#[DUAL GPU SYSTEMS] Performance Issues and Troubleshooting
1 messages · Page 3 of 1
Well it does have a disclaimer about HDR. Maybe it's too obscure
bought used ^^
for how much?
HDR is one of the notes in the spreadsheet
4k turns it way down again
and not saying its anyone's fault, sorry if it came out like that. just disappointed 😛
55 euro
Paper weight
😭
i have a wife and kids who games at 2k or less. maybe they want some generated frames 😛
77x2 at 1440p with HDR makes sense for an rx6400. So it's just worse than expected for 4k
there we go
not a waste after all
Lol
BTW what did it drop to at 4k?
With SDR
Was it something like 35x2 or 40x2?
52x2
100% res scale?
yes
That's actually pretty decent
and 1 frame latency
I'll put it in the chart when I get the time
well my 3080ti alone does 89 😅
so maybe for some lower cards its ok?
The RX 6400 is an absolute unit, just not one for 4k
Only in this server someone would mention "absolute unit" and "RX6400" unironically in the same sentence 😆
hows the RX6600? using it for 4k?
"RX 6500 Xt my beloved"
⚰️
Yes, 4K 60/120 no problem.
My max was 120/240 or 60/480 at 25% Res scale
What psu you running? Only have a 850, afraid I might need more juice
nah your good
you are meant to undervolt the cards
If only I could find my cables then 😅
my 1070 used 60w when used for lsfg and undervolted
My RTX 3060 12GB + RX 6600M + Ryzen 5 5500 are all undervolted and in total consume like 220W average
Wow, that is impressive
850 as well
I'm sure I never use more than 500-600 with a 3070 Ti
Because of these I feel like my 750W PSU was overkill lol
170 -> 130 is insane on that 3060
better safe than sorry
Yeah that was my thought process
I spent a while fine tuning the undervolt on it, I got a satisfying result
its not like psus are expensive either
Mine costed me 55 dollars
It's an xpg pylon
bronze rated?
Yes
well its not the best but its good you went for a 750w instead of something less
I think mine was like 150 ish in terms of usd
850w platinum
its nice to be worry free I guess
Bigger psu is almost always nice since they don't use the fan if components draw below a certain threshold. So it doesn't get clogged with dust years after.
Yeah I have a 600W one and I already changed it's fan once 😆
damn
I have tried to undervolt but my card ain't having it. But why would I undervolt it? Have I misunderstood the intention of LS. I thought if I could generate 80 fps without ls I could turn up the eye candy limit it to 60 fps and then generate 60 more using LS and another GPU.
you lose nothing when you undervolt and you keep the power cost low
means less strain for the psu and less heat generated
performance can even be improved if undervolted and overclocked at the same time
I have tried over and over. Must be doing something wrong. Eventually when gaming it just crashes on me.
might be too extreme of an undervolt
My B570 uses 60w when used for LSFG when stock (attempting to undervolt severely corrupted drivers)
wth corrupted the drivers?
60w is really good for stock though
The rx6600 uses like 50w
100% usage though?
Yeah lower usage means less wattage
My RX 6600M is very harshly undervolted and 58W is the most it draws
I didn't undervolt the RX6600 at all, 50W for normal usage is completely fine for me
@onyx warren just be clear, its 6600 not 6600 xt right? also do you use HDR?
Yes and no
But it should easily be able to do 4K 60/120 even with 25% less performance from the HDR
Okay Thanks. It costs 3-4 times more than the 6400. Need to think about it a bit 😅
4 times!?
no way
2nd hand 6600?
I got one for 150€ used
I got a 6600xt used for around 150 too
And a 6400 would've run me almost 100€ used
There's no second hand in my region 😅 not atm anyway. eBay is as expensive as new
4090s rn I bet
might be cheaper than that 6600
Yeah and they cost more than when they were bought 😅 people think they got pure gold in their hands
the market is so stupid lol
What about the RX 5600/5700 XT?
135 euro
Denmark
I see
.
108 euro
Alright, the 5700 xt offers a great uplift for that extra you would pay for it
5600xt is cutting it too close imo aim for something higer
5700XT for 135€ is a steal imo
Yes but imagine the 5600 xt costs 60 euros
yea no idea why it costs so much in his area
Is your 4K120 screen connected with DP or HDMI?
HDMI
An RX 5600 XT is a great option when it costs way less than a 5700 XT
This isn't the case so it's better to just get a 5700 xt
what about the normal 5700?
Denmark has some of the craziest taxes. You know the tarifs trump talks about. That is our everyday for everything.
5000 series don't have HDMI 2.1 and neither the 6000 series below the 6600
that sucks rip
I never tried the normal 5700 for LSFG, I just know my brother has it and it runs well, should be only like 12% behind a 5700 xt
But people love to price the rx 5700 as if it were the xt
its less than 12%
its like on par even
with the bios flash
So also 6400 fails in that field? 😅
it would be interesting to see benchmarks for it with lsfg
European tariffs are something else, it's dumb
Yes, you could only output 4K60 with that one
That's why I went with the 6600 in the first place.
