[SOLVED] RTX 3080 alarmingly low performance

[BoondockSaint080]

So I recently paired my rtx 3080 with my r5 3600 (no this is not a bottenecking situation) and fired up rainbow six siege (using vulkan, as it has the edge in performance after numerous optimizations). In 1440p maxed out, it maintains 165fps at around 75-95% gpu usage. This is alarming to me as other benchmarks see it at 260fps avg using similar setups (albeit faster cpu's) but the gpu usage is evident of it capping off at around 200fps. With this information, I quickly turned to TAA to put my worries to rest. At TAA 4X @60% render scaling (just over 4k), the game manages a measly 100-120fps at full gpu usage, when it is supposed to easily reach 145-155fps. This is a near 30% decrease in peformance over benchmarks and i'm completely confused. I use the same settings as the benchmarker did and this is entirely gpu dependent, as i can see it maintain significantly higher fps averages through 1440p native with no antialiasing. Any help is appreciated

Edit 1: parts list: https://pcpartpicker.com/list/sVr7ht

Edit 2: userbenchmark: https://www.userbenchmark.com/UserRun/34192878

 

[Phaaze88]

The more info you can provide, the easier it'll be for others to help you narrow down the source of the issue.
For now, start with the PC's specs - using PCpartpicker's list should simplify things: https://pcpartpicker.com/list/

 

[alceryes]

Posted benchmarks are usually the top .1%. They are not the norm - they are the outlier.
No one's going to post 'meh' benchmarks. Benchers/reviewers tweak, bench, tweak, bench, tweak, bench, etc., until they get a score that they deem 'worthy' and only then does it gets posted. These are usually people who work with technology for a living and know the ins and outs of settings and components. Many of them also know how to 'cheat' the system by running the benchmarks with unrealistic settings. They change everything from CPU priority, to benchmark config changes, to doing borderless/alt tab crap. In short, don't worry about hitting published benchmark numbers.

As @Phaaze88 says, full system specs are needed. A quick UserBenchmark run with a link to the results would also be helpful.

 

[geofelt]

Submitted benchmarks will often be from maximally tweaked setups.
That will include overclocked processors.
Those with dog processors will be silent.
r5-3600 is not the fastest of processor chips and it does not overclock well.

Be happy with a "measly" 100-120 fps at 1440P
Or, upgrade your processor.
3080 is designed for 4k gaming where graphics power is the most important factor.

 

[thepersonwithaface45]

So I recently paired my rtx 3080 with my r5 3600 (no this is not a bottenecking situation

using similar setups (albeit faster cpu's)

hmmm

 

[BoondockSaint080]

hmmm

So I feel as though this response is beating a horse I preemptively shot point blank at the start, as I clearly stated that the cpu is not the issue and is completely capable of reaching 165+hz. This is also ironic as the quotes you used were specifically written to disparage someone from generating such a conclusion. I'm wondering why my gpu, at a very high utilization percentage at 4k (where the gpu is the clear bottleneck) is heavily underperformed compared to its peers.

 

[thepersonwithaface45]

So I feel as though this response is beating a horse I preemptively shot point blank at the start, as I clearly stated that the cpu is not the issue and is completely capable of reaching 165+hz. This is also ironic as the quotes you used were specifically written to disparage someone from generating such a conclusion. I'm wondering why my gpu, at a very high utilization percentage at 4k (where the gpu is the clear bottleneck) is heavily underperformed compared to its peers.

All I know is that there's diminishing returns at any resolution below 4k for the 3080/90. Not sure if your monitor is 2k or 4k, but I think from your post it's 2k and you ran it in 4k and got your test bench marks from there.
Wonder what your fps would look like if you plugged in a 4k display

 

[BoondockSaint080]

Submitted benchmarks will often be from maximally tweaked setups.
That will include overclocked processors.
Those with dog processors will be silent.
r5-3600 is not the fastest of processor chips and it does not overclock well.

Be happy with a "measly" 100-120 fps at 1440P
Or, upgrade your processor.
3080 is designed for 4k gaming where graphics power is the most important factor.

I get the impression you didn't fully read my post, as I clearly wrote that it was rendering at basically 4k. Also, the people I'm referring to aren't professional overclockers, but rather regular people who clocked everything stock. Though I wasn't referring to 1440p with my measly performance claims, I would like to add some of those numbers that are noticeably lower than other people with the same setups.

In warzone, at 1440p high settings, on an r5 3600 and rtx 3080 stock, I've seen people get 170fps with an impressive 160fps 1% low. My hardware manages to get the same average of 100-120fps. Although this IS a cpu bottleneck, as indicated by the low gpu usage I experience, I'm still left wondering why other stock 3600's are getting 50% more frames. I will be editing in a pc part picker list as per request of another responder so feel free to use that. I appreciate you taking the time to respond.

