and looking the issues.
I was extremely happy with my Acts 1 and 2 frametimes but yes, upon reaching Act 3 performance has worsened sub-30 fps with a Ryzen 3600 and an RTX 2060
Yeah that was to be expected tho the third act is the one that probably have less than a year of work in the technical part i remember divinity original sin 2 had terrible performance om the final act when it launched and it was fix some time later
@@allansolano5587 Hopefully they do improve it then. I have some faith that they will because my first experience with Divinity Original Sin 2 was the PS4 Definitive Edition,, and that was actually a pretty solid port from start to finish imo. I never encountered any bugs or performance issues with that.
Difference is the PS4 version of DOS2 launched 13 months after the PC version, so they had a lot of time to fix things. BG3 is releasing on PS5 1 month later... so that might be a problem. In a way, the Xbox Series X players might end up being better off with the longer wait lol.
Expected? I'll tell you what's expected. It's expected for the game to work properly. I mean, wtf as become of pc players in the last decade? ALMOST EVERY SINGLE AAA GAME WE PLAY ARE BEING RELEASED ALMOST UNPLAYABLY FOR LIKE 40% OF THE CONSUMERS!
It is more about the amouth of things that are in the town. The game engine is same. No matter what act you are in. Act 1 and act2 are just less demanding to CPU…
@@yewtewbstew547 yeah basically you got the best version of divinity because it actually had more features and stuff in it that the original version on pc
Focusing on E cores in modern processors instead increasing Performance Cores was a huge mistake.
E cores are a stupid fad to try and bolster core counts for normies.
Thinking we need more than 6 cores is a fad. At least E cores is a better way to handle that fad.@@thefilmdirector1
The C&C music in the background brought back some memories... :-)
In case anyone watches this video now - check out the new patches. Performance has changed a lot for the better since the new patches. It still needs more optimization, but it's improved a lot, quickly.
I knew there were some issues with frametime due to CPU, even on my 13700K on Act 1. Game seems like it's stuttering a lot at around 90fps with Gsync.
Hi Alex,
I was wondering, if you could also test with an 12 or 16 core Ryzen for scalability.
Because to me it looks like more, that most of the games only use the P Cores for heavy duty work and are simply ignoring the others.
So to see if the games could scale to 10 ore more cores, if they are all "P" Cores, it would be very helpful if you would test this with a Processor, that has more then 8 P Cores.
You gain performance up to 10 threads, even if just a little, 6 threads gives you 95%.
I would be interested to see if the X3D cpus make a difference with these dips. My 5800x3D has not had nearly these types of issues in this act. I'm curious to see if the 3D Vcache is making a difference with this issue and runs better than a standard CPU. How does the 5800x3D handle this versus a 12900 or 12700? How does the 7800x3D handle this versus the 13900? These are interesting comparisons that might make a difference.
My bet is on pathfinding, there's a good amount of times your companions can't follow if left behind for a while, the same reason they disabled the move orders on the map, to prevent long distance pathfinding.
I have yet to reach Act 3, so my question to you: Does the settings "dynamic crowd" On/Off have any impact on performance for your stated city worst case?
How you guys manage to create such indepth videos in such a short length of time is beyond me. Very cool to see this content
Experience and dedication
Competence and effort, which I can't say the same for the programmers of this game.
Sadly act 3 is terrible. Really drags down the game tbh. I recommend waiting a year or so until the definitive edition or at least until patch 2 yo give it a shot.
so it is AAA garbage afterall. They just made the first 20 minutes seem good
@@Wylie288 It's actually way worst than that, because, fun fact: I can't find a game that performs worst than this. Not Redfall, Jedi Survivor, The Last Of Us Part 1, or any bad port with CPU performance issues.
