Question underperforming 3070

[derben]

Hi all,
I bought my Gigabyte RTX 3070 Aorus Master last year, and has been playing minecraft/roblox on it ever since. Until now when i wanted to try other big games, i have noticed some abnormal fps activities with my gameplay, the gpu usage would be super unstable ranging from 70%-98%, getting me significantly lower fps then i should be getting. i have tried all the possible fixes for this, and have just contacted gigabyte support, but anyone have any ideas how this happened? cheers.

on userbenchmark it shows my gpu as the 13th percentile >>

https://imgur.com/a/i0zxZtR

View: https://imgur.com/a/i0zxZtR


i have an 1080p 240hz monitor so the frame dips/inconsistency frames are more noticeable compared to a 60hz monitor.
on forza horizon 4/5 i am only able to get 40-90 inconsistent fps with stutters and occasional frame dips (ultra settings)
on valorant 250-300fps (max settings)
on minecraft 200-400fps
on roblox 200+ but sometimes dips to 20-30fps and stays like that until a system reboot is made; then persists again.

those fps are ridiculous for my specs, which is:
i7-10700k
6x fans running max speed
850w corsair psu
ddr4 4000 ram
samsung 970 evo plus
Anyone got possible ways to fix this?? gigabyte service team is just.... taking ages to reply, thanks a lot.


**edit: the screenshot posted above was captured 5 minutes after a clean windows reinstallation, the scores have 0 improvements compared to the pre-reinstallation of windows.

 

[Fix_that_Glitch]

Try using ddu uninstaller to erase gpu drivers, link is in my window, then install newest gpu driver and see if this helps. GeForce Game Ready Driver | 517.48 | Windows 11 | NVIDIA You are using two-year-old drivers.

 

[Karadjgne]

the gpu usage would be super unstable ranging from 70%-98%, getting me significantly lower fps then i should be getting.

You misunderstand usage. Usage is the amount of resources used by the gpu to get the frames on the screen, Not the amount of gpu used. That includes time of use for cores and the pauses between use. There's a subtle but definite difference. You don't ever want to see 100% because that means the gpu is using everything, and has no headroom for more.

If your dude was sneaking through occupied territory and at 100% and a Panzer bursts out of a building, that explosion adds a massive amount of data, all the particles flying in every direction, a huge amount of flying objects and colors and smoke. Your fps just went in the toilet because there's literally no more time for the gpu to process all that and get it on screen,

Instead, you'd be far better off with usage closer to 50-60%. My guess with a 3070 at 1080p is you are running 4k DSR possibly with Ray Tracing enabled. Otherwise that card is strong enough that usage in most games at 1080p will be far lower than 70-98%.

Cpu = fps, Gpu = eye candy. The cpu creates any and all fps, the gpu can only limit the amount of fps if details/resolution is too high. That's why pro reviews use the biggest gpus when testing a cpu, so that the gpu is not a limiting factor to fps output.

Post the userbenchmark list, there's a share link on the page, but I suspect that there's exactly nothing wrong at all with your gpu that's not attributed to anything other than detail settings.

Userbenchmark puts your gpu up against 100 other exact same gpus. What it doesn't do is put your gpu up against other users. 13th percentile will include all those other users who have jacked the OC to the moon on their cards, while your card doesn't have that advantage, but for some reason is still better than 87% of the rest of the users, whom undoubtedly are still running at stock settings.

 

[hotaru.hino]

GPU % Utilization is simply an average of some sort of how many SMs in the GPU were doing something during the sample period (usually 1 second). From NVIDIA's system management interface documentation (page 13):

GPU Percent of time over the past sample period during which one or more kernels was executing on the GPU. The sample period may be between 1 second and 1/6 second depending on the product.

So the GPU could be told to transfer one byte over the PCIe and if every SM is doing it, it'll be considered 100% utilization. Otherwise it wouldn't make sense as to why my GPU activity is spiking to 42% to render a 1080p YouTube video. Or that a simple DX9 based application is using 70% of the GPU.

 

[Karadjgne]

Put a nail into drywall with a hammer.
You'll use 100% of the muscles in your hand to hold the hammer.
You'll use 100% of the muscles in your arm to swing the hammer.