Cheapest card with hdmi 2.1 that works for 4K120 lsfg
DisplayPort my beloved
this
A lot of TVs still don't have it 
smh
4 HDMI ports and that's it
My chinese monitor does 
Monitors and TVs are very different
hdmi to dp adapters still suck right?
Oh yes
Oh wow, I really screwed up on that one. At least the wife seems interested in generating some frames. Guess Sims at 240 is something 😅
lucky that you managed to find something for it to do atleast
I had to get one for my rx6600 since it only has one HDMI port and that is reserved for my TV 
And my 3rd monitor is so old that it only has HDMI and DVI
truly a based 6600
6400 has HDMI 2.1 and DisplayPort 1.4.
I researched before I got one and didn't find one that had hdmi 2.1
Wait no, I think it was because they all just had 2 outputs and I have 3 monitors

That was the problem
Bro casually gaslit everyone and we fell for it ⚰️
🍷
oh good. thought i did my research on the HDMI. put wouldnt be the first time i was wrong today 😛
If i could get a 4080 at the same price as a RX660 would that be worth it? then my 3080 ti would become the second gpu ofc. but generally amd is better as a secondary right?
In what universe are you finding 4080 that costs the same as rx 6600 😆
I like AMD but not even RX 6600 for 200 bucks vs RTX 3060 12GB for 160 bucks comes close to how much of an overwhelmingly obvious choice this is
hard choice
i dont know what to say man 😛 i dont make the used market around here. i just saw it and went "wouldnt this be better, and for the same price"
if its somehow real I would be on that in a nanosecond
ill give it a go. looking at the other listings from that person it seems fishy. but they live kinda close, so could just ask for meet up.
Yeah there is no chance. Must be broken or something
Even then people want more than RX 6600 😆
Hey, I just wanted to double check with dc before buying
Monitor: WQHD/240hz
Mobo: b650 tomahawk wifi that has 2x 4.0 pcie that is shared with m2 (only impacts two way speed of SSD as I understand)
Will RX6400 for 1440p work? From what I've seen it's 14-15% slower on pcie 3.0 4x (4.0 2x)
I found used RX6600 for 90$ 
Should work fine if 4.0 x2 works the same as 3.0 x4
X2 as in 2 pcie lanes?
Sounds daring, my recommendation is making sure you can return it if the thing doesn't work
I got a used rx 5700 for 90$
specs from msi
Doesn't make much sense (cuz it's already 2x) but from what I understand SSD uses both ways while GPU uses one way, so it will only negatively impact my SSD (which is nowhere those speeds anyway)
Wow MSI really went for that 4¢ savings with this
I'd test that scenario before buying anything. @clear roost
flexing now are we?
You're not only gonna gimp your 2nd gpu but your ssd as well which sucks
Some GPU's suffer from having less lanes even if they get nice bandwidth from running on pcie 4.0
yea that's difficult for me 😭
5700x3d for around 120 usd was good
Based
My best is the legendary 65 usd 5600 xt that was like new
🙏
I just hope it works without issues
then I will truly ascend
👀
crazy fps
Get those spreadsheet numbers please
my resolution isn't on there 😅
I need to get a generic undervolt for it too first
got to minmax
What's your resolution?
3440x1440
Hmm, are you unable to set it to 2560x1440p?
I'll try to
so I just 2x my base fps
and see how much it can do
on stock settings
hmm what game
hmmmm
why is it like capped?
it only goes as far as my max refresh rate
its doing 330fps but it can do more if I could uncap it
Base FPS cap out at your refresh rate
LSFG can't capture beyond that
what's your refresh rate?
take a screenshot with usage numbers
Is this 2560x1440p?
Nah it's fine
my res 💀
If you are at 165/330 with 93% usage, then I guess it's fine to assume the limit is somewhere between 340 and 350
anyways
I'll try changing it to 1440p on windows
@rain hull
its far stronger than what the chart said
the rx 6600 xt is still proving to be on a different league
more than once we scratched our heads about it
Because my RX 6600(and the M)'s only did 115/230
What's the second PCIE slot?
I just want to find out why is the RX 6600 XT so many leagues ahead
Compared to the two RX 6600's I tested
yea its an incredible difference
Could be physical slot size, could be bandwidth, or could be a combination of that added to the fact it's an rx 6600 xt
But my processor only allows pcie 3.0 so I can't test 4.0 vs 3.0 and compare the loss of performance
same
but 3.0 x8 vs 3.0 x4 would make an interesting test
I'm very limited but it works
lemme go and check
So after testing the 6600xt I can confirm that nvidia sucks for dual gpus imo
Atleast mix with amd being render
So many issues it was unreal
Tell me something new
I think its x4 now
Time to check
Amd control panel with 2 gpus is seamless
Crazy stuff
I can set my RTX 3060 that's on my main slot to be the LSFG GPU with pcie 3.0 x16 lanes
When I did it could only do 100/200 fps on X2 at 1440p
t-r-a-s-h
😭
Meanwhile my RX 6600M that was running on pcie 3.0 x4 could do 115/230
Nvidia overhead moment
Yea idk
I don't think I can x4 it
nevermind
huhhhhhhh
why the hell is my main gpu x4 but not the 6600xt
what does gpu z say?
