No performance increase from SLI GTX 1070?

[Joel_64]

So I recently purchased a gtx 1070 FE to replace my r9 390x. I immediately liked the upgrade but I was still wanting better performance so I bought a second gtx 1070 FE to SLI. Only problem is I saw zero performance increase from the second card. If anything I scored lower on Heaven and 3dmark for benchmarks. Before anyone asks, yes I set it up correctly throught the Nvidia control panel, both cards are on the most current drivers and I bought the best bridge Nvidia had to offer for SLI. I'm just wondering if its a PSU issue or a bottlenecking issue now. Here's my setup:

FX 8350 overclocked to 4.2 GHZ
Nvidia GTX 1070 FE X2
Nvidia HB SLI Bridge
DDR3 G-skill ram 2133
MSI 970 gaming MOBO
Corsair H110 AIO CPU cooler
Rosewill 850w bronze PSU
Hitachi 2tb HD
Samsung evo 512GB SSD

 

[Isokolon]

your 8350 is bottlenecking heavily, it's bottlenecking even one 1070

 

[rcald2000]

+Joel_64 Besides the benchmark, have you attempted playing a few game titles with your SLI setup? Check the CPU utilization? Is it 80%+? If so, check out your CPU temperature to make sure that it's not thermal throttling. Also look at the utilization on both your GPUs.

 

[sizzling]

Sli carries an overhead, I expect your single 1070 was slightly bottlenecked by your CPU and adding sli gave your CPU the overhead of supporting it and hence your slight reduction in performance.

 

[rcald2000]

I saw a synthetic benchmark that showed the FX-8350 with even performance to the i5-6500. I'm not sure whether the benchmark is accurate to a real-world gaming scenario. I'm also not sure sure if it's counterpart, i5-6500, would bottleneck two GTX 1070's. My previous questions are phrased in a way to determine if the bottleneck exists.

Question: what resolution are you playing games in? My personal experience is that CPU bottlenecks are somewhat alleviated when the gaming resolution is increased.

 

[sizzling]

I saw a synthetic benchmark that showed the FX-8350 with even performance to the i5-6500. I'm not sure whether the benchmark is accurate to a real-world gaming scenario. I'm also not sure sure if it's counterpart, i5-6500, would bottleneck two GTX 970's. My previous questions are phrased in a way to determine if the bottleneck exists.

Question: what resolution are you playing games in? My personal experience is that CPU bottlenecks are somewhat alleviated when the gaming resolution is increased.

Every game is different but generally the FX8350 is beaten by a current i3. Here is just one example http://www.techspot.com/review/917-far-cry-4-benchmarks/page5.html


Even in BF1 which can use all 8 threads the FX8570 here only comes level with a 6100T http://www.techspot.com/review/1267-battlefield-1-benchmarks/page4.html

 

[bystander]

Those benchmarks he tested are really light on the CPU usage. I wouldn't expect it to be so heavily bottlenecked in the Unegine Heaven benchmark. If you used the 3dmark Time Spy or Firestrike benchmarks, those shouldn't be bottlenecked either. However, some of their other ones may.

Try the heaven benchmark with higher settings. Use 3dmark Time Spy or Firestrike if you did one of their others.

And when you use SLI, it is important to note that you must be in Fullscreen mode. SLI/CF does not work in Windowed or Fullscreen Windowed modes.

 

[SgtScream]

I saw a synthetic benchmark that showed the FX-8350 with even performance to the i5-6500. I'm not sure whether the benchmark is accurate to a real-world gaming scenario. I'm also not sure sure if it's counterpart, i5-6500, would bottleneck two GTX 970's. My previous questions are phrased in a way to determine if the bottleneck exists.

Question: what resolution are you playing games in? My personal experience is that CPU bottlenecks are somewhat alleviated when the gaming resolution is increased.