I have a 4080 with 32 GB 3200 Mhz RAM, and an old i9 9900k. My NVMe is the second fastest in the market, and my motherboard can only use like half it's speed, but I'm more than fine with that to. The i9 performs about the same as a Ryzen 5600x, except in games that take advantadge of more cores like Cyberpunk. In that case it's 20% faster, so it's a little bit better prepared for future games. From my experience, all I can say, from my experienced, is that I never played a game under 60 fps since I got this CPU, and I play A LOT of games. I actually was about to finish Act 3 before Digital Foundry upload their first Baldur's Gate 3 video, meaning, the previous one.
Now, if you do a little bit of research, you'll notice a funny thing: there is a video of someone showing optimized settings for this game. The video is easy to find, and he talks about the CPU performance issues. That youtuber has a Ryzen 5600x, and this game runs with the framerate on the 40s with that CPU on the city. There isn't a single videogame that runs that poorly with the 5600x. Not a single one, so this could truly be the worst optimized game I've ever seen, which is SO FUNNY, because I've seen a lot of people using this game as an example of a polished product.
When you see that a game barely runs well on an i9 12900k with a 4090, you have to at least consider the idea that this is very poorly optimized.
The first 50 hours are great, so, obviously, most youtubers gave fast conclusions before actually finishing the game. They'll just not admit that after all the drama, because it would look bad. Reviewers to, and you can tell that. I was actually surprised that IGN, from all the RUclips channels, actually had a good review video, but the problem is they gave this game a 10/10, because if you hear the video, it sounds like they did a great job, and they know about all the issues the game has in terms of the ending, bugs and performance, so how can that be a 10/10?
@@joseijoseiBut all of the people with buyer's remorse insist that an unstable 24fps is enough for a turn-based game like this. You are not entitled to functional video games or a stable 60fps. You're not entitled to 30fps either, even on the highest end PC. The human eye can't tell the difference above 10fps anyway. Poor devs shouldn't have to work extra hard to tune the performance above that threshold. Just buy the latest CPU to handle their spaghetti code!
It could also be a polling issue with the input of the mouse and keyboard. That it doesn't actually poll but use the raw data and causes a lot of vector/motion input changes.
Sir I think you just explained my issue with Hogwarts Legacy using a dualsense controller. Whether I'm plugged in directly to the USB or just using the connection from bluetooth, my GPU usage always dropped and my frame rate went down by around 5 or more every single time. I couldn't find the answer to this on google.
@@Stardomplay It's been known that high polling rate perhipherals can take up massive amounts of cpu cycles if used raw. So yeah, especially if that thing has wonky drivers it can definitely have that effect.
Is there a quick timestamp we can skip to if we don't want any location spoilers etc? I'd just like to see a chart so I know what I'm in for!
Also how much would RAM speeds affect these frames? The CPU isn't alone in its struggles to process all that info after all!
What about performance degradation over time in act 3? A lot of people, including myself have to completely restart the game every 10 to 30 minutes because performance gets consistently worse till it's almost unplayable. This seems like a much bigger issue than the what was mentioned.
Memory Leak? I did long sessions in act 3 and never experienced a performance degradation.
@@riccardopiccinini666 that's what I thought at first or maybe my vram usage was spilling over but I monitored both ram and vram and neither of them ever got maxed out. So I'm not sure what's going on.
Happens in Act 1 instead expect it takes like 2 hours or so for it to happen.
Are you implying most reviewers (and not many reviews out there) didnt complete this massive game before posting their review…?
Not that shocked to be honest…
@@bingobongo1615 More likely it's just not an issue that affects everyone. Not everything is a grand conspiracy.
For me this is a game that could run at 30 for all I care so long as it's stable since it's turn based. I'll take the FPS hit for prettier visuals in a game like this.
I really appreciate you making this video … it has saved me from throwing away money on a game that I can now see would perform horribly on my old i7 6700k. (The recommended game spec seemed to suggest I’d be fine, but it’s now becoming clear the published specs are not in line with reality.)
I played through act 3 with an average fps of like 20. Act 1 I (after some hoop jumping) managed to get to a stable 60. Quite the difference. Act 3 is the longest act though (for me at least) so I almost forgot what smooth gameplay looked like after 50h of stuttering.