Power to LS
Unless i move the card to main slot this ain't happening
Poor pc is struggling as is 😭
Barely fit this big boy in
Its bigger than my main gpu
Oh well, it's not that important
either way I'm real impressed by this card
omg wait
lemme redo the test
yep my ls settings were reset 💀
actual real numbers now
doesn't make a difference in 1440p but for my native res it should be much less now
What changed?
res scale 😅
still the same 1440p numbers though
it can go further than that
but my native res should change now
I'm bottlenecked here
pcie?
render gpu can't push out any more frames at that point
but with a lower res scale it can do my refresh rate 2x
and render gpu can push out 20 more fps
@rain hull it is confirmed that the 6600xt can do 330fps on 1440p but I think it can go further than that if I wasn't refresh rate bottlenecked and pcie bottlenecked too
Can't you unlock the rr?
apparently its capped by my monitor's rr
ls won't capture any more frames than that
It did for me
how did you do it?
already off for me
Weird
I mean it wouldn't do much more anyways, you're at 95% usage
This is pretty much the highest stable rate you're gonna get
true I'd guess like 10 more fps probably at most
And then it would be unstable anyways
Undervolted my 6600 yesterday, first thing my PC did was crash
rip
But got a stable one now
how did you undervolt?
I think I put it to 1080mV
whats the default?
And the clocks like 2200 min and 2700 max
1150?
I guess I'll push mine lower
only 1100 rn
1080 seems stable
1065 = crash
1070 seems stable
it may look like I need to do the flexi fix 🥲
Yea, estimates on the chart
Look at the LSFG 2 section of it, it's safe to say that I rounded down
Damn
dw about those results
I'd give higher multipliers and VSR a try
My ls settings got screwed with but later got it fixed
Wdym
Low res scale or smth?
Ah
A bit higher up
It might do like 340 or even 350 fps at 1440p but I'm pcie bottlenecked i think
And refresh rate locked too
I believe the previous person to test it was also limited by a 165hz monitor
A750 results too, it was just barely refresh rate limited so I marked it as an estimate and rounded up a little
Goes to show how strong they are for lsfg
That aside, does it work well in games?
Any odd perf hit from outputting from the 6600xt or activating LS?
It did
But it wasn't the same as the 1070
Summarize
So I lose fps even without ls like last time and when I use ls, I lost even more fps
It didn't seem hardware related so I'm doing software rn
But if it doesn't work I will 100% know what's doing it and fix it
I have an idea...
Like a hard cap or percentage worse FPS? Was the render GPU using lower wattage?
Not a cap but general worse fps
It might be because of my pcie layout of my devices tbh
If it is I'll move my stuff
Yea that would make sense
I have similar issues when I enable low latency mode in nvcp
So it can be software
Just made a comparison to stock, undervolting doesn't even drop a single Watt in normal usage 
So I got my 4060, went through the dual gpu setup guide and got it working. My problem is that while running the same config in games and in ls, didn't change any settings anywhere, I'm getting a lot of frame drops and stutters. But both GPUs, the 4070 for graphics hovers around 70% and the 4060 around 50% usage in helldivers 2.
At 1440p, x3 from 46 fps scale at 70, ls1 performance and scaling factor at 1.3.
Those exact settings worked perfectly on the 4070 alone, but they don't on the 4060, even lowering them doesn't fix the stutters. Same problem in god of war ragnarok.
This is my motherboard for dual GPU
Chad
yea no idea whats going on
I'll try to retrace my steps then
Try disabling resizable bar in bios
and HAGS
It was already disabled apparently
After disabling it, it gives me either a black screen when enabling ls or puts back to the desktop screen
Edit: forget that, my game was fullscreen XD
Reset NVCP settings
Mine is running 912 mv 
@rain hull I think I solved the fps drop without ls
but the ls one still happens
wait I think I kinda fixed it a bit
You ~ fixed it
I wish
Thanks for the support earlier this week everyone. Now I got this 😀
Excellent choice
Thanks 😀
Unfortunately I just can't find my psu cables. Seasonic shipped me some new ones but wont arrive until Monday.
Just to make sure, @elder smelt @onyx warren you both are using Windows 11 right?
I am
Make sure they're compatible with your PSU
Cables from different PSU models are often wired differently, and mixing them up can fry your PC
Yeah that's why I wrote directly to seasonic 😀
Alright
I stubbornly run Windows 10 with the same install since 2020 and haven't ran into that gradual performance loss thing yet
you got it in w11?
I haven't tried Windows 11 on my desktop PC
But I have yet to see someone that uses Windows 10 get that issue you and Flexi had
hmm
most people use w11 now so maybe thats why
still I can't tell if this is software or a pcie bottleneck
installing the amd chipset made my performance much better but its not the best it was like before
Jesus
Yes
Amd has to pay the dev of this amazing tool. Finally giving folks an actual good reason to buy their goofy aah graphic cards.