You will not use 100% of the strength in you hand to hold the hammer, it does not require a death-grip to hold it.
You will not use 100% of the strength in your arm to swing the hammer, it's only drywall, you aren't trying to drive that nail into the other room.

Same with a gpu. It's going to hit 100% of its boost, clocks, power limits, fps etc. That's the gpu doing all it can with what's given. That's using all its muscles.

Usage is simply the amount of strength, resources, sm's etc required in order to bring that 100%. The only number concerning usage that has any actual relevance is 100. If you hit that, the card has nothing left to use, so any additional instructions by necessity will have to wait on available resources, and fps goes in the toilet.

 

[hotaru.hino]

Put a nail into drywall with a hammer.
You'll use 100% of the muscles in your hand to hold the hammer.
You'll use 100% of the muscles in your arm to swing the hammer.

You will not use 100% of the strength in you hand to hold the hammer, it does not require a death-grip to hold it.
You will not use 100% of the strength in your arm to swing the hammer, it's only drywall, you aren't trying to drive that nail into the other room.

Same with a gpu. It's going to hit 100% of its boost, clocks, power limits, fps etc. That's the gpu doing all it can with what's given. That's using all its muscles.

Usage is simply the amount of strength, resources, sm's etc required in order to bring that 100%. The only number concerning usage that has any actual relevance is 100.

Then explain this

The GPU load is 100% despite:

• Power not being 100% (TBP is rated for 220W)
• Running at the base clock speed
• VRAM usage at 20%
• Performance limit is "utilization", which means "not enough work to boost"

The GPU is clearly not using 100% of its capabilities, and yet it's reporting a GPU Core Load of 100%.

If you hit that, the card has nothing left to use, so any additional instructions by necessity will have to wait on available resources, and fps goes in the toilet.

This is true in the sense that if every SM has something to do, then the next Warp in line is going to have to wait until there's a free SM. But this does not mean each execution unit in an SM is being used.

Utilization is more like you have a team of people and you're scheduling work for them. You can schedule everyone with some job, but that doesn't mean they'll be busy all the time. For instance, if someone is put into a position where they'll be on standby, they're not technically idling: they've been assigned a job, but they're not doing any real work. And you can't assign them another task because they're on standby.

 

[derben]

You misunderstand usage. Usage is the amount of resources used by the gpu to get the frames on the screen, Not the amount of gpu used. That includes time of use for cores and the pauses between use. There's a subtle but definite difference. You don't ever want to see 100% because that means the gpu is using everything, and has no headroom for more.

If your dude was sneaking through occupied territory and at 100% and a Panzer bursts out of a building, that explosion adds a massive amount of data, all the particles flying in every direction, a huge amount of flying objects and colors and smoke. Your fps just went in the toilet because there's literally no more time for the gpu to process all that and get it on screen,

Instead, you'd be far better off with usage closer to 50-60%. My guess with a 3070 at 1080p is you are running 4k DSR possibly with Ray Tracing enabled. Otherwise that card is strong enough that usage in most games at 1080p will be far lower than 70-98%.

Cpu = fps, Gpu = eye candy. The cpu creates any and all fps, the gpu can only limit the amount of fps if details/resolution is too high. That's why pro reviews use the biggest gpus when testing a cpu, so that the gpu is not a limiting factor to fps output.

Post the userbenchmark list, there's a share link on the page, but I suspect that there's exactly nothing wrong at all with your gpu that's not attributed to anything other than detail settings.

Userbenchmark puts your gpu up against 100 other exact same gpus. What it doesn't do is put your gpu up against other users. 13th percentile will include all those other users who have jacked the OC to the moon on their cards, while your card doesn't have that advantage, but for some reason is still better than 87% of the rest of the users, whom undoubtedly are still running at stock settings.

Sorry I didn’t explain well enough; I meant that during games my gpu would run at 70-90% usage while giving me 40 frames, my cpu would be doing fine at 30ish usage, not very sure but using the example u explained, my gpu has no muscles are is super weak. With a 3dmark timespy bench I only got 10k gpu score with the 3070, which is 40% below the average 3070’s performance. https://www.3dmark.com/3dm/80058941?

it is difficult to type everything out on phone, I’ll definitely provide more specifics details when I get home later, anyways thanks for the help.