Really interested to see the sacrifices needed to get this to 60fps on consoles.
They could literally remove 1/3 or half of the NPCs, and the city would still feel alive, not sure why they did that if it performs bad.
Not like the game needs 60 FPS. If they can get it to run at smooth 30 FPS that's easily enough for this type of game.
@@Naryoril I think Larian made a forum post regarding the one month delay for PS5 and it was due to trying to get the game to 60fps. I'm pretty sure that's the target, I just don't know how they're going to do it.
@@MrAlienwthey did it because it's their creative decision. I'd much rather want them to be ambitious with their game designs instead of catering to these consoles and in turn hampering the PC version.
@@retroreplay1980shouldn’t be too hard. Ps5 is a actually little monster (little in price). Also consoles use dynamic resolution and likely they’ll reduce the npc density like they always do. It’ll probably run great
Game seems to favor high cache CPUs. The AMD x3D chips are churning through it
I'd love to see Alex test high cache CPUs more often
Interesting. I haven't seen anything on that yet. They have any comments like the FPS they are getting with them?
Have not seen any evidence that they are any different than intel CPUs or AMD non-3d cache visavis the scaling in this act at all, would you mind posting back some areas where you saw reporting on that?
The user reports under my last video that caught my eye most that you can even see scrolling in the video here mention users with 5800x3d seeing big CPU limitations in this area of the game.
I get the odd stutter with my AMD 7950 X3D and 4090. Usually when I'm picking up things or entering inventory.
@@Relex_92 When you use a memory bandwitch limited cpu with a jedec spec ram kit for sure it performs like shit and X3D do better
I would like to see some testing with the X3D parts. I have a 5800X3D and RTX 3070 - I am running 4k with balanced DLSS max settings - i downloaded an ACT 3 save and ran around all over Baldur's Gate and had a much smoother experience than this video showed. There's occasional stutters but not constant like this test showed...
could you maybe do some kind of hardware GPU scheduling revisit if it has been improved?
I really *REALLY* want to see a performance test for Acts 2 & 3 using more moderate hardware overall.
Act 1 is easily the best optimized part of the game, but there were still a lot of time my performance took a hit.
However, 50+ hours in it's happening way more often. I still think the game deserves all the praise in the world, but it's kind of silly seeing the preemptive trashing of Starfield bugs when Larian just released a hot fix apparently addressing 1,000 bugs 😅
Completely valid opinion but Starfield budget and team are also a lot larger and is being set up as THE game for Xbox, whilst BG3 was a more niche title so these do play a part in community opinion and behaviour/response.
@@aadam4035 That's not true. Larian employs more people than Bethesda...
Also, I don't even have an Xbox. Starfield is a PC game first & foremost to me, and it will easily sell more on PC as well. Console fanboys can unfairly whine about nonsense like that, but it only serves to prove how willfully dumb & hypocritical they are.
@@aadam4035And please don't try to tell me you know the budgets for the games... BG3 is 100% a AAA game through & through.
It's a 10/10, massive game, in development for nearly as long as Starfield. Just because Larian is independent, doesn't mean they're small.
@@aadam4035 If I were in your position, I'd reload any save prior to posting a reply on this OP. It's like exploring a dungeon/basement and running into traps, now you're covered in acid.
@@donloder1 lol alright ill take the L
5800X3D has been the best investment I’ve made for my PC. Bought and quickly sold the standard 5800x, wasn’t worth the upgrade from 3700x
Does the cache help this area, I have a 5800x3d
I just changed 7700x to 7800x3d. For most games at 4K and 1440x5120, you don’t see a jump in fps but you definitely can feel the smoothness with 1% low and such. I’m curious as well if 3d cache will help with this game.