Every game is different but generally the FX8350 is beaten by a current i3. Here is just one example http://www.techspot.com/review/917-far-cry-4-benchmarks/page5.html


Even in BF1 which can use all 8 threads the FX8570 here only comes level with a 6100T http://www.techspot.com/review/1267-battlefield-1-benchmarks/page4.html

I agree with this. Return both of the gtx 1070's and just get a gtx 1060 or an rx 480 (Or just buy a new system). Your essentially paying all of this money for your graphics card to not run at 100% capability (Pissing your hard earned money away).

 

[Joel_64]

+Joel_64 Besides the benchmark, have you attempted playing a few game titles with your SLI setup? Check the CPU utilization? Is it 80%+? If so, check out your CPU temperature to make sure that it's not thermal throttling. Also look at the utilization on both your GPUs.

Yeah I tried several titles. CPU utilization never went over 69% and temperatures are solid. Now that you mention it though, my second cards utilization is super low on the logs. Almost like its not even actually doing anything, just sitting at an idle.

 

[Joel_64]

I saw a synthetic benchmark that showed the FX-8350 with even performance to the i5-6500. I'm not sure whether the benchmark is accurate to a real-world gaming scenario. I'm also not sure sure if it's counterpart, i5-6500, would bottleneck two GTX 1070's. My previous questions are phrased in a way to determine if the bottleneck exists.

Question: what resolution are you playing games in? My personal experience is that CPU bottlenecks are somewhat alleviated when the gaming resolution is increased.

I'm running games at 5760x1080, I have a triple monitor setup.

 

[Joel_64]

Those benchmarks he tested are really light on the CPU usage. I wouldn't expect it to be so heavily bottlenecked in the Unegine Heaven benchmark. If you used the 3dmark Time Spy or Firestrike benchmarks, those shouldn't be bottlenecked either. However, some of their other ones may.

Try the heaven benchmark with higher settings. Use 3dmark Time Spy or Firestrike if you did one of their others.

And when you use SLI, it is important to note that you must be in Fullscreen mode. SLI/CF does not work in Windowed or Fullscreen Windowed modes.

See I did Timespy and Firestrike and ran benchmarks with the single and sli setup. Both score were pretty disappointing for what they should be realistically scoring. My single GTX 1070 was doing like 7 fps on the physics test of Firestrike.....

 

[Joel_64]

I saw a synthetic benchmark that showed the FX-8350 with even performance to the i5-6500. I'm not sure whether the benchmark is accurate to a real-world gaming scenario. I'm also not sure sure if it's counterpart, i5-6500, would bottleneck two GTX 970's. My previous questions are phrased in a way to determine if the bottleneck exists.

Question: what resolution are you playing games in? My personal experience is that CPU bottlenecks are somewhat alleviated when the gaming resolution is increased.

Every game is different but generally the FX8350 is beaten by a current i3. Here is just one example http://www.techspot.com/review/917-far-cry-4-benchmarks/page5.html


Even in BF1 which can use all 8 threads the FX8570 here only comes level with a 6100T http://www.techspot.com/review/1267-battlefield-1-benchmarks/page4.html

I agree with this. Return both of the gtx 1070's and just get a gtx 1060 or an rx 480 (Or just buy a new system). Your essentially paying all of this money for your graphics card to not run at 100% capability (Pissing your hard earned money away).

I definitely won't return the 2 GTX 1070's. If anything I'll upgrade the processor, mobo and RAM if thats the route I need to take. Which it sounds like I'm going to. My build is something I'm slowly upgrading, I'm aware the 8350 is a somewhat outdated processor and I would be better off with a more current intel chip like a 6700k, I just didn't think I would be getting THIS bad of performance. At the time it was a good choice paired with my R9 390x.

 

[rcald2000]

i7-6700K is old tech now. Buy the i7-7700K / Z270 chipset motherboard / DDR4 3200 RAM.

 

[bystander]

Those benchmarks he tested are really light on the CPU usage. I wouldn't expect it to be so heavily bottlenecked in the Unegine Heaven benchmark. If you used the 3dmark Time Spy or Firestrike benchmarks, those shouldn't be bottlenecked either. However, some of their other ones may.

Try the heaven benchmark with higher settings. Use 3dmark Time Spy or Firestrike if you did one of their others.

