This is why GPU Drivers Might Not Matter
Vložit
- čas přidán 25. 06. 2024
- We used to assume that GPU drivers allowed for much better performance after a game launch but that might not be the case. In this deep dive into GPU benchmarking we discuss GPU drivers vs game updates and how some changes might completely break performance while others will benefit FPS in a big way.
Check out the Thermaltake Tower 300 (Sponsor) - geni.us/Tower300
Timestamps:
0:00 - A New Way of Benchmarking led to THIS...
0:46 - The Problems & A Year Long Project
1:49 - Alan Wake 2 Updates vs Performance
3:05 - Rushed Launches = Issues
3:46 - Sponsor Spot
4:16 - Good Examples of Update Stability
5:02 - Starfield Updates vs Performance
5:29 - The Baldur's Gate 3 DISASTER
8:08 - What HAPPENED with BG3?
8:42 - Updates can Make or BREAK a Game
8:58 - The GPU Driver Narrative
9:41 - Old vs New Drivers (Alan Wake 2, Avatar & Robocop)
11:14 - The MYTH of Starfield Driver Improvements
12:34 - Baldur's Gate 3 Proves a Point
13:51 - Newer Drivers don't Matter Much...EXCEPT
15;28 - A MASSIVE Challenge for GPU Benchmarking
16:56 - Where do we go from Here?
GET MERCH ►hwcstore.com
SUBSCRIBE ►bit.ly/SubHWC
This video is sponsored by Thermaltake. As per Hardware Canucks guidelines, no review direction was received from manufacturer. As an Amazon Associate we earn from qualifying purchases.
Follow us for more updates!
Twitter bit.ly/HWC_Twitter
FaceBook bit.ly/HWC_Facebook
Instagram bit.ly/HWC_Instagram
--------------------------------------------------------------
#hardwarecanucks - Věda a technologie
5:11 "The effect on the A770 was relatively minimal." Might want to recheck your facts there, Mike. The A770 saw nearly a 20% performance uplift with that patch, while the AMD and NVidia GPUs saw a 7% and 11% performance increase, respectively.
Yeah, mea culpa.
Drivers are not the problem , this is a good video because its to the point , game developers no longer optimize games.Preorders have destroyed the gamimg comunity and semifinished games are a standard now , prealpha game testers were replaced to early access
It's not really the fault of preorders... I've been preordering games since the SNES. I think the bigger issue is the relative ease of releasing patches.
"Back in the day" before digital content delivery, you physically manufactured your game. You didn't have an opportunity to patch it post-release, so the game had to have a scope reasonable to finish and playtest it in time for it going gold and ready for manufacture.
Now you can have a game-breaking bug on release, and it's not really a big PR hit as long as you've got a Day 1, or maybe Week 1 patch.
@@brayoungful I just mentioned the same thing else were. Why I love physical media.
I can't believe I've never seen anyone control for game version in diver comparisons. When you think about it, this should be obvious, but I guess it can be easy to fall into norms during testing.
Good job on breaking the mold HC! Really great research!
05:04 Going from 28 to 34 fps is much more substantial than 88 to 97 fps
As soon as I heard him say that I had to stop and do the math. This is not good science.
Going from 1 to 2 fps is much more substantial than 40 to 70 fps, yet we all know which is far more meaningful.
nor really u wont feel 28vs 34 if it would be 40fps yeah but not 34. But maybe it will mean u will be able to set one fsr higher to achive 60. But that 10fps on 144hz is good change (not really too fast fps brakes so many games its crazy i played two old games i had got for free on epic (old plague tale innocence and SOTTR and both games was stuck in quest because i couldnt turn lanterns/doors in those games but cap at 120fps fixed both games. (nvidia please made DLSS that will scale 4k to lower RES like DSR that does work in game not this shit DSr that is so buggy its crazy)
Be careful saying that because it’s not entirely correct … people don’t notice a percentage change in fps … they notice a physical change in fps and the impact will be different at different base frame rates. For example, going from 1 fps to 5 fps is a massive 500% increase in fps however the improvement is not meaningful because the experience will still suck. Whereas going from 26fps to 30fps may feel much more significant even though it’s only a 15% increase. You can use math to tell very different stories, so I would be careful making statements like this. If you have a 60Hz monitor you would be correct but if you have a 120hz monitor I would disagree.
