Terrible GTX 980 SLi framerates.

remilafo · April 2015

** in reponse to the OP, i haven't read all the commentary**
Your computer isn't the issue, it is the game and you are not going to be able to fix it.

Albeit your CPU is a big weak for that GPU setup, regardless it is still more than enough for this game.

You issue will only happen in crowded AND open world areas, like busy towns.

Your FPS is probably very nice in non-open world areas like dungeons and/or with nobody else around.

Your issue is the way the game server handles update requests from players, it's not a problem per se and there might not be a solution either at least with this game engine.

Your fps drops because the server is waiting on position/action updates from you and other players around you and then it must broadcast that to you and the players around you. In short internet latency is your problem.

BUT BUT my ping is perfect blah blah!! yeah maybe but even 1ms ping is still much longer than the communication speed between your cpu/gpu futhuremore isn't just about you, it's also about the people around you and their ping. AND how fast the server can process all that and send you an update.

so, in short no amount of tweaking or driver updates is going to fix your issue. It's not you.

Good luck

SapphireThunder · April 2015

remilafo wrote: »

Albeit your CPU is a big weak for that GPU setup, regardless it is still more than enough for this game.

You issue will only happen in crowded AND open world areas, like busy towns.

Your FPS is probably very nice in non-open world areas like dungeons and/or with nobody else around.

Your issue is the way the game server handles update requests from players, it's not a problem per se and there might not be a solution either at least with this game engine.

Your fps drops because the server is waiting on position/action updates from you and other players around you and then it must broadcast that to you and the players around you. In short internet latency is your problem.

1: i5 4690k is nowhere near too weak for dual-SLI 980. It can easily handle them. provided, it can hit the limit before gpu does.
2: the latency doesn't exactly affect fps, it affects how you see some things. For example, if you have high latency, you will see monsters or other people warping/be somewhere else/moving weirdly.

I'm sorry to say, but you have very little actual knowledge about what you said.
You do have the right idea, and thus not completely wrong. But you are blaming data transfer for image processing. Which just isn't true.
FPS and latency are not connected to each other.

remilafo · April 2015

SapphireThunder wrote: »

1: i5 4690k is nowhere near too weak for dual-SLI 980. It can easily handle them. provided, it can hit the limit before gpu does.
2: the latency doesn't exactly affect fps, it affects how you see some things. For example, if you have high latency, you will see monsters or other people warping/be somewhere else/moving weirdly.

I'm sorry to say, but you have very little actual knowledge about what you said.
You do have the right idea, and thus not completely wrong. But you are blaming data transfer for image processing. Which just isn't true.
FPS and latency are not connected to each other.

1. It is actually but not for this game (ESO), see benchmark results for Dragon age Inquisition for SLI and Crossfire users. A google search can help you with that.

2. You are making a broad assumption, although you aren't incorrect your statement lacks specificity. Latency can most certainly affect frame rate, this is especially true for full online games such as MMO's

3. way to be super rude and insult me, jerk. I am infact a linux sysadmin and know quite well whats going on here but i don't expect anyone to care or know that.

Before leaving yet another jerky remark why don't you test for yourself.
Same game,
same area,
same hardware specs,
different time of day and population

ex. record your FPS in a specific location during busy time (lots of other players) and do the same recording during non-busy time (aka ghost-town) .. Compare FPS difference.. draw your own conclusions.

BUT BUT!! more people means more textures OFCOUSE my FPS will be lower!

No ***. So instead examine your GPU load instead, more textures should imply a bigger gpu load right? RIGHT?
If your gpu load is infact lower during busy time but yet higher during none busy time that should get your brain thinking. Enjoy..

TL:DR .. it is latency related, maybe not your latency specifically but the ESO servers load and what it has to send to you; and most importantly what the game engine in your eso client uses. Your hardware can only update your screen on the info the eso servers provides you.

PS*, unless you want to download a third party app to add a software frame buffer to artificially force your FPS to always be 60fps. but thats a story for another time and probably not supported by the EULA.

SapphireThunder · April 2015

remilafo wrote: »

ex. record your FPS in a specific location during busy time (lots of other players) and do the same recording during non-busy time (aka ghost-town) .. Compare FPS difference.. draw your own conclusions.

BUT BUT!! more people means more textures OFCOUSE my FPS will be lower!

No ***. So instead examine your GPU load instead, more textures should imply a bigger gpu load right? RIGHT?
If your gpu load is infact lower during busy time but yet higher during none busy time that should get your brain thinking. Enjoy..

TL:DR .. it is latency related, maybe not your latency specifically but the ESO servers load and what it has to send to you. Your hardware can only update your screen on the info the eso servers provides you.

