Reference RX 5700 XT Testing

Shlouski

VIP Member
My new card has just arrived today, so lets see how the new RX 5700 XT performs.

In all of the test I will be using wattman to record gpu parameters and unigine superposition to benchmark the card, as I'm using an older cpu I will be running all the tests in 4k to reduce the possibility of any cpu bottleneck. As wattman isn't cooperating with superposition when it comes to averages I will be using 2 loops of heaven benchmark to record averages.


First I will start of with stock out of the box performance, no hardware or software changes have been made.


stock wattman.png
stock super.png

Stock Results:

Idle Temp: 43c
Max Temp: 93c
Max Boost clock: 1983
Average Clock: 1887


Now I will set an aggressive fan curve to see how temperatures are affecting performance:

fan curve wattman.png
fan curve super.png


Fan curve results:

Idle Temp: 33c
Max Temp: 78c
Max Boost clock: 2004
Average Clock: 1902

Now I will move on to the washer mod.
 

Shlouski

VIP Member
I feel like a 3770K would throttle these maybe?

I don't see it bottlenecking at 4k in unigine, remember this cpu has the same single core performance as beers ryzen 2700x in your cpu-z benchmark thread, so as long as unigine doesn't need more than 8 threads I think it will be ok. Once I have pushed this card to the max I can swap it into my i7 8700k machine and run the benchmarks again.
 

Shlouski

VIP Member
I did some testing to find out what is the minimum voltage need for the card to run stable, stock the card was running 1202mv and the card becomes unstable around 1060mv, currently I have it set at 1080mv which seems stable so far.

Lowered voltage benchmark with a less aggressive fan curve:

reduced voltage fan curve.png


reduced voltage fan curve super.png


Results:

Idle Temp: 33c
Max Temp: 77c
Max Boost clock: 2026
Average Clock: 2010

5% Performance increase in superposition over stock.

Next I will test an aggressive fan curve to see if even lower temperatures impact performance.
 
Last edited:

Shlouski

VIP Member
Aggressive fan curve and lowered voltages results:

reduced voltage agg fan curve.png

reduced voltage agg fan curve super.png


Results:

Idle Temp: 32c
Max Temp: 74c
Max Boost clock: 2019
Average Clock: 2016

I have found an issue with superposition which causes the minimum framerates to jump around from run to run, usually around +/- 5 fps even while using identical gpu settings. I have observed that superposition stutters in one of the scenes in the exact place every time and this stutter is recorded as the minimum framerate, so I believe its more likely be a software problem than the gpu itself. As I can do multiple runs until I get a score with a higher minimum fps, I also don't want to be here all day so I'm not doing that, but I will try a couple of different benchmarks after I have finished overclocking to make sure the gpu is stable and not suffering from dips.
 

Shlouski

VIP Member
+50% Power Limit results, also the ambient temperature is 3c higher than stock test:

50+ power wattman.png

50+ power super.png

Results:

Idle Temp: 34c
Max Temp: 80c
Max Boost clock: 2047
Average Clock: 2040

6.6% Performance increase in superposition over stock.
 

Darren

Moderator
Staff member
I have a feeling you're gonna see a noticeable bump with a newer CPU, definitely interested on those results.
 

Shlouski

VIP Member
I have a feeling you're gonna see a noticeable bump with a newer CPU, definitely interested on those results.

I think I will have time tomorrow to try it on my i7 8700k build, I also will try clocking this 3770k all the way back and we will see how they compare.
 

Shlouski

VIP Member
My 3770k was running all core 4.5ghz for the previous tests posted above, increasing the frequency up to 4.75ghz had no effect on performance, but I will still try the 8700k system at some point.

Now I have increased the stock core frequency from a max boost of 2090mhz to 2150mhz, which is maximum wattman will allow without soft modding. This change alone will not give higher boost frequencies, to achieve higher frequencies the power limit has to be raised and sufficient voltage supplied to keep the card stable at increased frequencies.


Max core boost of 2150mhz, power limit set at +10% and the voltage increase to 1204mv:

+10pl 1204mv watt.png

+10pl 1204mv super.png


Results:

Idle Temp: 33c
Max Temp: 87c
Max Boost clock: 2115
Average Clock: 2106

7.7% Performance increase in superposition over stock.


This is pretty much the max core clocks I can get without increasing the voltage over stock.

Not bad so far, and I haven't even touched the memory yet.
 

Shlouski

VIP Member
After playing with the voltages:

NICE2.png

I'm pretty much at this cards limit when it comes to core clock, I'm happy to leave it like this and move on to memory overclocking, 9% increase so far which is great :).
 

OmniDyne

Active Member
Definitely neat seeing those results. Great job on the results, reporting and mod. Was it as easy as Gamers Nexus made it out to be?
 

Shlouski

VIP Member
Definitely neat seeing those results. Great job on the results, reporting and mod. Was it as easy as Gamers Nexus made it out to be?