@@erisium6988 You'll feel an increase from 28 to 34 much more than you'll feel an increase from 84 to 90.
Another thing that drove me crazy for a few days that was killing my FPS in a few games (especially CPU bound games) was a Windows defender setting that was killing my FPS. It is under windows defender in Win 11 > device security > core Isolation > Memory Integrity. Once I turned off Memory Integrity my performance went back to normal with GPU utilization going up and CPU utilization going down.
Wait till you see an upcoming video about how MS bloat is killing performance.....
@@HardwareCanucks Debloating windows 10, going from 150 processes to 80, gave me around 10 to 30 fps depending on the game.
@@HardwareCanucks Looking forward to this, thank you!
This is not just for games, I work as an IT specialist and we've turned this off on our 2000+ computers in my organisation since it was just creating issues left right and center
That's a fine tip, however users need to understand what kind of degraded security protection this results in. Memory integrity does offer nice protections in case your computer is breached (attackers not being able to access some system-level stuff). But it is a bit resource intensive since it is essentially virtualization. And you might need to turn it off for other reasons anyways (like if you're running virtual machines).
Great work! Always wondered about this, thanks for this incredible testing job.
Interesting trailer fro next video. Will wait for next part!
No game needed this testing more than Jedi Survivor
Driver updates mostly matter, when you experience bugs related to them or instabilities. And sometimes newest driver version can actually "unwork" the device - in 2019, when i assembled my current rig, i had great difficulty to get my intel LAN working. What ultimately helped me, was not getting an update, but looking for older drivers for the same NIC - i ended up using X470 intel LAN driver on a X570 board and i use it to this very day.
Very interesting, thanks for covering this!
Great video … I was recently struggling with a related issue while benchmarking … this helps me realize I wasn’t going crazy 😉
On companies releasing broken / unfinnished games
Big patches and bug fixes for online games is hardly new..
Do you remember Half Life launch.
I remember having to download a 700MB patch to make it playable on 16Kbps connection (should have been 28kb dial up but our line was sucky)
That took me like 29 hrs IIRC
And thats nothing to say about the times it got interupted because someone called the house.
No resuming file downloads and no download manager / launcher at that point. Steam didn't even exist.
Great video! That's one thing I've mentioned a few times, but I couldn't find anyone doing the proof (and I can't do it alone), so it stayed as a gut feeling. Now, I'll use this example.
Larian needs to look into this! Have you contacted them?
We've been trying to contact them, even through our contacts at AMD & NVIDIA without success. We would have loved to have added an official statement to this video.
Great video Mike 👍😇!
It’s great to see gamin* results from this perspective…..I thought my BG:3 was buggy or something as my rig (7950X3D in 7800X3D mode and RTX 4090) saw about 40% drop in fps and I decided to leave the game for bit and come back to it ….. it maybe worth me trying this downloaded save file you have as well for the city test for me ….. was the save from Nexus?
great info Mike, really interesting topic :)
Amazing work, Mike!
I wonder how the Linux MESA drivers that are not geared for any specific games would scale, probably about the same I would recon.
I still use the earlier gtx 950 drivers for years and everything works flawlessly, even callisto protocol worked smoothly
That was an eye opening video.First because all the reviewers sould write the patch that the game has at that time and second, us the consumers should always check a benchmark with the latest game patch when we search which gpu we will buy at that time to know what to expect when we have it in our rig! Thanks HC!
Thanks for putting care into this.