PS*, unless you want to download a third party app to add a software frame buffer to artificially force your FPS to always be 60fps. but thats a story for another time and probably not supported by the EULA.

Latency plays no part in there. You can have 10ms latency and still get the same FPS drop with same amount of people in same place.
Or you can have 300ms latency and still have as high fps as ever.
GPU load also does not correlate quite that way. Sure it's normal for it to lower during gaming when the amount of stuff in screen decreases, but it usually still jumps back up. UNLESS cpu is bottlenecking it.
In which case the cpu is hitting computing limit before gpu does. (as you know)

If the amount of people on map lessens --> fps increases because gpu can use it's power without being bottlenecked by cpu
You can do this artificially by increasing resolution. It can improve the fps in crowded places by taking usage away from cpu and making it more demanding for gpu. The average fps often will lessen, but fps becomes more stable.

This is why when I changed resolution from 1920x1080 to 3860x2140, my fps in town increased. Meanwhile fps in elsewhere, with no people, decreased.
in town before: 30-35fps
after: 35-45

elsewhere before: 50-70
after: 40-55

This was all during same day, same time. Latency 90-100ms
Later that day, I had latency of 190-210ms and the fps was still untouched.

remilafo · April 2015

SapphireThunder wrote: »

1. Latency plays no part in there. You can have 10ms latency and still get the same FPS drop with same amount of people in same place. Or you can have 300ms latency and still have as high fps as ever.

2. GPU load also does not correlate quite that way. Sure it's normal for it to lower during gaming when the amount of stuff in screen decreases, but it usually still jumps back up. UNLESS cpu is bottlenecking it.
In which case the cpu is hitting computing limit before gpu does. (as you know)

3. If the amount of people on map lessens --> fps increases because gpu can use it's power without being bottlenecked by cpu
You can do this artificially by increasing resolution. It can improve the fps in crowded places by taking usage away from cpu and making it more demanding for gpu. The average fps often will lessen, but fps becomes more stable.

4. This is why when I changed resolution from 1920x1080 to 3860x2140, my fps in town increased. Meanwhile fps in elsewhere, with no people, decreased.
in town before: 30-35fps
after: 35-45

elsewhere before: 50-70
after: 40-55

This was all during same day, same time. Latency 90-100ms
Later that day, I had latency of 190-210ms and the fps was still untouched.

- added numbers to paragraphs for discussion purposes.

You made a good point and actually did a test, cool. im liking this discussion; can i make a few "givens" that we both seem to agree upon so that we don't bring it up again in our dialogue.

- a 4690K is NOT a bottleneck in this game, can we agree to remove this variable from discussion.
- also dual gtx 980's in terms of raw compute power should be more than enough to run this game at 100fps@1080p and over 60fps@4k.. so the gpu's arn't a bottleneck either. hence remove from discussion.
*agreed with these givens?

The Problem: Game performance is much lower than expected despite overpowered hardware, supporting evidence is that GPU_load (reported by afterburner or gpu-z probably) is not 100%, hardly 50%. Furthermore there seems to be drastic changes in FPS negatively affecting gaming experience again despite overpowered hardware which should be maxing out this game without breaking a sweat.

The cause:*
- my blame is on the client/server relationship and how it's handled, which i will explain in a moment.
- Sapphirethunder what are you blaming?

my thoughts: I think the game client is waiting on the game server for updates on dynamic objects like other players and only renders the changes once it has gotten said update. For static objects like walls and environment once the game client has the info it renders the object and doesn't check for moment by moment updates. Support for my thoughts is the dramatically lower FPS in busy areas hence alot more dynamic objects.

Your thoughts: ????????

Regarding our earlier debate seems you are really focusing on the client side of things but i ask you to consider how latency is behaving on the server side of things and speculate how that would behave/ripple in effect to a client.

PS*edit: i just re-read the OP post and i found this interesting. "I have tried disabling one of the cards, and performance actually goes UP", i really can't explain that based on my experience with this game. I personally run a quad-fire amd 7970 setup and i experience the low 25-45% gpu_load *** as well but in general i do get 4x scaling with my gpu's. Even with one gpu the load for me tops out at about 50-66%
Still if the OP did their homework correctly i in no way think their system is to blame for this problem.

remilafo

SapphireThunder · April 2015

1: Yes, we agree about those.

2: The problem for low fps is not the hardware or network. The problem is engine and how the engine/server handles the data.

3: as you noticed, the multi-gpu scaling in this game can be bad. It is only good on mid-end gpus like 760 or R9 270/270x.
770/780/960 are on the edge. 970 and 980 seem to be on the worse side. My guess is they are just too new and too powerful for the engine.

