Lower than expected GPU performance

[KRKATANAKID]

So my roommate and I both have identical EVGA GTX 980 ti's. We both are supposedly GPU bottlenecked (usage constantly shows 100% in game). He is running a Ryzen 5 3600, and I have an overclocked X5680. We both play at 1080p, and he easily pulls 50%-100% more frames in games like Apex compared to me. Both of our CPU's are completely bored.

 

We used Passmark PerformanceTest 9.0 to do some testing. He scored almost a 15,000 on 3D, while I scored only a 7900! With the same card! I used DDU and removed the drivers, reinstalled, and re-ran. I got my score up to 10,000. Progress, but I am still looking for the other 5,000 points. I noticed the main difference is that his DX12 subscore is a 75, while mine is a 45, even with redone drivers.

 

I'm sure some of you guys will think my CPU is old crap and I need to upgrade it, but I hold the overclocking record for air cooling on hwbot.org. I'm scoring similarly to a 6700K on benchmarks, so I honestly don't see how my CPU (at 20% utilization) could be a bottleneck, especially since GPU usage is hitting 100%.

[Jurrunio]

Because your CPU is too old to support common instruction set AVX.

[SpookyCitrus]

8 minutes ago, KRKATANAKID said:

I'm sure some of you guys will think my CPU is old crap and I need to upgrade it, but I hold the overclocking record for air cooling on hwbot.org. I'm scoring similarly to a 6700K on benchmarks, so I honestly don't see how my CPU (at 20% utilization) could be a bottleneck, especially since GPU usage is hitting 100%.

The Ryzen 5 3600 is significantly better than your CPU, it's on par with the 8700k where as yours is a lot less. Most GPU testing software like passmark or 3Dmark also are affected by CPU performance. Same thing with Cinebench R15's graphics test, you can use the same card same clocks and everything but use a different CPU and get a higher or lower score. Also your friends rig is much more modern depending on his board might have PCIE 4.0, the program is probably on an NVME drive and yours maybe isn't? There are a ton of factors but the main one is his CPU is much better and newer than yours and has improved architecture.

[KRKATANAKID]

The 6700K was not even out when the 980ti was new, and IIRC people were benefitting from SLI with that CPU and 2x 980ti's, so it seems there would still be plenty left on the table. I'm also seeing 100% GPU usage. Is there any way I could have a CPU bottleneck when the GPU is maxed out? 

 

Also this is happening in game as well.

[Vishera]

It's your CPU,while not technically bottlenecking the GPU,

It's old and has poor single core performance,

I don't have a bottleneck in my system but the Ryzen 5 3600 will give me more FPS due to it's single core performance.

[KRKATANAKID]

Continuing with my example of Apex, could it be a CPU bottleneck when no core ever sees more than 40% utilization? This seriously looks like a GPU bottleneck

[Vishera]

Just now, KRKATANAKID said:

Continuing with my example of Apex, could it be a CPU bottleneck when no core ever sees more than 40% utilization? This seriously looks like a GPU bottleneck

It's not technically a bottleneck but your CPU does limit your GPU.

[KRKATANAKID]

According to passmark I am closer to an 8700k than an 8700k is to an R5 3600. Just leads me to be confused as to how any of those could be limiting a 4 year old GPU. Sorry for wasting everyone's time

[Blatcher2]

21 minutes ago, KRKATANAKID said:

I have an overclocked X5680.

Have you overclocked it. I overclocked my X5675 by almost 50% and have a GTX 1070 which performs just a couple % better in most games.

 

I'll run Passmark rn and let you know what I get

[KRKATANAKID]

Just now, Blatcher2 said:

Have you overclocked it. I overclocked my X5675 by almost 50% and have a GTX 1070 which performs just a couple % better in most games.

I've been running anywhere from 4.4-4.8GHz throughout the entire time I've had this CPU. It's done well for me, and kicks the ass of most CPU's that cost 4x

[Vishera]

Just now, KRKATANAKID said:

According to passmark I am closer to an 8700k than an 8700k is to an R5 3600. Just leads me to be confused as to how any of those could be limiting a 4 year old GPU. Sorry for wasting everyone's time

Do a Cinebench R15 single core or a Geekbench 4 Single core benchmarks,

Compare the scores to other CPUs and you will see how it makes sense.

 

Video games heavily rely on single core performance,that's why the 9900K is the "king of gaming" and not the 3900X.

[Blatcher2]

24 minutes ago, Vishera said:

It's your CPU,while not technically bottlenecking the GPU,

It's old and has poor single core performance

While yes it is a factor, its not why the OP is seeing such a big difference. 