@@anthonyrizzo9043It doesn't at all. I have a 5800x3D and I'm 150 hours in and it stutters all the time in the city. Honestly even worse than the video shows.I am capped to 60 cause I'm running at 4k and while the framerate never drops, it hitches every 5 seconds.
@@rage8010Mine did the same on act 2 with vulkan, a stuttering mess. Then tried dx11 and it went away somehow.... And dx11 did not work well on act 1
I'm hoping to see a Mac video as well whenever that gets final release. Would be interesting to see the comparisons vs the PC version.
Nice follow up, people asked about this a lot
Yeah my 3600 literally goes bananas in this cities it feels very janky on those 30 fps moments hahaha
Hopefully those issue will be addressed on the ps5 version or it won't matter in consoles if there's a proper 30fps cap
They say game will be 60fps on Ps5 but looking at our recent times i seriously doubt that unless they downgrade the graphics heavily
@miguelsanto8123 it's not that heavy of a game graphically. I believe DOS:2 was 60 fps too unless I'm mistaken. This doesn't look that different than DOS2.
@@miguelsanto8123 graphic settings has nothing to do with CPU bottleneck in ACT3. Developers said that they are close to caping the game at 60 fps, so I believe that PS5 api is much better than PC counterpart.
@@miguelsanto8123graphics are not the issue with bg3. There is more behind a game like bg3 than how it looks.
3:37 I saw this in my Steam Deck review. The player emits a vision cone similar to enemy stealth checks, and the game has to figure out who you might see, and if they see you.
I have a GTX 960 and AMD FX 6120 and couldn't resist taking a chance with BG3. The game is mostly functional on medium settings for me, usually about 25-40 fps. But have had dips down as low as 4 in some fights. Still in Act 1, looks like I might need to start saving up for an upgrade or 2.
With my 7800x3d in act3 around the wizard tower with 50+npcs i have around 90fps +/- 20 when rotating the camera (cpu limited). Usually u have 120fps cap on, on ultra settings. Works totaly fine for me but its a killer cpu.
Thanks to let us all know that you have a killer cpu. Congratulation
@@lastonestanding1641by any chance you want to know about my killer GPU too? :^)
@@lastonestanding1641 np ;)
@@n.erdbeer A 4090TiSuperXTX?
@@EarthIsFlat456 nothing less is good enough to run this game, anyone should buy one :^)
act 3 is the worst in everything it has a lot of bug too, from camera glitch in cutscene to unresponsive game when interacting with npc and also straight up quest not working or working for half the content. I hope they patch it soon enough because is near game breaking in my opinion.
Thanks for the follow up video!
I've been playing the game a bit ON and OFF between the Steam Deck and my main PC, but Act 3 was a PC only affair. The Steam Deck just cannot cope with it in any meaningful way.
Also, I would start including in the CPU comparisons test a high cache CPU, like the 7800x3D, because in some istances the bottleneck can be found in the amount of cache and it would be nice to get a comparison with one of those CPU also, but I get that it could be much more work for not a meaningful difference with the 12900k.
5800x3d reporting in, 100mb of cache hasn't spared me from the stuttering.
Thanks mate, appreciate the effort this would've taken and ignore the negative nancies.
At 12:00 the minimap shows how the game struggles with all the npc update calls. That should normally be completly smooth.
tl;dw : game's unoptimized as fuck and act 3 is a stutterfest even with the best rigs, lets hope their "huge" patch fixes this
should have focused more on optimization instead of sex scenes
The usage of the CPU cores is very odd, half of them are hovering around 0% with the other maxed out. Maybe this needs a DX12 implementation?
it would be identical to vulkan. the issue has nothing to do with the graphics API.
I am glad I upgraded to a 5700x from a 2600x, its been perfect for playing BG3 and should be good for Starfield next month. Don't have much to spend on upgrades atm so this was a nice cheap one I could do with the CPU :)
I have a 5700x and the fps in Act 3 is horrible (40-60fps), Act 1 and 2 is pretty consistent above 100fps.
We definitely need an Act 3 breakdown.