The RX 6600 was already a banger even before LS
But I agree for the rx 6400 and the 6500 xt
Cards no one wanted but now plenty want
Those were pretty much ewaste until now
And who gonna try 9070xtx
I'm going to assume it'll be powerful enough for framegen that testing it will be impossible due to LS not being able to capture beyond refresh rate
Except for maybe 4k240 or 4k360
E.g, on a 165hz screen your testing is limited to 165x2
Someone (Dranas) tested the 7900xtx and 4080 and the 7900xtx was a good 20% more capable, but they had to use multipliers above X2 because they were refresh rate limited
So I couldn't put it on the chart 🥲
So for lossless scaling 7900xtx was around 20% faster than 4080?
We know AMD is faster for frame gen than NVIDIA. But it's it the case also in 5000 series?
Or NVIDIA did something to improve this type of calculations
No idea
I've been busy lately as you might have noticed. Got heavy works to do
🤯
So I didn't catch up with hardware news for the past 2 weeks
Ok I think I may be pcie bottlenecked ngl
there is some kind of really weird overhead
@clear sundial did you ever have less framerates on dual gpus before you went to pcie 4?
Hmm if there is some kind of overhead and its not from pcie, what would it even come from? I have to brainstorm 🤔
it looks like a cpu bottleneck but isn't specifically one
maybe ram related too
Don't go off by CPU usage, check usage of both GPU's
yea my gpu usage gets lower
I never check cpu usage because its always low
but my fps takes a hit
looks like a cpu bottleneck but it isn't one
PCI-E 4.0 to PCI-E 3.0 was always a performance drop for me on dual GPU (x8 CPU) on the LSFG GPU.
On the main card, not really an issue. I didn't even feel a difference going from PCI 4.0 x8 to PCI 4.0 x16 or vice-versa
PCI 4.0 x8 (CPU) to PCI 4.0 x4 (Chipset) was a massive drop in performance in both LSFG GPU and main GPU
I had similar jank from having low latency mode enabled
I'm pretty sure I got everything disabled in my control panel unless its doing similar behaviour without a setting
so gen 3 x8 vs gen 4 x8 always has less performance?
3.0 x8 second GPU, 4.0 x4 render GPU, around a 60fps base I see basically 0 hit on base FPS or frametimes outputting from my B570, and CPU usage is placebo range, maybe lower
so technically we have the same bandwidth
I'm at 1440p 16x9, you?
Potentially, video copies through RAM, but RAM bandwidth significantly exceeds PCIe bandwidth so it shouldn't be a limit unless you're running PCIe 5.0 x16 with a single DDR4 stick
I'll keep that in mind because my ram setup is very goofy
also also
low latency mode changes what?
if its buffer frames maybe I need to change the default amount of frames my gpu buffers
For me yes, but I do suspect it being the 4060 ti's fault. I didn't test with another GPU yet
And this is also on 4K. 1440p should be easy to drive in comparison
4k oh
how much performance loss are we talking?
3440x1440 is nearly half of 4k so it shouldn't be too bad
Around 33% #1328337747399938138 message see here
146 fps -> 109 fps
ok it does help a lot but it still does a hit
pcie issue here?
and I'm talking about when using lsfg
without lsfg, performance might get worse in some cases still though
ofc
😭
a game could be running like 100fps with single gpu or display gpu being render too but as I use a secondary card as display it drops to 80s or 90s
yep
it can even go high 70s
it could be pcie but it might be more than that
I'm not sure
so dual gpus is not perfect in the end
sad, i fear i wont achieve my 100.00% stable framerate/frametime with 5090/7800xt 5k 240hz
using dual mode
for me yea but for others they seem fine

I'm sure we got some great minds here in general chat
It's perfect for me
Maybe because I'm using an AMD gpu
both my gpus are amd but here I am 😭
this^
Whenever I try to test my 3060 for LSFG it just becomes hyper jank
You can tell from Ravenger's notes on it in the chart that none of us is happy with the results it gave
no
oh?
nvidia + amd is good
perfect
example here without lsfg
usually I can hit 175fps
but it got lowered
due to some kind of overhead
if its with lsfg 💀
nein
I'm using RTX 3060 12GB + RX 6600M
this is the cpu bottleneck looking thing I was talking about
looks really great for you
What game is that?
NEIN?!!! WARUMMMM
7days to die
This screams CPU bottleneck to me, what's your current CPU?
ICH HEAR EIN DEUTCHEN PERSONEN SPRECHE
5700x3d
but it isn't
it shouldn't be
Also check with task manager if you have windows doing anything dumb on the background
usually its fine without secondary gpu
One time I was getting CPU bottlenecked quickly with LSFG enabled because Windows 10 was doing a background scan
and it took 20% of my CPU
background scan? tf
Malware scan
my cpu usage is less than 40%
Windows defender jank
disable that crap
it may look like a cpu bottleneck but it isn't
The process to disable it on Windows 10 isn't straightforward
dcontrol
1 button fix
works on w11 too
Now after the Windows Reinstall Elden Ring doesnt work anymore 
Nvm had to run it as admin for some reason
Can you guys run an experiment for me?. I think I've spotted a weird behavior but I am unable to solve it until I can confirm this happens for someone else.
