Polaris 20 Voltage Curve

[Sabotaged_Enigma]

Yesterday I attempted to overclock my Sapphire RX 580 2048SP Pulse.
Wait... What is a 580 2048SP? Oh, yes, it's a Chinese market only SKU. It can be simply regarded as an RX 570 with better binning and as a result possible higher frequencies. I think AMD's purpose of launching this model was to distinguish this from those have mined, e.g. RX 570, RX 580 and other GCN-based cards.

We all know that GCN cards come with a default voltage that's too high. So, nice tweaking could reduce both power consumption and thermal performance and meanwhile proper overclocking could bring nice framerates.
This model runs at 1306 MHz @ 1150 mV at default, and its VRAM manufactured by Hynix runs at 1750 MHz.
Throughout the process, I overclocked VRAM to 2000 MHz and called it a day. I overclocked and undervolted the core with Afterburner software and tested it with Furmark 1.26 at 1080P preset.

Here are the results:

In daily gaming, I set the card at 1024 MHz @ 900 mV for a good gaming experience.
The frequencies above 1306 MHz are all overclocking, but we can see that none of their voltages is higher than default voltage - 1150 mV. That's why we say GCN cards need undervolting. (I don't know why manufacturers make it like that... )
And at 1411 MHz which some models of RX 580 (2304SP, the original Polaris 20 XTX) are rated at, my card could do it at 1075 mV.

And about voltage I have to say something more. In the HWiNFO64 software, I saw there was always a little voltage drop while running, no matter it was running Furmark or any game. I seemed not to see this happening back a year ago and I've only been using the card for only two years. So when running Furmark at 1411 MHz, I actually gave it 1085 mV to avoid pushing the card off the edge. But as it turned out, its actual voltage was 1.077 V, only a tiny little more than 1075 mV. I can't tell whether that's a hardware problem or monitoring fault.

Undervolting makes a huge difference to power consumption and thermal performance. I increased the TDP limit by 20% (100% = 150 W) but even 1411 MHz only took 158 W in Furmark. If it's running at 1150 mV, it could be about 180 W. And in common games like GTA5, it stayed under 100 W and worked silently.

I hope this could help you tweak your own cards. Thx.

 

[eidairaman1]

Yesterday I attempted to overclock my Sapphire RX 580 2048SP Pulse.
Wait... What is a 580 2048SP? Oh, yes, it's a Chinese market only SKU. It can be simply regarded as an RX 570 with better binning and as a result possible higher frequencies. I think AMD's purpose of launching this model was to distinguish this from those have mined, e.g. RX 570, RX 580 and other GCN-based cards.

We all know that GCN cards come with a default voltage that's too high. So, nice tweaking could reduce both power consumption and thermal performance and meanwhile proper overclocking could bring nice framerates.
This model runs at 1306 MHz @ 1150 mV at default, and its VRAM manufactured by Hynix runs at 1750 MHz.
Throughout the process, I overclocked VRAM to 2000 MHz and called it a day. I overclocked and undervolted the core with Afterburner software and tested it with Furmark 1.26 at 1080P preset.

Here are the results:
View attachment 215464
In daily gaming, I set the card at 1024 MHz @ 900 mV for a good gaming experience.
The frequencies above 1306 MHz are all overclocking, but we can see that none of their voltages is higher than default voltage - 1150 mV. That's why we say GCN cards needs undervolting. (I don't know why manufacturers make it like that... )
And at 1411 MHz which some models of RX 580 (2304SP, the original Polaris 20 XTX) are rated at, my card could do it at 1075 mV.

And about voltage I have to say something more. In the HWiNFO64 software, I saw there was always a little voltage drop while running, no matter it was running Furmark or any game. I seemed not to see this happening back a year ago and I've only been using the card for only two years. So when running Furmark at 1411 MHz, I actually gave it 1085 mV to avoid pushing the card off the edge. But as it turned out, its actual voltage was 1.077 V, only a tiny little more than 1075 mV. I can't tell whether that's a hardware problem or monitoring fault.

Undervolting makes a huge difference to power consumption and thermal performance. I increased the TDP limit by 20% (100% = 150 W) but even 1411 MHz only took 158 W in Furmark. If it's running at 1150 mV, it could be about 180 W. And in common games like GTA5, it never broke 100 W and was silently running.

I hope this could help you tweak your own cards. Thx.

This should be a guide, also you may be able to jump in bios flash threads as an enhancement to make cards run at lower MVs.

I see your style card often. Typically 1002 6FDF 1DA2 E347/E353/E366/387

 

[Sabotaged_Enigma]

I see your style card often.

Yes. Months ago when BTC and ETH were temporarily in "low" positions, I saw that many users here posted some weird problems with their Polaris cards and asked for help on the "proper" BIOS. I think that's when my model has shown up.

Oh my, I wish BTC were never invented...

 

[Taraquin]

Yesterday I attempted to overclock my Sapphire RX 580 2048SP Pulse.
Wait... What is a 580 2048SP? Oh, yes, it's a Chinese market only SKU. It can be simply regarded as an RX 570 with better binning and as a result possible higher frequencies. I think AMD's purpose of launching this model was to distinguish this from those have mined, e.g. RX 570, RX 580 and other GCN-based cards.

