late to this party, but anyways. I started to oc my Fury X again, after I went back to stock when ReLive came live and now I want to share some OC results with you.
I use OPs custom BIOS and changed MVDDC back to stock 1.300V. My card at stock runs VID on DPM7 with 1.187V which I changed to 1.231V. I used the already set +25mV offset but decreased VID of DPM0-DPM6 accordingly to match stock values again. So DPM0-6 is stock and DPM7 is around 1.256V under usage.
Core Clock on 1140mhz at 1.256V runs stable so far, running multiple stress tests. OC in percent = +8%
HBM can run stock with -25mV but for 545mhz, 1.300V is necessary Going higher results in memory error. OC in percent = +8%
GPU gets around 48-50°C, VDDC Temp get as high as 65°C and MVDDC is around 49°C. Logged with HWiNFO64.
I use a pretty aggressive cooling curve though, running 1,900rpm most of the time, which is an acceptable noise level for me.
And before I forget, my HW is: Windows 10, AMD ReLive 17.3.1, Radeon Fury X, Xeon E3-1231v3, [email protected]
of DDR3 ram
So I ran a few benches to see the performance difference between stock, core oc, HBM oc and both overclocked. I ran each test 3 times and used the average of those for the diagrams.Valley Benchmark (Click to show)
So first of, Valley benchmark with Ultra Preset. Min and Max FPS is not that important. 0,1/1/99-percentile would be much better, but I wanted to keep it simple. Let's talk about avg. FPS then.
We can see, that HBM oc only already gives us a boost of around +12%. Core OC gives us a boost of +15%. OCing both, results in literally the same, even a bit less. Even though the card maintained their max clock speeds, some kind of throttling seems to take place.3D Mark (Click to show)
Next up, 3D Mark. In Fire Strikes High Performance test, HBM oc gives +2%, core and both oced give around +8% performance boost. Again Core gives you more performance per % of oc, but oc on both tends to be even a tidbit worse than ocing only the core.
Heading over to 3D Marks Time Spy benchmark, we see a very interesting result. HBM oc now gives us -1.7% performance in overall score, on graphics it's withing margin of error. Core oc however increases our performance by around +5%. Again, OC on botch results the same.Deus Ex: Mankind Divided (Click to show)
On Deus Ex: Mankind Divided (DX:MD) I used the built in benchmark on DX12, combined with a custom preset (Ultra with tweaks) I set up for my game. We can now see, that HBM oc does nothing and even the 8% core clock increase gives us only 4,8% FPS increase. I was curious whether the increased HBM clock really does nothing, so I also measured the frametimes using OCAT/PresentMon with those results:Frame Times over 30sec in DX:MD (Click to show)
While definitely not the best frametimes, we see that HBM oc performs exactly as stock. Only core clock increase results in a better avg. result but oc on both give us the shortest spikes and overall best timings. Which is interesting, considering the reported FPS is the same.Conclusion:
While not the most comprehensive benchmark, I get a similar pattern throughout each test and I try to extend that when I got time. HBM oc does literally nothing at all, with the exception of Valley benchmark. Increasing the core clock results in minor performance increases, but you're very limited on how far you can push the card. A typical combination of both, seems to run the card into some kind of throttling (The clockspeeds stayed at max though). We can often see a slight decreased performance, even though it should be always slightly higher at minimum. But temps are okay, so are the used voltages. As of now, it seems that you gain nothing from HBM OC, but it may vary in different applications. I'll try to look further into this.
Oh and yeah, here are some AIDA64 screen I took. One is full stock (1050/500), the other full oc (1140/545):AIDA64 (Click to show)