Anecdotally speaking: I have personally tried testing some very fine-grained, nitty-gritty performance things around gaming/3D APIs on PC, and the amount of variables to consider is pretty much staggering, at least if you let it be. I think the best we can do (especially speaking for myself as a hobbyist) is to remain calm, be honest about what we know and what we don't (especially if going to making public claims -- best to give clear context/caveats/notes when sharing what we do/don't know), check prior assumptions as much as we reasonably can (like this video!) and flag things for special attention if they don't have a reliable explanation yet (like the investigation of the Baldur's Gate 3 performance issue, good job following up on that and pretty strongly ruling out a driver issue there!).
So, all that is to say, science is not easy sometimes, but being, well "methodical" about applying the scientific method, we can advance truth, and highlight any areas of uncertainty for further investigation. Everything I saw in this video made me think HW Canucks will be on that right track in that regard!
I definitely don't have resources to go out and buy all the major PC parts! We (the public) definitely rely on reviewers there. So, I would just say, thank you so much trying to put the best info out there using a good process.
(P.S. it kind of makes me wonder about motherboard firmware updates, and Windows Updates, mostly thinking about CPU/memory/storage performance, hmm... But that's a thought for another day, I think. Heh. At least nothing's forcing a mobo update on a test system, generally. And for offline games I can get away with an older Windows update for a long while.)
well, reviewer's have moved away from any "scientific method" for testing because it takes too long, & you can't do it with the games they review with hardware now. Since they use games that everyone is playing & not an older stable game that doesn't have or need constant updates. This was done to please their audience.
Extraordinary work! Thank you,
Very interesting to see. And also a bit scary when you look at some framerate evolution over time ...
I feel like it would really benefit content like this to dive into how code actually works. When you are fixing bugs or adding new features, you get them working first, then get them running fast. This is why you have observed dips and then returns to normal. What makes code run fast varies between engines, languages, and compilers, but as a general example it could be something like converting a nested for loop into a single for loop. Or taking a function that scales quadratically and rewriting so scales linearly instead. Another example could be simplifying checks that are made in the main event loop so they take up less frametime each cycle.
Good stuff! Keep it up🎉
Have you reached out to Larian about the save file difference? I think they'd be pretty interested in the differences between them
We have and tried through our contacts at NVIDIA & AMD as well. No response yet. We would have loved to include some of their perspectives here.
Not sure if its a recent thing but the reason Istopped playing Baldurs gate was that one of the patches messed up garbage collection and eventually the game would cycle through all the garbage which caused issues. At that time, the issue was people would kinda get mmo-lag spikes along with lower fps. There was a community made tool that released at some point that let you clean up the save files.
Edit: Looked through my post history to find the fix but apparently this bug first appeared on patch 3 and was supposedly fixed patch 4 but I guess it wasnt fixed in its entirety? The original save file cleaner post seems to have been deleted on reddit as well :(
Cool! You should gain a lot of subscribers if you include both the game patches AND gpu driver updates in the results!
If possible, please also include performance of different OS on the same hardware, like Windows 10, 11, and 1 or 2 Linux distros.
Amazing work !
Interesting. One comment being: how about drivers for gaming laptops where manufacturers (ASUS in my case) do not deliver updates timely? I had a laptop with an RX 6800M at the time Starfield launched and I could not play the game at all. Installing the latest drivers from AMD was not easy, and resulted in system instability... Any thoughts?
impressive tests!
Keep up the good work! 👍
I've got some nice FPS increase in Ghost of Tsushima with the Driver that added support for this game on my RX 6600.
did you control windows version in this trial?
I knew it! I knew it was weird that my 4090 was running so low on BG3. So I started playing other games while waiting for a patch to fix it.
There's a lot of factors to consider, from stability issues because of specific hardware combinations (personally encountered this on Horizon Zero Dawn and Dead Space Remake), and sometimes it really is a driver issue where a good example of this is Palworld where in the case of AMD - they never released any drivers specifically addressing this game. So there were times when I update my drivers, I had to rollback (clean install with DDU of course) because with the latest drivers at that time it had micro stutters even when just walking about (game is notorious with shader stutters but its not the cause).
Let's casually forget about Windows 10 and Windows 11 updates.