We all know that GCN cards come with a default voltage that's too high. So, nice tweaking could reduce both power consumption and thermal performance and meanwhile proper overclocking could bring nice framerates.
This model runs at 1306 MHz @ 1150 mV at default, and its VRAM manufactured by Hynix runs at 1750 MHz.
Throughout the process, I overclocked VRAM to 2000 MHz and called it a day. I overclocked and undervolted the core with Afterburner software and tested it with Furmark 1.26 at 1080P preset.

Here are the results:
View attachment 215464
In daily gaming, I set the card at 1024 MHz @ 900 mV for a good gaming experience.
The frequencies above 1306 MHz are all overclocking, but we can see that none of their voltages is higher than default voltage - 1150 mV. That's why we say GCN cards need undervolting. (I don't know why manufacturers make it like that... )
And at 1411 MHz which some models of RX 580 (2304SP, the original Polaris 20 XTX) are rated at, my card could do it at 1075 mV.

And about voltage I have to say something more. In the HWiNFO64 software, I saw there was always a little voltage drop while running, no matter it was running Furmark or any game. I seemed not to see this happening back a year ago and I've only been using the card for only two years. So when running Furmark at 1411 MHz, I actually gave it 1085 mV to avoid pushing the card off the edge. But as it turned out, its actual voltage was 1.077 V, only a tiny little more than 1075 mV. I can't tell whether that's a hardware problem or monitoring fault.

Undervolting makes a huge difference to power consumption and thermal performance. I increased the TDP limit by 20% (100% = 150 W) but even 1411 MHz only took 158 W in Furmark. If it's running at 1150 mV, it could be about 180 W. And in common games like GTA5, it stayed under 100 W and worked silently.

I hope this could help you tweak your own cards. Thx.

I've had several Polariscards and you can probably run 1200-1250@900mv if binning is okay, 1024MHz should require spmething like 800mv or below. My worst 580 with an asicquality of 6% did 1260@950mv, my best 580 did 1315@950mv with an asicquality of 78%. A early 480 sample with asicsquality if 8% did 1230@1000mv.

Btw voltagefloor on most 570s is 900mv, on 580 950mv and 470/480 1000mv due to vram controller voltage

 

[Sabotaged_Enigma]

1200-1250@900mv if binning is okay
Btw voltagefloor on most 570s is 900mv

Wow, then I'll give it another shot and see where it can go.
And by voltage floor you mean the lowest voltage required at default frequencies?

 

[Taraquin]

Wow, then I'll give it another shot and see where it can go.
And by voltage floor you mean the lowest voltage required at default frequencies?

No matter if you set volt lower it won't work. You can manually lower vram voltage in radeon settings, but it tends to reset, only voltage reset, not clocks etc so it doesn't crash. I ran one of the cards at 1150@850mv, but it reset from time to time to 950mv.

Consumption with 1024@900mv and 1200@900mv is close to the same, but you get 10-15% more performance

 

[Sabotaged_Enigma]

My worst 580 with an asicquality of 6% did 1260@950mv, my best 580 did 1315@950mv with an asicquality of 78%.

What software did you use to test it? In my case, Furmark is great stress and if voltage goes down to where you said, screen output started to tear apart and the whole thing got unstable, lol...

 

[Taraquin]

I used unigine superposition and heaven If they are stable I rarely get instability. Try work your way upwards from 1100, increase 25MHz at 900mv and see where you land

 

[Sabotaged_Enigma]

Update: Memory can be easily overclocked to 2250 MHz, which is also the limit by the BIOS. At this frequency, memory copy speed goes above 200,000 MB/s. But to squeeze more performance, overclocking the core is still the play.

 

[Sabotaged_Enigma]

@Taraquin Yes, bro. My card could pass ~1200 MHz @ 900 mV, but it seems to me there is some frame drops. I don't whether that is what really happened or that I'm too worried about killing my card.
Well then, my curve needs correction. And all of a sudden my stats became invalid...
AND! Furmark 1.26.0 probably went wrong. My card running at ~1000 MHz could have higher scores and framerates than it running at ~1300 MHz and at the same time lower power consumption, which totally confused me. And if I was lucky, the software probably got back normal and gave me the correct results... I don't know whether anybody has encountered this problem.

 

[Taraquin]

Dont use furmark, use a more real life test like superposition

@Taraquin Yes, bro. My card could pass ~1200 MHz @ 900 mV, but it seems to me there is some frame drops. I don't whether that is what really happened or that I'm too worried about killing my card.
Well then, my curve needs correction. And all of a sudden my stats became invalid...
AND! Furmark 1.26.0 probably went wrong. My card running at ~1000 MHz could have higher scores and framerates than it running at ~1300 MHz and at the same time lower power consumption, which totally confused me. And if I was lucky, the software probably got back normal and gave me the correct results... I don't know whether anybody has encountered this problem.

You cant kill your card by undervolting, there is often compensatory mechanisms in drivers to counter furmark, I've seen cards boost way lower at lower temps during furmark, that never happends in superposition. You could try maxing pwr limit while doing 1200@900mv, that can reduce drop in real world applications with very limited increased consumption.

 

[Sabotaged_Enigma]

You could try maxing pwr limit while doing 1200@900mv, that can reduce drop in real world applications with very limited increased consumption.

Thanks, bro. Actually when it runs at 900 mV, power consumption always hovers round 100 W, which is far less than TDP of 150 W.
I'll go try other benchmarks. I don't like Furmark anyway since I think it's too extreme.