4: few other reasons are that the client is 32-bit, while not directly affecting the performance, the amount of RAM it can support and handle is limited compared to 64-bit. This can limit some high end hardware. Also DirectX11 itself due to how it uses cpu cores. Majority of utilization is core 1, with core 2 having just around half of that, core 3 having 1/3rd of core 1 and core 4 having just 1/6th. core 5 and 6 having just 1-2%. 7 and 8 cores being completely unused. (unless corrected with game engine)

addition: you are actually losing overall performance on your Quad-Crossfire. Since the cards are only 4x speed each. with 2 you would have each card at 8x speed. A single card would be 16x speed. While computing power increases, the speed decreases so it will actually be worse for some applications. This is one of those. Generally 2 and 3 cards are the sweet spot. 4 cards has very small performance increase.

EDIT: How latency affects the game, is how your position in the map, the monster AI, loot, npc AI, skill usage and similar things are handled.
With very high latency, you may find your attacks not landing to enemies because they either were actually outside the range or you may find there to be lag on attacks. You may also see rubberbanding where your character or other characters suddenly teleports backwards/forwards. Or something shows up to be stuck somewhere/not moving.

Also: after changing to 4K resolution, my gpu utilization jumped from 60-70% to 95%+. Because now it was more demanding on the gpu as I stated before.

remilafo · April 2015

SapphireThunder wrote: »

2: The problem for low fps is not the hardware or network. The problem is engine and how the engine/server handles the data.

exactly, i don't see how you can stand by your point that "it's not latency" and claim the above statement. Seems you are only looking/thinking skin deep as it were on the client side.

SapphireThunder wrote: »

addition: you are actually losing overall performance on your Quad-Crossfire. Since the cards are only 4x speed each. with 2 you would have each card at 8x speed. A single card would be 16x speed. While computing power increases, the speed decreases so it will actually be worse for some applications. This is one of those. Generally 2 and 3 cards are the sweet spot. 4 cards has very small performance increase.

if it was Q4 2011 maybe.... you are out of date. this is long corrected. I get 400% (more like 380%) scaling in all my games
even the ones that DON'T have a crossfire profile, you just need to override in the driver control panel and enable the default one that comes with CCC (amd driver control panel). Although experience will vary from one to another, with the x79 or x99 chipset from intel you can get 40X lanes of PCI goodness and quad-fire/sli it up no problems.

SapphireThunder wrote: »

EDIT: How latency affects the game, is how your position in the map, the monster AI, loot, npc AI, skill usage and similar things are handled.
With very high latency, you may find your attacks not landing to enemies because they either were actually outside the range or you may find there to be lag on attacks. You may also see rubberbanding where your character or other characters suddenly teleports backwards/forwards. Or something shows up to be stuck somewhere/not moving.

ALthough this is true you are still just thinking about client side. Anyways im done with this convo, im getting bored. Peace!

SapphireThunder · April 2015

remilafo wrote: »

exactly, i don't see how you can stand by your point that "it's not latency" and claim the above statement. Seems you are only looking/thinking skin deep as it were on the client side.

Latency != how server handles data
Latency is the time it takes data to transfer between client and server.
How server handles the data, is all about the server's hardware and architecture along with their server engine.

remilafo wrote: »

if it was Q4 2011 maybe.... you are out of date. this is long corrected. I get 400% (more like 380%) scaling in all my games
even the ones that DON'T have a crossfire profile, you just need to override in the driver control panel and enable the default one that comes with CCC (amd driver control panel). Although experience will vary from one to another, with the x79 or x99 chipset from intel you can get 40X lanes of PCI goodness and quad-fire/sli it up no problems.

You can't fix the PCI-E lane amount except with giving more PCI-E lanes on cpu and motherboard. You will gain the power, but you are losing the speed.
You will STILL have the computing power of 4 cards. Just the data transfer speed between them and cpu is lowered due to low amount of PCI-E lanes.
Many games do not care about this so you will still get the the full performance. But there are some games and there will be games which will be affected by it.

wilsonirayb16_ESO · April 2015

I use an i7 2600k at 4.6ghz with two R9 290s in Crossfire on stock timings, on a 1440p 120hz monitor (96hz effective for ESO).

In cities and populated areas(or rather I should simply say in heavily populated areas due to other players) I average (no science here, just observation) around 45-60. When I leave and head to an open world area, with less player density (but still some players), I get about 80-100, with the majority of the time on the 100 side of things. And of course in dungeons/instances I hold 100 until the screen gets crazy with spells then it drops a little.

All in all, I think given how CPU dependent the game is (an obvious draw back of generic MMO coding methods...) the game performs OK.

The biggest performance factor(s) seems to be Draw Distance, shadows and then some others such as particle levels (for battle purposes). If you can live with the loss of visuals, using 75 or so setting for Draw Distance, will help quite a bit.