 

[hotaru.hino]

Sorry I didn’t explain well enough; I meant that during games my gpu would run at 70-90% usage while giving me 40 frames, my cpu would be doing fine at 30ish usage, not very sure but using the example u explained, my gpu has no muscles are is super weak. With a 3dmark timespy bench I only got 10k gpu score with the 3070, which is 40% below the average 3070’s performance. https://www.3dmark.com/3dm/80058941?

Can you get a picture of all those results graphs 3DMark makes after doing a test from the Time Spy run? They'll show stats over time which is very useful.

 

[Karadjgne]

I'm 100% on the phone, I almost never get to use my pc, the few odd times I can, I'm gaming lol.

The GPU is clearly not using 100% of its capabilities, and yet it's reporting a GPU Core Load of 100%.

Exactly. Core load. Not Usage. Memory controller is only 7% etc. What's the Usage? Usage isn't strictly Core Load afaik.

With a 3dmark timespy bench I only got 10k gpu score with the 3070, which is 40% below the average 3070’s performance.

I'm not so sure on that. Look at the competition. Gaming pc, high end pc, premium gaming pc. That's not against other 3070's, that's saying vrs a 2070 upto a 3090. For example.

I'd run userbenchmark and post that link. It at least compares you to others using the same equipment, but allowances must be made for others using the equipment differently, like OC levels etc.

 

[hotaru.hino]

Exactly. Core load. Not Usage. Memory controller is only 7% etc. What's the Usage? Usage isn't strictly Core Load afaik.

You tell me, it's all there.

For what it's worth, HWiNFO's tool tip says GPU Core Load is "Average utilization of the GPU core as reported by the GPU hardware."

 

[derben]

I'm 100% on the phone, I almost never get to use my pc, the few odd times I can, I'm gaming lol.

Exactly. Core load. Not Usage. Memory controller is only 7% etc. What's the Usage? Usage isn't strictly Core Load afaik.

I'm not so sure on that. Look at the competition. Gaming pc, high end pc, premium gaming pc. That's not against other 3070's, that's saying vrs a 2070 upto a 3090. For example.

I'd run userbenchmark and post that link. It at least compares you to others using the same equipment, but allowances must be made for others using the equipment differently, like OC levels etc.

Here you go, UserBenchmark results.

https://www.userbenchmark.com/UserRun/55593426

 

[derben]

Can you get a picture of all those results graphs 3DMark makes after doing a test from the Time Spy run? They'll show stats over time which is very useful.

Here are the 3d mark graphs.

https://imgur.com/a/0zHH7J7

View: https://imgur.com/a/0zHH7J7


Thanks.

 

[hotaru.hino]

Based on the 3D Mark graph, the video card appears to be working just fine. Just the only thing that's odd is the FPS chart, because that looks more like what my RTX 2070 Super gets.

As silly as this sounds another thing on my mind to try is using MSI Afterburner to dial something back a little. Like lowering the power limit to 90% or set something like -100MHz on the clock speeds.

 

[Karadjgne]

Ubm looks fine. 17th % is also vs ppl manuallyl overclocking, so overall isn't bad for stock numbers that can fluctuate depending on temps and the resultant boost hit.

 

[alceryes]

Hi all,
I bought my Gigabyte RTX 3070 Aorus Master last year, and has been playing minecraft/roblox on it ever since. Until now when i wanted to try other big games, i have noticed some abnormal fps activities with my gameplay, the gpu usage would be super unstable ranging from 70%-98%, getting me significantly lower fps then i should be getting. i have tried all the possible fixes for this, and have just contacted gigabyte support, but anyone have any ideas how this happened? cheers.

on userbenchmark it shows my gpu as the 13th percentile >>

https://imgur.com/a/i0zxZtR

View: https://imgur.com/a/i0zxZtR


i have an 1080p 240hz monitor so the frame dips/inconsistency frames are more noticeable compared to a 60hz monitor.
on forza horizon 4/5 i am only able to get 40-90 inconsistent fps with stutters and occasional frame dips (ultra settings)
on valorant 250-300fps (max settings)
on minecraft 200-400fps
on roblox 200+ but sometimes dips to 20-30fps and stays like that until a system reboot is made; then persists again.

those fps are ridiculous for my specs, which is:
i7-10700k
6x fans running max speed
850w corsair psu
ddr4 4000 ram
samsung 970 evo plus
Anyone got possible ways to fix this?? gigabyte service team is just.... taking ages to reply, thanks a lot.