Time have learned gamers that those can degrade performance too.
We maintained the Windows version throughout testing and have a separate piece covering Windows Updates planned. However unlike the game updates, Windows updates typically have very little to no performance impact...though some major ones do.
@@HardwareCanucks That's very great of you. You are very thorough in this process. Nice video tbh!
Good video and insightful, but if I was going to be running this test I would have ran It at 4K so the test would be more GPU bound. At 1440p the load is more spread across the GPU and CPU, and what about chipset drivers..?
Big, big fan of this Canadian CZcams channel.
I'd like to see a similar piece on NVIDIA's driver overhead, specifically when compared to AMD's in DX12 titles. My biggest concern as an NVIDIA user is how the overhead has been antithetical to NVIDIA's, "brute force it with horsepower" philosophy.
Hardware Unboxed have a video covering this.
Wow, this is shocking. Good reporting, Canucks!
It might be worth testing alternative drivers/platforms, such as dxvk & vkd3d, as well as Linux
Thanks for investigating.
Skyrim on the PS3 had a similar issue with save files, and from my admittedly limited and potentially incorrect understanding it was something akin to a memory leak causing save files to balloon in size.
Great great video. Nice job.
Turns out modern AAA games are not focusing on game performance or fun, I wonder what they are focusing on nowadays.
i have crazy problem here and got no fix for it
i have i5 14600k + 4070ti
say i cap a games on 90fps but each few seconds it hit 88fps for no reason i have tested it many times many games and that was on menu too not a game play yet it happen in game play too idk a fix for it or even what to call it to search for it
anyone can help?
You might want to clarify the title to avoid confusion. The title sounds like a general statement, as opposed to a statement about game performance in some games.
@HardwareCanucks With regards to Baldur's Gate 3, I don't remember exactly, but they introduced a very nasty bug in a patch. Basically, for quest progression the game would check a flag at some point in the story. After the patch the game would check that flag on everything you did afterwards in the story, the further you played, the more checks it would perform and the slower the game would get. It was supposed to be fixed in patch 5 but I think it is possible that some saves carry the bug forward.
That means you have to start a new game to de-active the flags.
I think their testing also contradicts that too because they two saves one the same patch that play completely different from each other.
One has high fps like 130fps while the other has half 60fps or blow the fps that the one has.
Did you all make sure to have ReBAR turned on specifically for the Intel GPU?
Great vid
On launch Baldur's gate 3 was causing my RTX 4070 to panic and run the fans on max at some points, it was crazy, like I had never heard that card before, and now it was just screeeaming like crazy, but then the next game update just fixed that. It never happened since. I was worried it might damage my GPU like it happened to those 3090s and 3080s with that Amazon MMO game.
Most tech reviewers are basic. Especially the big guys. I like what you've done here.
4:54 Tbf, it's a singleplayer game. Why would it have a ton of players right now? People finish it, then move on, so I'm not sure why you're mentioning that there are only a "handful" of people still interested in it. Thousands of people will keep playing it, just not at the same time like you see in multiplayer games. For instance, a few thousands might play it in a month, even if SteamDB says only 159 played in the last 24 hours.
8:09
Uhhh maybe this is part of the problem but did you not notice that the GPU on the right side is drawing significantly more power than the one on the left?
Were OS updates taken into account? Maybe that dip for AW was just Microsoft releasing a weird update.
I should have mentioned it but OS updates were paused during testing.
A similar issue seems to be happening in Ark: Survival Ascended. It seems that depending on which wild creatures are on the island performance can decrease as well as the GPU usage. I have been forced to use admin commands to wipe all wild creatures once every few days when frame rate tanks. This seems to have started when they moved from Unreal Engine 5.0 to 5.2.
Honestly Unreal engine 5.0+ is just bad in itself, it's trash
@@kevinerbs2778 I'm starting to think that might be the case, but I will have to check on other games running on it to be sure.