Still, my gripe is the fluctuation in frames during open world activities. There are times where it will go from 100 down to 85 and back a bit. That trips up the frame latency and causes a bit of jidders in the motion smoothness. I'm close to setting the monitor refresh rate to 80 and locking the frames there.

LaughingJack · April 2015

well i just got my hands on, and installed, 8GB DDR3 1066 (dual ch.) and now all my stuttering and horribleness is gone. game runs smooth-as at around 50'ish fps. in general, around combat and busy areas, i get down to 40-50, and in dungeons i can go up to 90. i'm only running a HD4870(512MB) after all.

@Sapphire Thunder: yes i fully agree it would be nice to have a solid 60fps, and that would be my average. i guess my point is that you dont need to run this game at 4k@100+Hz for it to look fantastic. FACT: most people cannot discern frames once the framerate goes above about 75Hz, so going any higher than 80-90 is pointless - you are churning out frames you will never see, unless of course you need it for the bragging rights, which is still pointless. it is Not absolutely essential in the global sense, it may be for you because you cannot bear to even look at a game unless it looks like its running on a supercomputer, but thats your choice. personally i go for immersion over pretties for my gaming, and immersion does not equal pretties.

as for Reaction times? well this game only requires reactions times that are fairly slow compared to FPshooters

what i find really funny about all this is the fact people have spent good money on SLI/CS setups with mutiple cards when one really good card would suffice. always seems to be the case - those pushing the bleeding edge end up having the most issues.

LaughingJack · April 2015

wilsonirayb16_ESO wrote: »

I use an i7 2600k at 4.6ghz with two R9 290s in Crossfire on stock timings, on a 1440p 120hz monitor (96hz effective for ESO).

In cities and populated areas(or rather I should simply say in heavily populated areas due to other players) I average (no science here, just observation) around 45-60. When I leave and head to an open world area, with less player density (but still some players), I get about 80-100, with the majority of the time on the 100 side of things. And of course in dungeons/instances I hold 100 until the screen gets crazy with spells then it drops a little.

All in all, I think given how CPU dependent the game is (an obvious draw back of generic MMO coding methods...) the game performs OK.

The biggest performance factor(s) seems to be Draw Distance, shadows and then some others such as particle levels (for battle purposes). If you can live with the loss of visuals, using 75 or so setting for Draw Distance, will help quite a bit.

Still, my gripe is the fluctuation in frames during open world activities. There are times where it will go from 100 down to 85 and back a bit. That trips up the frame latency and causes a bit of jidders in the motion smoothness. I'm close to setting the monitor refresh rate to 80 and locking the frames there.

could not agree more. i get around the same performance as you +-10fps or so.

brandon02852cub18_ESO · April 2015

SLI issues are nothing new. I do not see
nVidia making much progress on proper driver support for SLI, so I stick with the highest performing single GPU I can get my hands on.

Which is why I bought a TITAN X. Overclocked at 1450MHz, custom cooler, custom bios...overclocked 3770K....it rips this game to shreds.

I am playing at 3840x2160 (DSR) on a 1080p monitor...max settings...config maxed for quality (-3 skip mipmap load level).

The lowest fps i get is 35 in the most crowded places. I am usually between 60 and 120 in the overworld and 90-120 in dungeons.

Screw SLI

Egonieser · April 2015

OP, you get the same framerates as I do with my specs:

CPU: i7 4790k
GPU: MSI GTX 970 Twin RazR (CPU/GPU watercooled)
RAM: 16GB 1600Mhz Crucial HyperX
SSD: Samsung EVO 840 250GB

I run it with near max (shadows/water on medium) and DSR on 1080p, and FPS varies between 30-80, but never really dips below 30 even in Cyrodiil. This leads me to believe that the game doesn't utilize GPU that much, but is more CPU intensive as MMO's tend to be, which is probably holding you back.
SLI setups are prone to excel in graphics intensive games with few players like Battlefield etc/Single Player games where you only render the world/visuals and/or a handful of players. MMO is a different beast entirely, players and objects/interactables are mainly rendered by CPU rather than GPU, you could even have SLI Titans on board, it wouldn't make the FPS any better if you don't have a beastly CPU to compensate.

Besides, it is already well established by now that the game is poorly optimized as it is. Their goal was to have at least stable minimum 30 FPS (which is the console standard, and the game is coming out on consoles soon) so don't think they will bother that much to opimize it further for those with high-end PC's. Their version of bare minimum is met, and that's (i believe) is where they stop investing anything else in it.

I do think it's frustrating as I too would enjoy better FPS (I can achieve this in other games, yet not ESO). It is a good looking game, but it's not THAT good to get such low FPS like we're getting now. Fair enough, it's far from unplayable, but steady 60+ FPS is better than random jumping between 30-80, sometimes without even a particular reason.