Whenever I load a game that does not have native HDR support in HDR (even if it doesn't enable AutoHDR) the GPU2 (used for LS) starts at 60% and never lowers while doing basically nothing. This restricts the power of the GPU for actually doing LSFG so I had to lower the FG Res to 25% to have a lag free experience.
If the same game is loaded in SDR mode, GPU2 is not taxed and I can use it at 100% FG Res without latency.
The experiment is simple. Open up any game that does not support HDR, such as Dying Light 2 or Once Human, but in HDR mode, watch the second GPU (the one connected to the monitor) and tell me if the load in your case goes up to 60% (it should stay below 30% if it's working fine).
Guys @rain hull and @elder smelt and anyone else on the AM4 platform, I need you to do a quick test to confirm a theory of mine, if you can. I might've found out the dual GPU "fix".
Just go here https://www.aida64.com/downloads and download AIDA64 portable and run the memory test and share a screenshot of this window
👀
isnt that the trial versioon?
like half the info
its funny tho I literally ran the benchmark a few hours ago
yes, but it's fine
The values I need are shown on the trial version too
it was in safe mode though if that makes a difference
yes it might affect latency (positively), but that's not the value I'm after
I think it's L1, L2 and L3 cache related
well you may be on the right track considering my issue looks like a cpu bottleneck (it isn't)
But I need people to test that for me, even people with AM5 or other platforms. I'm looking specially for people that don't have issues with dual GPUs to test
@versed knoll are you on am4?
7800X3D
yes they are
9950x same memory config and timings
so much faster cache
I made a big mistake in not testing dual GPUs when I got that CPU, because I was 100% sure it couldn't be CPU bottleneck, so I wrote off the 4060 ti completely. But yesterday I installed the GPU back in and noticed that RTX super resolution doesn't lag anymore
wait you think its a cpu bottleneck?
I reinstalled this GPU over 5 times with my 7800X3D, always using a flashlight to confirm it was 100% plugged in, since I'm very paranoid about bad mounts. There's no way I installed it badly 5 times in a row
wait
is it an x3d issue?
the cache speed difference
its a drastic change
multiple times faster
I think it's L1, L2 and L3 cache related. So, indirectly, it might be a CPU bottleneck yes, since higher cores = higher cache speeds
hmmm
But, like I said, I don't think I'm correct. It's a very stupid theory
But I've tried everything, this was driving me mad
whats the cache speed on a non x3d cpu?
Oh, yes
can you do his test?
Extreme?
yeah, it can be trial version please
and I want to ask you if you have issues with dual GPU or not (dropping frames)
I highly doubt that I'm suffering through a pcie bottleneck too so it could 100% be cpu related
Even though Ravenger is running a 3900x, looking through the web, the aida64 numbers are higher than even 7800X3D. That's why I want to confirm this theory
I see
so aside from me and you who else had issues like this?
flexi?
I think he got it fixed though
3900x (from the web) https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fl89dv57sjeu41.jpg
yea quite higher
@clear sundial
crazy latency 👀
I knew someone would point out my cas latency 🥲
That's because of my ram, don't look at it
the reason I thought about this theory is that the drops happened when the GPU wasn't at full load. I was positive it was a memory controller bottleneck, but that didn't explain other people using an RX 6600 with success, so that's not it
so dranas the cpu you had this issue with was an x3d cpu?
Thank you, did you find issues with dual GPU?
yes
Nope, only wonky rtx 3060 results but the rx 6600m was very smooth
now i'm curious, is this an x3d issue???
who has an x3d but is also running dual gpus here
we need them
https://youtu.be/FZlwfWZ-0JM?t=140
https://youtu.be/5z1mmyORVGA?t=35
https://youtu.be/3LPsDKEv3V8?t=8
PC Used:
-B550 Plus AC HES
-Ryzen 5 5500
-40GB DDR4 3200Mhz
-RTX 3060 12GB
-RX 6600M
The RX6600M was off-loading the Frame Generation when Lossless Scaling was used to show the benefits of using a secondary GPU compared to using an in-game implementation.
Timestamps:
Settings: 0:00
No Frame Generation: 0:27
FSR 3 Frame Generation: 1:06
LSFG 3....
PC Used:
-B550 Plus AC HES
-Ryzen 5 5500
-40GB DDR4 3200Mhz
-RTX 3060 12GB
-RX 6600M
The RX6600M was offloading the Frame Generation whenever Lossless Scaling was used to show the benefits of using a secondary GPU compared to using an in-game implementation.
Timestamps:
Settings: 0:00
No FG: 0:18
LSFG 3.0 X2: 0:30
FSR 3 Frame Generation: 0:50
...
Specifications:
-B550 Asus Plus AC HES
-Ryzen 5 5500
-RTX 3060 12GB (Main Graphics Card)
-RX 6600M (LSFG Graphics Card)
-40 GB DDR4 3200 Mhz RAM
Tested game settings:
-Main GPU (RTX 3060 12GB) was running at PCIE 3.0 X16
-LSFG GPU (RX 6600M) was running at PCIE 3.0 X4
-1440p 165hz monitor
-82 FPS cap in RTSS to ensure smoothness and avoid teari...