28 minutes ago, Jurrunio said:

Because your CPU is too old to support common instruction set AVX

This is probably why OP has such poor performance in APEX. Most games don't use AVX but APEX does for some reason. Not sure what you can do about that.

 

41 minutes ago, KRKATANAKID said:

He scored almost a 15,000 on 3D, while I scored only a 7900! With the same card! I used DDU and removed the drivers, reinstalled, and re-ran. I got my score up to 10,000.

Alright, so I ran the test using x5675 @4.4GHz and 1070 at 1860MHz. Simply put, your friend's score is most likely a mistake. The average GTX 1070 scores a 11375 and I scored a 12666 (right about the median based on the graph). He should re-run the test. I'm also interested to see what the average 980 ti score is, since I used to have one.

 

For reference my individual scores were: Dx9 182, Dx10 239, Dx11 255, Dx12 61 and GPU compute 6682. 

[Vishera]

2 minutes ago, Blatcher2 said:

Alright, so I ran the test using x5675 @4.4GHz and 1070 at 1860MHz. Simply put, your friend's score is most likely a mistake. The average GTX 1070 scores a 11375 and I scored a 12666 (right about the median based on the graph). He should re-run the test. I'm also interested to see what the average 980 ti score is, since I used to have one.

The 980 Ti has slightly more "horse power" than the 1070 so it should have a better score than the 1070.

[Blatcher2]

19 minutes ago, Vishera said:

The 980 Ti has slightly more "horse power" than the 1070 so it should have a better score than the 1070.

More cores but at a slower clockspeed and different architecture. 

https://babeltechreviews.com/the-gtx-1070-versus-the-gtx-980-ti/3/

[Jurrunio]

59 minutes ago, KRKATANAKID said:

The 6700K was not even out when the 980ti was new, and IIRC people were benefitting from SLI with that CPU and 2x 980ti's, so it seems there would still be plenty left on the table. I'm also seeing 100% GPU usage. Is there any way I could have a CPU bottleneck when the GPU is maxed out? 

 

Also this is happening in game as well.

Not all benchmarks use all available instruction sets in a certain CPU, more often than not they only care about single-precision floating point performance. Your X5650 might be good there, but when new instruction sets are used heavily it falls apart, like trying to make the general American read German. They could with online translation tools, but that takes time.

[KRKATANAKID]

He ran my 980ti in his system, and reran several times. Both cards score over 14,000 in his system. I believe the average for a 980ti is just over 11,000

 

Thank you for benchmarking your 1070 & X5675 combo. Helps me feel less insane for expecting >10,000 out of my rig.

[I AM WOLFII]

49 minutes ago, Blatcher2 said:

While yes it is a factor, its not why the OP is seeing such a big difference. 

This is probably why OP has such poor performance in APEX. Most games don't use AVX but APEX does for some reason. Not sure what you can do about that.

 

Alright, so I ran the test using x5675 @4.4GHz and 1070 at 1860MHz. Simply put, your friend's score is most likely a mistake. The average GTX 1070 scores a 11375 and I scored a 12666 (right about the median based on the graph). He should re-run the test. I'm also interested to see what the average 980 ti score is, since I used to have one.

 

For reference my individual scores were: Dx9 182, Dx10 239, Dx11 255, Dx12 61 and GPU compute 6682. 

 

I am his roommate, and this was my passmark score as of 2 minutes ago. (NO OVERCLOCK)

[KRKATANAKID]

I threw in a 1060 6GB I had laying around and went from a 10,000 to a 7,200. I feel like this sort of confirms a GPU bottleneck/ software issue of some kind?

[Blatcher2]

1 hour ago, KRKATANAKID said:

He ran my 980ti in his system, and reran several times. Both cards score over 14,000 in his system. I believe the average for a 980ti is just over 11,000

 

Thank you for benchmarking your 1070 & X5675 combo. Helps me feel less insane for expecting >10,000 out of my rig.

Literally makes no sense to me why one system score would exceed the average by over 30%. The only test I did better on was GPU compute which happens to be the only off screen test. Sounds like your system is running as expected, just make sure to enable high performance in Nvidia control panel for power management. 

[KRKATANAKID]

I messed with my overclock, turned down my multipliers and turned up my baseclock. I disabled hyperthreading, went through the settings in nvidia control panel, and overclocked my 980ti.

 

Getting my baseclock up brought me to the 10,000 mark. Disabling hyperthreading and the rest got me to 11,128. This is on par with the normal expected for a 980ti. I still have stability testing to do, but I'm gonna call it a win. 

 

Thank you Blatcher2 and I AM WOLFII for your help. To everyone who just dismissed this and told me to buy an entire new PC, thanks for nothing.