**edit: the screenshot posted above was captured 5 minutes after a clean windows reinstallation, the scores have 0 improvements compared to the pre-reinstallation of windows.

Your GPU does appear to be underperforming in UBM and Time Spy.
Run through Time Spy with HWiNFO64 running in background (sensors only, logging on) and go through the log or upload it somewhere we can have a look.

 

[derben]

Your GPU does appear to be underperforming in UBM and Time Spy.
Run through Time Spy with HWiNFO64 running in background (sensors only, logging on) and go through the log or upload it somewhere we can have a look.

This is the HWiNFO log, please have a look, thanks
https://www.dropbox.com/s/2tnxlzde9jhoybv/hwinfo log.CSV?dl=0

 

[derben]

Based on the 3D Mark graph, the video card appears to be working just fine. Just the only thing that's odd is the FPS chart, because that looks more like what my RTX 2070 Super gets.

As silly as this sounds another thing on my mind to try is using MSI Afterburner to dial something back a little. Like lowering the power limit to 90% or set something like -100MHz on the clock speeds.

i just tried that but still no help ):

 

[alceryes]

This is the HWiNFO log, please have a look, thanks
https://www.dropbox.com/s/2tnxlzde9jhoybv/hwinfo log.CSV?dl=0

Are you doing any overclock, overvolt, or timings adjustments on the CPU, motherboard, memory, or GPU (either manual or automatic)? If yes, reset everything to stock/default and don't load up any GPU frequency software (like afterburner) at startup. Also, reset NVIDIA CP setting to all default too.

What's the exact model of your corsair PSU? Although it's within tolerance level, your just under 12v on the power going to the GPU. Probably not causing an issue though.

Also, get rid of zero-RPM fan setting on the GPU. It can cause extra high spikes to hotspot temps. Start with a very low (quiet) fan setting and ramp it up from there.

I'm not seeing any smoking gun in these numbers unfortunately. Maybe a larger sample will help.

 

[Karadjgne]

Me either. It's pulling roughly the same TimeSpy as a 2070 Super/3060, which is 300points lower than most 3070's, a definite underperformance, but there's nothing pointing to an exact cause.

With its age, putting it in prime covid gpu expense, I'd not be surprised if it was an rma card that's really a 3060 with a 3070 vbios and cooler. Wouldn't be the first shady deal like that with ppl swapping out coolers etc.

 

[derben]

Are you doing any overclock, overvolt, or timings adjustments on the CPU, motherboard, memory, or GPU (either manual or automatic)? If yes, reset everything to stock/default and don't load up any GPU frequency software (like afterburner) at startup. Also, reset NVIDIA CP setting to all default too.

What's the exact model of your corsair PSU? Although it's within tolerance level, your just under 12v on the power going to the GPU. Probably not causing an issue though.

Also, get rid of zero-RPM fan setting on the GPU. It can cause extra high spikes to hotspot temps. Start with a very low (quiet) fan setting and ramp it up from there.

I'm not seeing any smoking gun in these numbers unfortunately. Maybe a larger sample will help.

I have the corsair RM850x (2019) with the stock cables from corsair, and i am having my RAM a little bit overclocked, just lower CAS latency and 166mhz higher than stock. I already tried running stock everything, newest bios but it does seem like a gpu problem. I’m just waiting for gigabyte support to get back to me now, thx for the help!

 

[alceryes]

Yeah. It could just be a faulty GPU.
Good luck!

 

[derben]

Thank you, have a nice day!

Yeah. It could just be a faulty GPU.
Good luck!

Thanks! Have a nice day!