Life is good in my PC when the nvidia GPU is doing the rendering
The only other person I remember having no issues with dual GPU was running a 9800X3D
But looking through the web, even the 9800X3D has crazy values on aida64
yea its the latest and greatest cpu
so 7800x3d and 5700x3d had issues
but 9800x3d was fine
it can't be a coincidence that both cpus are x3ds imo
it might relate to something
your poor CPU was getting hammered on the wilds benchmark jesus
playable by the way
But barely
So people with AM4 x3d CPU's are dropping frames?
looking through your vids you are affected though, but you also have no fps cap
I'm just extremely sensitive to jitter
playable but unbearable
we need more x3d folks here
lossless when you have a slight change in fps will result in jitter, even with gsync/freesync on
For the longest time I had the crappiest PC mankind knew, so maybe I'm not too sensitive to it
I've been there too 😅 I remember playing games at 30 fps at 480p and enjoying them. Nowadays even a slight microstutter / judder annoys me to the point of me reducing settings
same
I remember when I played with an intel igpu
0 pixels dying light
hd 6450 + phenom ii x6 1090t vs the world in 2018
I don't mind playing on lower fps as long as it's 100% stable. Having big oscilations feels horrible for me
12 fps and 200 ping in ultra low quality settings in 768p in war thunder
2019
2025
I remember in 2017 while waiting for my am5 system to arrive, I had to put my 980 ti on this PC 😅 https://www.userbenchmark.com/UserRun/4396612
beautiful
am5 in 2017 huh
early access
a time traveler
ah oops 😛
tree trunk
*am4, too lazy to change
Also in my old PC I was dumb af about taking care of it so the CPU was constantly thermal throttling
Somehow it survived the massive abuse
the cherry on top
Yeah I wasn't joking when I said it was the worst mankind could know
My first PC back in 2008 (hp compaq presario) literally burned to death. I had black paint in my hands from using the keyboard 😅 100ºC on the CPU while on energy mode in Windows
At least I had 32gb of ram for:
insanity
I even tried running skyrim on it. 480p, potato mod settings, 30 fps on the NVIDIA Geforce 9200 GE 256mb vram
btw that ssd there? I've been using that as an OS drive in this PC until 2 weeks ago. Still works with 55 TB written to it
Yeah, I hope my other ones do too! I've never had an HDD last that long lol
@lavish juniper you still have a 5700x3d?
Ah fuck, I'm on a x3d.... I won't be able to test until tomorrow though since I'm waiting for a PCIe psu cable.
its not like your gonna have issues
its just that all of the people who do have issues have an x3d
so yea
what cpu do you have?
7800x3d
uh oh
same cpu dranas had issues with
its the moment of truth
if you set everything up and you have issues then we will know
You were saying?😅😭
you're fine definitely fr
just believe
it does get me thinking though
you had issues with the 6400 yea?
but it didn't lower your base fps did it?
bro
thats the same issue
that im dealing with
💀
oh no
I hope it works for you
or else we found our culprit
Wonder if it's possible to fix in future releases
Yeah
I doubt it
its more hardware related
or software in windows
but it could just be a false alarm
might work fine for you
just make sure to let us stay tuned
we have to see where this ends
so cpttombstone is the 9800x3d guy who has a working setup
hmm
well we'll see soon enough
Don't know much about motherboards. Could there be a feature difference?
Or Ram? CL or amount
sure there can be but its crazy that 3 x3d users experience a similar issue
and on 2 different platforms
and lets not forgot that dranas fixed his issue by upgrading from a 7800x3d to a 9950x
I can't think that somehow all of it is a coincidence but then again who knows
also this issue looks like a cpu bottleneck but it shouldn't be happening
there needs to be more testing to be made certain
tbf rx 550 is very bad
but
it sounds like a similar issue
ugh, hope i can return this GPU if true
I hope it works fine for you I really do
It sucks for me if thats the case too
I got a spare ryzen 5 5600x but that means downgrading
so what cards are you matching @elder smelt
7900xt + 6600xt
and the same issue I've had it occur on my 1070 too but I thought it was too weak for my 7900xt so I ignored it
🙏
when it works it works
whats the running theory on why am4 has issues?
sorry playing catch-up
l1, l2, l3 cache?
Might be x3d related not sure
Because dranas also had issues and he was using a 7800x3d
Upgraded to 9950x and boom
Fixed
Same memory config, same mobo but different cpu only
It gets worse with lsfg but yea
Could be different issues same cpu type but not sure
Theres only like 5? Dual gpu users with x3d cpus
I'm having a weird issue with my 6950xt and 1080ti setup, in monster hunter wilds, for some reason even though everything is set up correctly, my frame rate is very low and my gpu isn't even utilized fully, I didn't seem to have that happen in ff7 rebirth and baldur's gate 3 (only things i managed to test)
Tried benchmarks like firestrike to test it out too and see if there's any limitations or something wrong compared to before adding the 1080ti in, but everything was fine.