@@Donivar When I looked into Unreal engine 5 they have a version of Nvidia's PhysX built into it under a different name that is newer but is still cpu only you can attribute a lot of stuttering problem in some games to that.
came here because of ghost of tsushima drivers from of amd that actually broke the game for alot of players on certain hardware, with my 7800xt and the optimized drivers for the game the game crashed about every few mins... rolling back to the old driver fixed the issue entirely
Tell my 5700xt that drivers don’t matter. I’ve grown used to failures and green screens.
I get fairly regular lockups on my bog standard no oc 4090 too
This is not a video. It's gods work right there 😊
Good video.
BSG devs of Escape From Tarkov getting a major pass on this test thanks to their constant state of it still being (needlessly) in BETA? Would love to have seen some of the numbers on it & many of their patches! Tired of spending more money on hardware just for ever decreasing gains....
Tldr : it depends
Case closed
I dont think this is *fully* correct but this does open up a seriously interesting topic for discussion. Nevertheless this was great to see.
You are right. It isn't meant to be accurate in every situation. It just represents what we've seen in the capsule of games we ran this test on. I mean look at Avatar and Robocop. A literal flat line.
5:02 relatively minimal? In single frames perhaps, but the bump for the poor Ingreep is was 19% vs the other's 11%.🤔
Like the PUBG with every update performance goes down somewhat.
The only this happen and when nvidia add rebar to a title Months after launch like Forza Horizon 4 and 5
Thanks, Mike. You just made me feel a lot better about being a 1080p gamer. 🤣🤣🤣
Maybe make the messaging clearer?
Its not all about fps numbers. Latency and stability matters too. This is why I never play games on launch.
If they thing drivers dont make a difference then they should test the amd 24.1.1 drivers in bf2042 for instance. majour stutter feast compared to 23.12.1. Then later on even the 23.12.1 started to act up(suddenly the fps went down to 30fps after a while in warzone2, and I had to upgrade to latest one to get the fps drops sorted. super strange like if the latest wz2 patch broke the old amd drivers or amd has a timer on older drivers causing u to update to the latest one.
That's the one they first introduced AFMF, think of it like a game update, since AFMF run on driver level it probably affecting things that shouldn't have.
Back in the crossfire days waiting for a driver only to have a game update break it 👍
Gpu drivers do make a difference I noticed it in cyberpunk after I reset my pc my gpu was amd 6900 xt
They're desiged to optimise performance and occassionaly, one model will suffer a performance hit, to increase performance of others. It shouldn't and it doesn't make much sense but accidents can happen. This is taking a very rare case, which was fixed within weeks. Software problems can happen but when they do, they're fixed. Problems exist specifically in driver files, in improving performance and patch management, can effect multiple cards, without being able to test every single driver on every single card. It targets the card for the update being made.
No one said they didn't...
@@DaKrawnik the title of the video suggests that and I made my comment around the time the video was uploaded before I could watch it CZcams clickbait will always exist
Yeah wtf is with Baulders gate
These game devs are relying on upscaling technology instead of optimizing their games...
Hey, Now that's our band-aid solution! 🤣
Id cooling released done new LP coolers.
I think i hear a hmmmm in the mic maybe a bad ground or something is miss configures
No it's the air conditioning in the lab.
So, day 1 drivers do matter!
Hooray, some info to prove the point that driver updates and finewine don't change the picture that much. Maybe the glazing of every driver bringing improvements will stop but who am I kidding
Makes so much sense since nobody's buying new AAA (or AAAA) games these days tbh
You don't necessarily have to do things DIFFERENTLY, just give all of your specific parameters within a test that CAN be retested among peers with the same parameters....... I mean, that's what disclosing methodologies is all about. That's it. Period.
You must have forgotten how Nvidia's SLI and AMD's Xfire all failed because of the drivers.
which is lie. It failed because people are lazy & idiots. This clearly shows that "Drivers" were never the issue with those, its literally game engine being the main problems.
@@kevinerbs2778 There is no easy in the drivers to do Nvidia SLI or AMD's Xfire and developers were left holding the bag unless they got Nvidia's direct help. AMD has a poor history of documentation and that's why AMD Linux drivers take FOREVER. You don't know what you are talking about kevin.