 

[BoondockSaint080]

Posted benchmarks are usually the top .1%. They are not the norm - they are the outlier.
No one's going to post 'meh' benchmarks. Benchers/reviewers tweak, bench, tweak, bench, tweak, bench, etc., until they get a score that they deem 'worthy' and only then does it gets posted. These are usually people who work with technology for a living and know the ins and outs of settings and components. Many of them also know how to 'cheat' the system by running the benchmarks with unrealistic settings. They change everything from CPU priority, to benchmark config changes, to doing borderless/alt tab crap. In short, don't worry about hitting published benchmark numbers.

As @Phaaze88 says, full system specs are needed. A quick UserBenchmark run with a link to the results would also be helpful.

I will be editing the pcpartpicker list in when I get a chance after work. I appreciate the response.

 

[logainofhades]

With a 3080, at 1440p, you can become CPU bound. This was evident in some of the day 1 reviews. Simply put, that is really meant for 4k gaming card. You also forgot to add your motherboard, to your listing.

 

[BoondockSaint080]

With a 3080, at 1440p, you can become CPU bound. This was evident in some of the day 1 reviews. Simply put, that is really meant for 4k gaming card. You also forgot to add your motherboard, to your listing.

I realized I forgot the motherboard as soon as I locked my phone haha I'll add it shortly. As for the resolution, I clearly stated that the render resolution was 4k, the gpu usage maxed out, and the framerate clearly being gpu bound. Please read the whole post as the word 1440p being in the post doesn't warrant total disregard of the potential issue.

 

[thepersonwithaface45]

I realized I forgot the motherboard as soon as I locked my phone haha I'll add it shortly. As for the resolution, I clearly stated that the render resolution was 4k, the gpu usage maxed out, and the framerate clearly being gpu bound. Please read the whole post as the word 1440p being in the post doesn't warrant total disregard of the potential issue.

But is render resolution the same as a native 4k input thru display? I don't know myself, I just feel like they wouldn't work the same.

 

[Phaaze88]

Edit 1: parts list: https://pcpartpicker.com/list/H9rdCz

I see 3 different storage drives. Can you tell us what each is for?

Not aimed at you specifically, OP:

View: https://www.youtube.com/watch?v=mrzqoeQVg4k


Ryzen 3600 is far from a terrible pairing with a 3080. The gap between it and the best only gets narrower at the higher resolutions.
Then again, benchmarks like those are going to be a sort of 'perfect world' scenario.

 

[jmrnilsson]

Hi,

I’d say like most else here, at very high frame rates you become very CPU-bound even if there is GPU-head room. I’d for sure crank up the resolution, if possible.

With that said this is what I’d do to the most out of GPU and minimize latencies.

1. Set future-frame rendering to 1 (or possibly 2). Make sure ultra low latency mode is disabled in Nvidia driver.

2. Enable XMP and find a stable yet higher than stock RAM frequency to run at. Also, CL18 is some pretty loose timings, make sure to tighten it or run the RAM at pretty high frequency.

3. Up resolution or detail in the game. High FPS is going make you very frequency and IPC dependent. I’d consider the 10600K or the upcoming 5800X if you have 240 or 360 hertz monitor.

 

[BoondockSaint080]

Hi,

I’d say like most else here, at very high frame rates you become very CPU-bound even if there is GPU-head room. I’d for sure crank up the resolution, if possible.

With that said this is what I’d do to the most out of GPU and minimize latencies.

1. Set future-frame rendering to 1 (or possibly 2). Make sure ultra low latency mode is disabled in Nvidia driver.

2. Enable XMP and find a stable yet higher than stock RAM frequency to run at. Also, CL18 is some pretty loose timings, make sure to tighten it or run the RAM at pretty high frequency.

3. Up resolution or detail in the game. High FPS is going make you very frequency and IPC dependent. I’d consider the 10600K or the upcoming 5800X if you have 240 or 360 hertz monitor.

Ram is being run at 3466mhz, the highest this motherboard will support. I'm currently waiting on eve to drop their spectrum line of monitors and will be getting the 4k version, which is why the 4k numbers I'm seeing using taa are so scary. It's supposed to be getting 176fps avg but I'm getting 115 probably.

 

[BoondockSaint080]

I see 3 different storage drives. Can you tell us what each is for?

Not aimed at you specifically, OP:

View: https://www.youtube.com/watch?v=mrzqoeQVg4k


Ryzen 3600 is far from a terrible pairing with a 3080. The gap between it and the best only gets narrower at the higher resolutions.
Then again, benchmarks like those are going to be a sort of 'perfect world' scenario.