@clear sundial tried it on my 5950X
Thanks for the screenshots guys, do you also have issues with dual GPU? Or does it run flawlessly? @left cape @mental peak
Even though most speeds are below am5, the L1 cache is a lot faster
And Pyro's setup demolishes the 3d chips
Not flawless no, I get stutters and frame drops even when both GPUs are below 70%, I did disable hags, reset the NVCP settings and resizable bar in BIOS was already turned off beforehand
How so? What numbers should I be looking at
Win 11? Also what motherboard are you using?
I thought the x3d chips had better performance and cache than the non x3d, why's that?
I cant say flawlessly, but it works more often than not.
in some cases I get 10-20 less base fps than with a single GPU set up. Other times it works absolutely flawlessly. cpu could be the culprit, but my game sample size is pretty small and I have had a lot of game specific issues
Ok im awake now
i have to test more, life's been busy
godspeed
I made a post on the ls reddit and I got some really good insights ngl
7950x3d was fine for someone
But then someone else on an intel system had the same issue
Its so random
Maybe it's not that?
yea its just speculation so far
wtf is this timespy result
https://www.3dmark.com/3dm/126165257 display connected to 7900xt
https://www.3dmark.com/3dm/126166144 connected to 6600xt
so it only happens in games?
so weird
huh. how many games/what games have you tested
I've tested 3
but 2 main ones
and both are different
one is cpu heavy other gpu heavy
maybe theres some kind of load that triggers it
wait I just tried something in 7 days to die rn
instead of normal fullscreen I did exclusive fullscreen and my fps skyrocketed
tf
normal fullscreen aka borderless
exclusive fullscreen
I have to try this on my 7900xt as display brb
If timespy was running exclusive fullscreen that explains the result
but it seems general performance improved a lot too with display connected to 7900xt
from like 95fps to 115fps while out of the pause menu
timespy is completely GPU bound. I've noticed similar results on my PC. Slightly higher performance connected to the LSFG GPU vs solo. I'm guessing since it's not displaying an image, it slightly increases performance? I dunno
They have less cores. With bigger core CPUs you have more L1 and L2 cache spread to all cores, so they can do more work overall
could be that my undervolt was on but it makes no difference imo
its quite the similar result
If they made, say, a 10950x with all cores having 3D cache, I wouldn't be surprised that those numbers would be even higher
could explain why a 7950x3d is fine too
I would try disabling SVM if you don't use virtual machines, see if it helps
with all the bullsht that windows 11 has running with virtualization in the background, I wouldn't be surprised if it gave you a performance hit. (I have that disabled by default)
@clear sundial thoughts on this?
its not perfect but more performance comes back
you're running less GPU usage, so something is interfering with your rendering GPU in borderless
Interesting, although I still have huge fps drops in generated frames that I didn't happen before with only one GPU.
I do 46x3 in helldivers (with gsync), it'll randomly go to 110 fps fg, and both GPUs are at 50%. No matter the resolution scale, I tried it at 25, x2 instead, 1080p and it still drops out of nowhere
What's very strange Pyro, is that you're running on a pretty beefy motherboard with very good PCI-E lanes throughput
Exactly
Basically the same system as I, but on a different platform
I would try disabling SVM in the bios and see if it does anything. Hags disabled is a must too
5950x
5950x
but you're also running some very high latency ram though
Can you check all core usage with afterburner? (individual cores)
does latency make a difference? mine was 59ns but here I am
It helps a ton in CPU bound situation, but it isn't a magic bullet
On 3D cache CPUs it isn't that big of a deal
Just checked. SVM was already disabled, so is hags, resize bar and I reset nvcp settings as well yesterday
So that rules out virtualization shenenigans... 🤔
So it might not even be a cache difference, then what the heck is it
It just doesn't make sense, how can a CPU change affect the GPU performance when it's not the bottleneck
you are the only person who managed to solve it thus far
and it was a cpu change
its bizarre
I'm using the same windows version, same drivers, same everything. Just plopped it in, enabled EXPO and the settings I had on my previous CPU
it makes absolutely no sense here
and pyro has the same issue somehow with a 5950x
we need more data
It also doesn't matter if the GPUs are undervolted or not, I tested both with and without and it still happens
same here
its regardless of the gpu
very stupid question, but do you guys run lossless in the same drive as the game's themselves?
no but it doesn't make a difference if my games are running crap without it 😄
No. I run lossless on my main ssd, most of my games are on my second ssd
"6800xt is connected to pcie 5.0 and 3050 is right below it at pcie 3.0 x4 cpu is a 12600k and MB is a z690 msi pro" same issue as us
So, so far it happens regardless of the GPU, cpu, motherboard, game and ssd
somehow dranas fixed it with a 9950x
3.0 x4 is very weak for a second card
I agree
but if he experienced a pcie bottleneck then are we as well?
even a RX 6400 throttles to oblivion just watching videos at that speed. I had one in the past
I think our situation is different
Just a coincidence
probably
You're running dual AMD cards now?
