Accidental and intentional methods for enhancing GPU performance

In my previous blog post I described how I fixed my Asus Strix GTX 970 DirectCU II OC graphics card by changing fans and thermal paste. I noticed 7% increase in measured performance. What I left out from the post that I also discovered and observed other methods for increasing the GPU performance.

Changing the GPU fans and thermal paste

Lets recap something from the previous blog post. I changed the fans and thermal paste of an Asus Strix GTX 970 DirectCU II OC (sic) graphics card. I used PassMark PerformanceTest 9.0 for testing. First results with “broken” fans:

So, 9107 points as the baseline for bad fans and bad thermal paste. Then for the changed fans and paste:

Impressive, 9738 points for the change! That’s around 7%.

Using x16 data lane PCIE 3.0 slot

This one I discovered accidentally. I moved the graphics card from one slot to another. Lets take a look at the map of my motherboard (Asrock Z77 Extreme4) PCB about how I did it:

So, what was wrong in “PCIE3” slot in comparison to “PCIE2” ? Lets take a look. Both were listed as PCIE 3.0 x16 slots on the map page, but then..

So, another page told that “PCIE2” was actually faster slot, because it used x16 data lanes instead of x8 the old slot used. This was measeured in PassMark PerformanceTest 9.0. First for the old slot:

Then the result after changing to new slot:

So, there was free performance increase from 9738 to 10025. That’s a 3% increase. Not bad.

Changing the CPU (from i5-3570 to i7-3770k)

Some systems exhibit a bottleneck on the CPU when GPU does not get “fed” data fast enough. Here is an example when benchmarking with Fortnite on “Epic” graphics settings and 1920×1200 resoluton. I used the microsoft gaming bar thingy to see what was happening. First the old Intel i5-3570 processor:

There is clear CPU bottleneck observable. When changing the CPU to i7-3770K the results are much better:

In continuous play the CPU load stayed under 100% which implies that CPU does not become the ultimate bottleneck of the system.

It is a bit philosophical question if the CPU should should affect the GPU 3D test points. Maybe it is impossible to decouple those from each other in a smart, universal way. Anyways, here is the 3D result for old i5-3570 (with case open):

10048 for doing all the possible tricks without CPU change. And then finally, repaired GTX 970 in correct slot and Intel i7-3770K CPU (with case open):

10224. This last bit was worth 1.8% performance increase test-wise, though subjective feel of gaming is that things run now a much much smoother. So maybe the test software tries to anticipate the effect of CPU?

Conclusion

We started from 9107 3D test points and ended up to 10224 with fixed fans & thermal paste, correct PCIE slot, more powerful CPU (and case open). Point increase is 1117, and percentage-wise then increase is some 12%. Nice.

Leave a Reply

Your email address will not be published.