So this benchmark actually indicates higher feamerates than what I'm referring to. 176hz is the bare minimum I should be getting but alas I'm getting a nearly 40% fps cut

Edit: the sata ssd is for windows, the 970 evo plus for games, and wd 1tb is for anything I deem unworthy of ssd space.

 

[BoondockSaint080]

But is render resolution the same as a native 4k input thru display? I don't know myself, I just feel like they wouldn't work the same.

Although I'm far from an expert, I've been lead to believe that that's the case

 

[Phaaze88]

the sata ssd is for windows

Oof.
120GBs is too small for today's Windows. At least 250GBs.

Could you run a userbenchmark and post a link to the results of that?

 

[BoondockSaint080]

Oof.
120GBs is too small for today's Windows. At least 250GBs.

Could you run a userbenchmark and post a link to the results of that?

so a couple of notes regarding userbenchmark before you actual look at them. it's actually been something i've been looking into strenuously in the last 24 hours, as well as unigine superposition benchmarks. My gpu is the pny variant for reference and does not breach the 80c threshold (well aware that that's not an achievement but i want to get thermal throttling out of the question).
In userbenchmark, it consistently stays below the 30th percentile I noticed that the gpu doesn't reach the 1920 boost frequency, but rather a cold 1710 for boost. if i crank that up, along with the power draw, i will get halfway decent results. i've been able to get a 210 boost clock on the gpu so far and it's stable, which isn't too much of a flex given it's merely the advertised boost clock that they took away in a panic to avoid the public shame of the capacitor incident.
The card does fine both thermally and stability wise at 1920mhz, reaching a 14900 in superposition. unfortunately I cannot verify that this will affect what userbenchmark has to say about my card, as i use afterburner, which relies on rivatuner, which crashes the graphics test.

TLDR: what you're seeing in userbenchmark is the stock settings for the 3080, as afterburner will not allow it to run through. I'm definitely underwhelmed by this. https://www.userbenchmark.com/UserRun/34192878

 

[alceryes]

Your Windows drive is filled past what is deemed 'optimal performance' for SSD drives and wasn't even able to complete the benchmark. SSD drives slow down when they get over 65-75% full. You should also run mfg. drive tools to check the overall health of all your drives.

Your 3080 is definitely underperforming.
I see the term 'thermal thresholds/throttling' thrown around a lot (not just by you). What many fail to realize is that just because a particular GPU (or CPU, for that matter) stays below a certain temperature it doesn't mean that it should be able to go faster. Thermal thresholds are just one facet in the performance jewel. The Ampere core in your RTX 3080 may just not be able go faster with 100% stability. Graphics card designers/manufacturers are also starting to get very fancy with their advertised speeds vs. what is seen in real life. They use complex algorithms to determine whether the card will even be allowed to reach a certain MHz speed. My Vega 64 is a perfect example of this.

One final note. NVIDIA says the RTX 3080 requires a 750W power supply (read 750W 'performance' power supply, here). It's not necessarily the combination of all your parts but also the wattage your PSU can deliver on the aux GPU rails. It could be that your GPU isn't getting enough juice right when it needs it the most.

 

[Phaaze88]

@BoondockSaint080

https://imgur.com/DdMo6v9

View: https://imgur.com/DdMo6v9


To finish the message in orange: the OS affects basically EVERYTHING.

 

[alceryes]

Hahaha! What you are pointing to saying, "Ignore this BS," is exactly what you want to pay attention to!
The score to the right, the actual benchmark score, is just how well this component did in the Userbenchmark. This is an absolute score so, naturally, if you have one of the fastest GPUs available it will still perform well even if it's just limping along. The "BS" score is the relative performance against other RTX 3080s running Userbenchmark. This is the EXACT score we want to focus on.

Also, please post the actual URL to the Userbenchmark score. There's much more info on the actual page.

 

[Phaaze88]

@alceryes
That 'performing below/above X percentile' is part of what makes UBM such a garbage site to use - it does have it's uses though.
Instead of using a stock Nvidia FE RTX 3080 as a comparison baseline, it compares to everyone else's regardless of condition:
-Drivers, whether good or bad
-Boost clocks are temperature sensitive, so operating temps do matter
-Vbios power limit hacks/betas
-Paring with good/bad power supplies
-Overclocking
-Silicon lottery
Too many variables - it just skews results.
Since the OP's score is within 10% of the average of all those possibilities, that's still acceptable performance.
If one is trying to min-max... then don't use UBM at all.


Regardless, the problem appears to be the OS being on the pretty much full 120GB SSD. That's what sticks out the most.

 

[alceryes]

@Phaaze88, I agree it's not perfect but that's when you, the user, needs to roll up your sleeves and do some work. Go through the hundreds/thousands of comparisons and narrow down as best you can with like hardware (and clock frequencies, etc.).
It just takes a little bit of work, that's all.

 

[Phaaze88]