I was going to suggest running display driver only, but since you're running both AMD cards, I don't think it would be worth it
If AMD (Rendering) + NVIDIA (LSFG), I suggest display driver only for AMD, full install on the nvidia card
the thing is this also happened with my 1070 too
Also, this was the driver I was running before selling my 7900 xtx https://www.amd.com/en/resources/support-articles/release-notes/RN-PRO-WIN-24-Q4.html
used to run it too
display only + full nvidia driver (without the app or geforce experience)
What I find specially weird, is that on fullscreen it works amazingly. Which shows it's something messing with the driver in borderless, since in exclusive fullscreen, windows leaves all the rendering to the GPU driver instead of using DWM
not for me tbh
ingame my game runs at like 95 fps with 6600xt as display regardless of borderless or fullscreen
didn't you say your FPS skyrocketed in exclusive?
but its 115fps with 7900xt
it did but that was in the pause screen
it sorta helped
its so odd
when its unpaused its pretty much the same as borderless
it seems like something is bottlenecking my cpu or something
My issue wasn't dropping base fps from the main gpu though, just unstable LSFG performance on the LSFG card with headroom.
But I'm still trying to think about your issue though
I'm not sure at all
rip
I'm gonna save up for an am5 system at some point but I'm not sure if that will fix it
I don't recommend it
I would've stayed AM4 if my 7 year old Motherboard didn't go poof
I loved that socket so much. Insanely fast boot times, solid performance, cheap. AM5 is finnicky, specially with RAM. You can reduce boot times, but it's still pretty slow compared to am4
14.1s Bios boot time, on my am4 system it boot in 6-7s. IF you disable memory context restore, it boots in almost a minute with EXPO enabled...
💀
The thing with memory context restore is that it skips memory training and if your ram isn't stable, you're going to have a pretty bad time, since it can cause a lot of issues. If it's stable, it's a must have... taking a minute to reboot a system is a big yikes
It's the Achilies heel of AM5 😓
ddr5 ram is insane too from what I've heard
huge performance difference depending on it
even more ram dependent than am4
Well, am4 in the beginning was like that too. DDR5 when it released was horrible, uber high latency, very expensive. Nowadays it's pretty good.
am5 has its 3rd gen now right? I'd expect it to get better soonish then
second
very early then
my issue is so stupid
timespy perfectly fine scores, hell its even better
then in games and when using lsfg it all breaks down
it makes 0 sense
I'm getting freezes and have to restart my PC anyone experiencing that?
My config is well... not great, intel 11800h and a dGPU RX 480
Trying to use the iGPU with the dGPU
It worked once and and now I don't know why it's not working
Maybe something with your PSU? maybe something's overheating?
Monitoring the stats while using it is the best way to find out
Freezes could be unstable RAM. I've started having the same, randomly, and saw that the tight timings I was running on my previous CPU aren't fully stable on this one. Dialed back the timings and it now runs stable.
Specially since you mentioned using the iGPU, which relies on system RAM to operate
The cable finally came. Have to head out to do some errands. Did an extremely fast test and it's the same result as the rx6400. Cyberpunk went from 84 fps to 50 and then doubled.
oh no
To clarify now it's a 3080ti with an rx6600. Also have a 7800x3d
does it look like a cpu bottleneck?
I didn't have time to check in depth. Will test more in depth in a couple of hours.
you know at this point I think I may have a motherboard issue here
dual gpu with 7900xt was still somewhat off I think
stutters were worse
I'm out of ideas too
not really sure what part of my system is the issue
Just to make sure, you followed all the recommendations i wrote here, right? #1328337747399938138 message
How come? Since others are having the same issue
its not exactly the same issue
but I don't know why there's such a discrepancy in performance for my system compared to others
so i have tested games on same and sepperate SSD. both times the rx goes to 97% and my 3080 goes to 50%-60%
do you lose performance even without lsfg?
gimme min. didnt feel like it but i will check
it isn't clear on some games but very clear on others
but when I use lsfg it doesn't matter
fps gets heavily reduced
cuberpunk looks fine before LSFG. but it says my RX660 is maxed even before LSFG is on
that seems odd?
that happened to my 1070 but not my 6600xt
it might be pushing the max frames it can send to the other gpu?
might be pcie
whats your setup again?
on a tuf gaming B650MB-plus
dont know where i see that?
my rx is 100%usage with LSFG off, that makes no sense to me
could be maxing out pcie or something
your 3080ti is doing x16 at pcie 1.1
usually power saving feature
so its fine there
pcie 4 at x4 is the same as me running gen 3 x8
not sure if that is the bottleneck or not
shouldn't be tbh
so i have NVME on my MB thoose wouldnt occopie some of the same lanes in the pcie right? it would show?
it would show
but it doesn't do anything for your case
@clear sundial @rain hull thoughts?
just so im on the same page. you are speculating in pcie bottleneck?
Welcome to dual GPU
its normal?
because right now desktop window manager is eating my GPU for some reason
Problems that make no sense are normal 
yep
then we have people like ravenger who have no issues
