I got my SAPPHIRE NITRO+ Radeon RX 470 100407NT+8GOCL in yesterday and I wanted to try and optimize it's performance and see if I could get near RX 480 performance out of it. I picked this card since it's GPU speed is 1260 MHz and it's memory is 2000 MHz unlike all the other 470s which utilize the slower GPU and memory speed. I also didn't want to exceed the top end voltages, so keep it as near to stock as possible.
First, I ran a lot of the free benchmarks to establish a base line. I used the Unigine Heaven Benchmark, 3DMark Firestrike and Time Spy, FurMark, and CompuBenchCL benchmarks.
So, the basic tool I helped to optimize my card is the Unigine Heaven Banchmark. I started it out and let it go to benchmark to 5 out of 26 before cancelling it, so I could have a base line FPS to work with. I also ran 5 out of the 26 every time I made a change to see the FPS increase or decreases. If there was a positive improvement, I would run the full Heaven benchmark and the other benchmarks as well to verify overall improvements.
Be sure that after you make changes, you click Apply in the upper right hand section of the screen.
So, to start, these are the (Stock) settings when you change everything to manual on the Sapphire card.
This is the end result (Optimized) settings I settled on from tweaking and testing.
Now, in regards to the Radeon 480 cards, I've read that memory clock speed and undervolting have helped out performance, so those were the 2 factors I was looking to start with.
In Wattman, I started out by setting the memory frequency to manual and raising the memory speed to 2025 MHz, then did the 5/26 on the Heaven benchmark. I increased the memory speed to 2050 MHz and ran the test again, but had a significant loss of performance, so I dialed it back to 2025 MHz for now.
Next, on the GPU settings, I set the Voltage to Manual, but left the Frequency to Dynamic. This way I could optimize the voltage. Now, 1050 mV seemed to be a popular setting at State 7, so I set it to that and ran the Heaven benchmark test. This resulted in one of the best performance improvements from tweaking. I tried doing +25 mV and -25 mV and +10 mV and -10 mV from 1050, mV, but I experienced more performance loss either way.
From there, I did -25 mV, +10 mV , and -10mV changes on the other States and tested performance. But, out of all the other States, only State 5 had any real performance change when it was bumped up +15 mV. State 2 was okay with a -4 mV change, so I left it at that, since I was trying to reduce power usage where possible as well.
Now that the voltages had been optimized, it was time to tweak the frequency, so I set the frequency to manual. With State 7, I wasn't going to touch the max 1260 MHz, so I started at State 6. Using +10 MHz and -10 MHz changes, I actually had the best Heaven Benchmark FPS results when it was set to 1215 MHz. But, when I ran CompuBenchCL, the BitCoin Mining MHash was lower, so in testing, the 1185 MHz was the happy medium. And as you can see, State 6 was the only state that showed any good performance increase with adjustments.
I went back and tested the memory speeds again, which is why I dropped it from 2025 MHz to 2020 MHz.
Since I had a Radeon RX 480 I was putting in another build, I had another base line for comparison, so here are my results from tweaking. I also had access to a EVGA GeForce GTX 1060 SC GAMING card, which has a base overclocked speed of 1607 MHz vs the stock 1506 MHz for the GTX 1060.
FireStrike
RX 480 - 10,285
GTX 1060 - 10,762
RX 470 (Stock) - 10,020
RX 470 (Optimized) - 10,293
Time Spy
RX 480 - 3,955
GTX 1060 - 4,058
RX 470 (Stock) - 3,795
RX 470 (Optimized) - 3,842
Unigine Heaven Benchmark - FPS/Score /Min FPS/Max FPS
RX 480 - 104.9 / 2643 / 29.7 / 215.4
GTX 1060 - 132.8 / 3344 / 20.8 / 272.5
RX 470 (Stock) - 103.4 / 2604 / 23.7 / 210.2
RX 470 (Optimized) - 104.3 / 2628 / 29.4 / 217.8
CompuBenchCL - In order they are listed in the benchmark
RX 480 - 119.59 / 36.661 / 2222 / 830.4 / 10.666 / 134.36 / 604.14
GTX 1060 - 159.51 / 27.433 / 1775.4 / 1242 / 8.5806 / 109.63 / 569.91
RX 470 (Stock) - 114.45 / 32.923 / 2205.7 / 791.42 / 9.9736 / 131.09 / 553.56
RX 470 (Optimized) - 114.44 / 32.943 / 2251.4 / 799.32 / 9.9827 / 130.6 / 577.11
Luxmark 3.1
RX 470 (Stock) - 14,062
RX 470 (Optimized) - 14,123
Someone might have a better way of tweaking the performance, but I figured I throw this out to start since no one really explains the numbers they used with Wattman. Using my Kill-A-Watt and checking power draw on FurMark after optimizations, I went up 1 W on power usage. With the memory speed increase and the State 6 increase, 1 W isn't anything significant for the performance I get.