This is an incredibly useful video, and its great to see how the later game scales, but it is making me wish you guys would add a 3D chip to your typical bench.
A 5800x3D would be a good "midrange" option between the 3600 and the 12900k, and for this game in particular, I was kinda hoping you guys would test the claims I've seen that the 3D chips have vast performance advantages in this game.
I run a 5800x3d, and a 4080, play in 4k max settings. It's cpu bound for me. 144 in cutscenes, drops to 80 when moving around populated zones. GPU useage under 99%>.
@@sgtBONGwater still mighty impressive. I'll upgrade from an 8700k to a 3d cache chip next year, can't wait
I've said for some time DF needs to overhaul their "AMD" testing. Or lack there of. Considering the prices either the 7600X@$230 or 5800X-3D@$330 paired with Richards 6800XT would be great and very doable as a business write off. They have a High-End Intel/Nvida System so that was in the budget. The whole AMD is for Budget Builds only isn't valid. Considering you can for an all Intel CPU/GPU build, but they don't want to put that perception out there, and where I've had my issues with DF over the years....questionable moves they make that only certain age groups and people who work in the tech industry can see.
@sgtBONGwater Yeah, I'd be amazed if the 5800x3D, or even the 7800x3D or the 7950x3D could completely eliminate CPU bottlenecks, but that still sounds like a much better performance than what they got with the 12900k. There aren't that many games where the 5800x3D wins that much of an advantage against the 12900k.
They're also still using a 12900k as their high end CPU, rather than a 13900k. Which, yes, the 12900k is still a high end CPU, but it isn't "the best possible CPU for gaming, if money isn't a factor" that technical benchmarkers prefer to use. There's more factors here than "they don't want to", I suspect. @@chrisbullock6477
Thank you! We needed this video. The game was phenomenal, but we can't just look away from the poor performance in Act 3. Hopefully Larian will be able to smooth this out a bit with some hot fixes and patches.
Can we get a test of the third act on steam deck? Is larian aware of third act drops?
My only remaining question is one of 3D Vcache. Do we know if the extra cache helps at all in these CPU-limited, NPC-dense scenarios?
I heard that the Act 3 (ie in BG itself) noticeable performance drops and stutterings are due to save game files. Seems like a bug.
It stutters as soon as I’m out of the ship, on a 5800x3D, 3090, 32GB, M.2 SSD, not looking forward to getting to the city then.
@@WhiteHawk77 That shouldnt be like that. Whats your resolution and are you using DLSS? 5600x, 3070ti, 32gb 1440p no stuttering.
shouldnt need dlss to get no stuttering lmfao
definitely stuttery and longer loads.... and worse some progress bricking bugs.
@@dave8413 4K, doesn’t matter if DLSS is on or not, my frame rate is fine, 70 to 90+ most of the time but moving around it’ll be fine for a few moments, then some small stutters, then fine again and repeats. I know some people can’t pick up on stutters and bad frame times but I can.
Vulkan is better for frame time consistency, or 'frame health' like you said. So even if DX11 might get me extra frames, I would much rather use Vulkan so the game is smoother. It is much more noticeable when you pan the camera using middle mouse button.
not my case, my fps can stay steady with vulkan with my 6600xt
I haven't even gotten out of Act 1, but my game produces crazy frame-time spikes whenever new UI elements (like tooltips in the inventory or skill descriptions are drawn) I drop from a framerate above 80s to something in the 30s. Just scrolling my mouse over my character's inventory dumps performance. i haven't found a solution to it, both vulkan and dx11 have issues on both my desktop (Ryzen 3700x/RTX2080/32GB 3600MHz ram) and my laptop (a Lenovo Legion 7 Slim w/ a Ryzen 5800H/RTX3060 and 16GB of ram) latest nvidia drivers and all.
Is it possible that the way the charater is facing and moving starts some preloading of where you will be heading. So when you run in a circle, the game is just constantly dumping and preloading content?