Disassembly of the card was easy, the only awkward part was disconnecting and reconnecting the cables on each side, one for the fan and the other for the LED, the cables are short so you can only separate the PCB and cooler just enough to get your fingers in there and you are doing it one handed if you want to support the card in the other hand.

I found a post by someone modding a different AMD GPU, I thought he had the good idea which was to put extra spacers under the metal clamp towards closer to the center of the gpu, hopefully allowing for more pressure and contact with the middle of the die.

untitled.png

I also tried this mod but I didn't really notice any difference in temperatures, also I didn't like how much extra pressure the bracket was now under and I worried that it may lose its shape and bend over time causing it to lose pressure and contact. I ended up bringing both ideas together and putting washers under the screws and spacers under the clamp, this in my opinion has helped to increase the pressure more evenly over the die and also stopped the deformation of the clamp, this also actually lowered the temperatures 1 or 2 degrees more :).

I do miss the old ways of overclocking, increasing core, memory and shader clock speeds until it becomes unstable and then dialing it back a bit and it was great when more gpu's started to allow control over vcore so you could push for the little bit extra.

I have overclocked many gpu's over the years, my most recent overclocking experience has been with the 980 ti, 1070 ti, 1080 and the RX 470, and the 5700 xt acts differently to all of them. Of course we all know that years ago there was no turbo boost, you just entered the clock speed you wanted and the gpu for the most part was stuck at that speed, and from my experience even when gpu's started to come out with gpu boost they would still boost to the frequency you set if temperatures and power limits permitted it, but the 5700 xt is different. Usually my overclocking goes as follows: Max out the power limit and check temps, find the minimum vcore needed to run stock settings, increase core clocks until unstable, increase vcore to stabilize, if temperatures permit continue to increase clocks and vcore until the card hits its max frequency or until the point that vcore increases only gain greatly diminished core clocks to the point that its not worth continuing, and then move on to memory.

With my 5700 xt you can set the core clocks to whatever you want, but it won't go much over 2100 no matter the settings. At stock the core runs in the 1880 to 1950 range, increasing the power limit to 50% brings it up to around 2030, then increasing the core from stock 2090 to 2150 gets you to around 2100. At this point the temperatures are maxing out in the high 80's, increasing the fan curve can reduce the temperatures by about 10 degrees, but this has no impact and clocks stay the same. Now the only way to raise the core speed is by reducing the voltage as low as possible and this will only get you an extra 10 to 15mhz and the power limit can be reduced all the way down to 10% before it effects the clock speeds. So now I have a card at a reduced voltage from stock, at a power limit of only +10%, running a max core temperature in the mid 70's, a max memory temperature in the high 60's and a max vrm temperature in the high 50's. This is where my 5700 xt differs from other cards, any other card I've had with voltage, temperature and power limit headroom would be trying (stable or unstable) to boost to a higher core clocks. Now this suggests to me I am not reaching the cards maximum overclock and I'm more likely to be hitting a limit built in by AMD, I guess if I want more I will need to do a SoftPowerPlay mod.
 

Shlouski

VIP Member
At stock the memory runs at 875mhz and I have found that my memory becomes unstable at anything over 910mhz, this is my highest result with the memory at 900mhz.

best mem.png
 

Shlouski

VIP Member
I have put the 5700 xt in my i7 8700k build and its like its a different card. Overclocks that would not work before are now stable, clock speeds are much more stable and the card is willing to draw much more power. Before I removed the 1080 out the 8700k system I saved its overclock settings and benchmarked it, then I loaded the same overclock settings and benched it on my 3770k system and it worked perfectly, it even got very close results in superposition at 4k. I do not know why this is but there seems to be some sort of conflict, in the past I have experienced problems between motherboards and graphics cards which were released many years apart, it is possible this card does not work well with the z77 chipset.

5700 xt running at stock in the 8700k system:

stock wattman.png

stock super.png

As you can see there is very little performance difference between the 3770k and 8700k at stock. The 3770k system supposedly hit a higher boost clock, but I have experience lots of anomalous readings in wattman when it comes to boost clock speeds, so I am sure they are inaccurate.

Now the i7 8700k with the same aggressive fan curve setting and stock settings as the i7 3770k, and the 5700 xt achieving a world record boost clock of 7336mhz stable ;)

stock agg fan curve.png

stock agg fan curve super.png


On the i7 8700k system I was able to use settings that were far from stable on the i7 3770k system, I was able to use higher power limits with lower voltages and the card was able to pull a more wattage, also I was able to get higher memory clocks.

best yet.png

best yet super.png
 

Shlouski

VIP Member
I saw Timmy Joe benchmarking and overclocking his anniversary edition 5700 XT:


I decided to download Time Spy and compare his results to my 5700 XT, even with higher cpu score my card still manages to beats his in graphics score.

timespy vs tj.png
 
Top