And when you use SLI, it is important to note that you must be in Fullscreen mode. SLI/CF does not work in Windowed or Fullscreen Windowed modes.

See I did Timespy and Firestrike and ran benchmarks with the single and sli setup. Both score were pretty disappointing for what they should be realistically scoring. My single GTX 1070 was doing like 7 fps on the physics test of Firestrike.....

First off, forget about the FPS you see in Timespy or firestrike. They aren't supposed to be playable FPS. They are strictly to test how your system performs.

2nd, the physics tests has nothing to do with your GTX 1070. That is a CPU test.

When you look at the results from a single GPU and SLI, did the Graphics score improve greatly? Can you link both results for us to view. The results given on their site will compare them to people with similar setups.

 

[Joel_64]

Here's the score results for my SLI setup
http://www.3dmark.com/3dm/17202045?

Here's the results for my single GPU
http://www.3dmark.com/3dm/17471276?

The single GPU score was higher then the SLI score

 

[bystander]

Your CPU is clearly a problem. Even the benchmark said your CPU is a problem. Compared to people with similar systems, your score is WAY lower.

Monitor your CPU clock rate and temps. I'm expecting you are thermal throttling.

 

[Joel_64]

Your CPU is clearly a problem. Even the benchmark said your CPU is a problem. Compared to people with similar systems, your score is WAY lower.

Monitor your CPU clock rate and temps. I'm expecting you are thermal throttling.

After running more tests I've come to the conclusion that its definitely my processor. The CPU usage was hitting 92% on DOOM. The only weird thing was that my CPU's highest temperature was 41 degrees C under full load which is pretty damn good. It is water cooled though. The fact that I'm getting better results from a single card says enough as well.

 

[bystander]

Your CPU is clearly a problem. Even the benchmark said your CPU is a problem. Compared to people with similar systems, your score is WAY lower.

Monitor your CPU clock rate and temps. I'm expecting you are thermal throttling.

After running more tests I've come to the conclusion that its definitely my processor. The CPU usage was hitting 92% on DOOM. The only weird thing was that my CPU's highest temperature was 41 degrees C under full load which is pretty damn good. It is water cooled though. The fact that I'm getting better results from a single card says enough as well.

What are the clocks running at during the test or DOOM?

 

[Isokolon]

Things to consider:

The Pascal chips scale pretty bad. The 60%-75% performance increase older series experienced with Sli is no more.
Many games don't support Sli or don't support it properly. Thus the 2nd card idling in some games.
You cpu can run at 50% and be at maximum capacity as the game can only tax one thread per module if it doesn't support all the threads the FX provided. It's similar to an i7. So while 4 threads are at 100%, the others don't even get used or just a little bit, resulting in 50% or 60% load when in fact your CPU is under full load.
Before upgrading the CPU I'd wait for AMDs Ryzen. Might be a disappointment, might be the next great thing. Should make Intel drop prices a little bit in any case.

 

[bystander]

Things to consider:

The Pascal chips scale pretty bad. The 60%-75% performance increase older series experienced with Sli is no more.
Many games don't support Sli or don't support it properly. Thus the 2nd card idling in some games.
You cpu can run at 50% and be at maximum capacity as the game can only tax one thread per module if it doesn't support all the threads the FX provided. It's similar to an i7. So while 4 threads are at 100%, the others don't even get used or just a little bit, resulting in 50% or 60% load when in fact your CPU is under full load.
Before upgrading the CPU I'd wait for AMDs Ryzen. Might be a disappointment, might be the next great thing. Should make Intel drop prices a little bit in any case.

The scaling is quite good, if you look at 4K benchmarks. The problem is they get bottlenecked at 1080p quite easily, so you do have to be mindful of that. This may be that games are becoming less demanding in comparison to the cards, and there is also a fair bit of games that simply do not support SLI that have been released recently. This could be a trend with DX12/Vulkan, but when it works, and not bottlenecked, the scaling is good still.

The problem isn't so much that is CPU is bottlenecking (it is), but that it is not currently performing like it should. Look at the links to his 3dmark benchmarks. They are WAY behind others with the same system. Something is wrong with his system.