@@michaelcarson8375 I know more than you because all of AMD's RDNA line from RDNA up through RDNA3 all support mGPU which what you talking about. Your clown ass doesn't have fucking clue about what you are talking in the first place cause you didn't even mention what the hell mGPU was. Another about developers are constantly lazy just like consumers too I mean look how they're implementing D.L.S.S & F RS to be mandator to be able to even run the games it's pure laziness with half-finished games that they patch later. So next time if you're going to reply to me at least have some fucking proof next time.
your conclusion isn't accurate for every game.. Eg World of warcraft on the 7000 series amd cards is still crashing when DX12 is selected a year after it was reported.
We can't account for every game. Just the ones we have this much experience with.
@@HardwareCanucksCould always do a in depth video on it, Would take awhile but would be worth the effort to show more of a showcase 💜
I don't understand why drives needs to improve performance, that is the game developer job, drivers are only for fix issues. Having an external company add drivers could destroy the performance intended for the game developer.
That's why I give Intel so much credit tons of newer games coming out and don't even list the arc cards in system requirements, So Intel has to go through many drivers in order to try and remedy the performance of the games.
Amd/Nvidia are just grand fathered in at this point for game devs so they don't have to try as hard for the effort.
Great video, love the testing. I always think hardware canucks has the best videos after all this time. Like how the pixel 8 came out and you guys said "hey why is it stuttering during scrolling lol' every other reviewer glazed over that.
As someone who has seen certain green tinted drivers completely hose a windows install on different cards, purple tinted screens, and random game crashes... Drivers absolutely matter. Hell RT still freezes in some games today. For me it's less about performance and more about saving me time by just working as it should.
But hardware unboxed did a better test for the performance side. Team green had more CPU overhead than team red. That would matter for folks who tip the budget towards GPU by getting a cheaper CPU.
How can you go through this much data without going mad ? this is the kind of stuff you go through when you're working on your thesis, but you guys are doing it every few days! kudos for the amazing effort.
Wow, great study, guys!!
All variables matter!!!
Wow! The A770 had infinite improvement!
NIIIICE
DirectX HAL was supposed to solve this.
@HardwareCanucks
Hello, I want to buy a 27 inch gaming monitor for my gaming pc (RTX 4080 super). I play a balanced mix of fps shooters and AAA titles. (Like Apex legends and cyberpunk)
I'm looking at a 4K IPS and a QHD OLED Modell.
Would you say you can see a difference between QHD and 4K on 27 inch?
So do you think it's better to play QHD and enjoy the better colours & HDR gaming content with the OLED if you can't see any difference on 27 inches?
Or is there a noticeable difference, which is why I should go with 4K IPS to get all the details of the game world and not waste the power of the GPU?
(The monitors I have in mind are the ASUS ROG Swift OLED PG27AQDM or Dell Alienware AW2725DF (for QHD OLED) and the MSI 27U (for 4k)
Thank you for ur help!
Just research online. This question has been asked millions of times now...
OLED monitor will blow your mind away with color depth once you compare it to other monitors you never will return to anything else my friend.
Im waiting until Nvidia abandon new game optimization driver for rtx 3000 series (aka Amd fine wine) before upgrade to newer series
The canucks did it again
Mike is a madman to welcome this kind of stress, he's pretty much the "I don’t want peace, I want problems always" guy. My hat off to you good sir.
I use linux so windows won't uninstall them mid update.
BG3 uses 100% of my 3060 Ti.
They do not matter ? try Dying light 2 first blood questline all the way up to skullface first cutscene with a 7900 XTX on 24.2.1 drivers or older and then try it again on 24.3.1 and newer, then tell me if drivers do not matter, the game is still broken today but not as much as it used to.
You're just now figuring out that updates to games can impact performance?
Sounds like the game developers did not beta test their games.
People who buy their games on launch
ot long after launch, are the beta testers.