First, I ran a lot of the free benchmarks to establish a base line. I used the Unigine Heaven Benchmark, 3DMark Firestrike and Time Spy, FurMark, and CompuBenchCL benchmarks.
So, the basic tool I helped to optimize my card is the Unigine Heaven Banchmark. I started it out and let it go to benchmark to 5 out of 26 before cancelling it, so I could have a base line FPS to work with. I also ran 5 out of the 26 every time I made a change to see the FPS increase or decreases. If there was a positive improvement, I would run the full Heaven benchmark and the other benchmarks as well to verify overall improvements.
Be sure that after you make changes, you click Apply in the upper right hand section of the screen.
So, to start, these are the (Stock) settings when you change everything to manual on the Sapphire card.
This is the end result (Optimized) settings I settled on from tweaking and testing.
Now, in regards to the Radeon 480 cards, I've read that memory clock speed and undervolting have helped out performance, so those were the 2 factors I was looking to start with.
In Wattman, I started out by setting the memory frequency to manual and raising the memory speed to 2025 MHz, then did the 5/26 on the Heaven benchmark. I increased the memory speed to 2050 MHz and ran the test again, but had a significant loss of performance, so I dialed it back to 2025 MHz for now.
Next, on the GPU settings, I set the Voltage to Manual, but left the Frequency to Dynamic. This way I could optimize the voltage. Now, 1050 mV seemed to be a popular setting at State 7, so I set it to that and ran the Heaven benchmark test. This resulted in one of the best performance improvements from tweaking. I tried doing +25 mV and -25 mV and +10 mV and -10 mV from 1050, mV, but I experienced more performance loss either way.
From there, I did -25 mV, +10 mV , and -10mV changes on the other States and tested performance. But, out of all the other States, only State 5 had any real performance change when it was bumped up +15 mV. State 2 was okay with a -4 mV change, so I left it at that, since I was trying to reduce power usage where possible as well.
Now that the voltages had been optimized, it was time to tweak the frequency, so I set the frequency to manual. With State 7, I wasn't going to touch the max 1260 MHz, so I started at State 6. Using +10 MHz and -10 MHz changes, I actually had the best Heaven Benchmark FPS results when it was set to 1215 MHz. But, when I ran CompuBenchCL, the BitCoin Mining MHash was lower, so in testing, the 1185 MHz was the happy medium. And as you can see, State 6 was the only state that showed any good performance increase with adjustments.
I went back and tested the memory speeds again, which is why I dropped it from 2025 MHz to 2020 MHz.
Since I had a Radeon RX 480 I was putting in another build, I had another base line for comparison, so here are my results from tweaking. I also had access to a EVGA GeForce GTX 1060 SC GAMING card, which has a base overclocked speed of 1607 MHz vs the stock 1506 MHz for the GTX 1060.
FireStrike
RX 480 - 10,285
GTX 1060 - 10,762
RX 470 (Stock) - 10,020
RX 470 (Optimized) - 10,293
Time Spy
RX 480 - 3,955
GTX 1060 - 4,058
RX 470 (Stock) - 3,795
RX 470 (Optimized) - 3,842
Unigine Heaven Benchmark - FPS/Score /Min FPS/Max FPS
RX 480 - 104.9 / 2643 / 29.7 / 215.4
GTX 1060 - 132.8 / 3344 / 20.8 / 272.5
RX 470 (Stock) - 103.4 / 2604 / 23.7 / 210.2
RX 470 (Optimized) - 104.3 / 2628 / 29.4 / 217.8
CompuBenchCL - In order they are listed in the benchmark
RX 480 - 119.59 / 36.661 / 2222 / 830.4 / 10.666 / 134.36 / 604.14
GTX 1060 - 159.51 / 27.433 / 1775.4 / 1242 / 8.5806 / 109.63 / 569.91
RX 470 (Stock) - 114.45 / 32.923 / 2205.7 / 791.42 / 9.9736 / 131.09 / 553.56
RX 470 (Optimized) - 114.44 / 32.943 / 2251.4 / 799.32 / 9.9827 / 130.6 / 577.11
Luxmark 3.1
RX 470 (Stock) - 14,062
RX 470 (Optimized) - 14,123
Someone might have a better way of tweaking the performance, but I figured I throw this out to start since no one really explains the numbers they used with Wattman. Using my Kill-A-Watt and checking power draw on FurMark after optimizations, I went up 1 W on power usage. With the memory speed increase and the State 6 increase, 1 W isn't anything significant for the performance I get.
Last edited:
Facebook
Twitter