TR Forums

Just saw this article at Tech Power Up where one of their readers successfully flashed his RX480 to an RX580 BIOS to achieve 1411MHz. Interested?

Brings back the fond memories of the ATI 9500 -> 9700...although the performance increase there was quite a bit more.

Not particularly interested, I'm afraid.

The RX580 gobbles down 35W more power for a paltry 3% performance boost. It would appear that the 480 is overvolted to the point of throttling under full load without increased power targets, and the RX580 makes this worse. You're much better off spending ten minutes playing with the voltages in Wattman to see how low you can go, since an undervolted 480 or 580 will spend more time at it's max boost clock.

If the default voltage is 1160mv at 1266MHz and you can get all the way down to, say 1025mv before it locks up and resets the driver, then you can probably run 1333MHz at 1100mv or something because there is so much power budget released by reducing the voltage.

My experience in undervolting would agree, but I wanted to point this out for the Xtreme gerbils out there. Sounds like the BIOS flash allows you to add more voltage than the RX480 BIOS would, hence allowing you to hit higher clocks. Looks like it took 1225mV to hit 1411MHz.

Chrispy_ wrote:

Not particularly interested, I'm afraid.

The RX580 gobbles down 35W more power for a paltry 3% performance boost. It would appear that the 480 is overvolted to the point of throttling under full load without increased power targets, and the RX580 makes this worse. You're much better off spending ten minutes playing with the voltages in Wattman to see how low you can go, since an undervolted 480 or 580 will spend more time at it's max boost clock.

If the default voltage is 1160mv at 1266MHz and you can get all the way down to, say 1025mv before it locks up and resets the driver, then you can probably run 1333MHz at 1100mv or something because there is so much power budget released by reducing the voltage.

Hmmm... the RX 580 was attractive to me but I wasn't looking at it this way. Then again, either way I'd be looking at pre-oc'ed non-reference cards e.g. Strix, Gaming X, so does that nullify 480s "advantage"?

Flash your RX480 to a brick

TwistedKestrel wrote:

so does that nullify 480s "advantage"?

Probably off topic for this thread. However, the RX580 can likely be undervolted also, although it's clocks are likely pushed far past the efficient range of the frequency/voltage curve. The RX580 really isn't anything more than an OC'd RX480 but you might get higher clocks with slightly less voltage due to process improvements. As I mentioned, it looks like the RX480 in the TPU article needs 1.225V to hit 1411MHz. The Powercolor RX580 in anandtech's review needed 1.205V to hit 1380MHz but that's without voltage tweaking. I'd bet you could run 1400MHz on an RX580 at less than 1.2V by, as Chrispy put it nicely, "spending ten minutes playing with the voltages in Wattman". My RX480 does 1305MHz happily at 1.03V FWIW. Keep in mind you're playing the silicon lottery in all cases, so YMMV.

I'd recommend taking pricing into consideration for your decision. I've seen higher end 8GB RX480s (better than this one) for $190 or less (after MIR) and they're "old news" now that the RX580 has launched, so the RX480 may see further discounts to clear stock. Keep that in mind when you're looking at RX580 pricing and the ~5% performance improvement associated with it.

PS, before the RX580 launch I lobbied TR to do some investigation into the RX480 vs RX580 frequency/voltage relationships using manual voltage adjustments. That's still something that nobody else has really exclusively covered and it would've given TR quite a bit of publicity. Instead, we have to go elsewhere to see the important aspects of this launch. /vent

DPete27 wrote:

TwistedKestrel wrote:

so does that nullify 480s "advantage"?

PS, before the RX580 launch I lobbied TR to do some investigation into the RX480 vs RX580 frequency/voltage relationships using manual voltage adjustments. That's still something that nobody else has really exclusively covered and it would've given TR quite a bit of publicity. Instead, we have to go elsewhere to see the important aspects of this launch. /vent

Interesting proposition, but for this test to have any merit there would have to be a sample of cards (N=5? 10?) so that variances can be estimated. Picking, by chance, a golden 480 and a bad 580 would lead to confusing conclusions, for example. Unless, of course, you think that the 580 is really, really markedly superior (apples vs oranges).

Intuitively, I don't think there has been major improvement in silicon and the difference, if any, will probably small, hence the need for several cards. No web site has money for this kind of review, but I guess readers could probably derive a very decent estimate of the RX480 volt/freq curve, if we agree (all of us owners of 480/580) to follow a specific protocol.

Chrispy_ wrote:

If the default voltage is 1160mv at 1266MHz and you can get all the way down to, say 1025mv before it locks up and resets the driver, then you can probably run 1333MHz at 1100mv or something because there is so much power budget released by reducing the voltage.

Managed 1344MHz @ 1.105 mV on my XFX RS RX 480.

Turns out though XFX fitted their RS RX 480 cards with crummy VRM components (similar to a certain PowerColor RX 480) so it's likely why I'm not able to overclock past 1.175 mV or 1350 MHz without heavy throttling, and locked out from exceeding 20% power limit. The VRM section on this XFX card is cooled by the main heatsink so heat from the VRM plus GPU causes the fans to ramp up quick.

Flashing the Sapphire RX 580 BIOS removes those restrictions but how long will the VRM section survive those spikes to 250 or above watts? Anyways been messing around with the Sapphire 580 BIOS, and I can go up to 1400 MHz at 1.190 mV. A few of the Performance States don't match up with XFX card so I've been using WattTool to re-adjust.

I'd really have to mess up to brick this card btw since it has the bios switch feature.

The RS series are the low-cost, slower-clocked variants, right? If you want serious overclocking that'll be XFX's GRT or XXX lines.
Does it have an 8-pin or a 6-pin connector?

I doubt they specced the VRMs any higher than necessary if it only has a single 6-pin connector and 150W power target for the whole board. I'm sure most boards and PSUs can handle >75W from both board and 6-pin, but I wouldn't want to push it too far.

biffzinker wrote:

been messing around with the Sapphire 580 BIOS [on my XFX card], and I can go up to 1400 MHz at 1.190 mV.

Interesting!! I hadn't thought of using the Sapphire BIOS on other manufacturers' cards. Pretty good freq/voltage you've got also. Thanks.

Chrispy_ wrote:

Does [XFX RX480 RS] have an 8-pin or a 6-pin connector?

8 pin. IIRC, only the reference blowers had 6-pin connectors since the whole debacle that created.

Fair warning: the VRMs on RX580s are far better than their 480 counterparts.

The VRM on the XFX RS 480 was one of the worst 480 VRMs. Don't raise the power limit while it is using a 580 BIOS.

Flashing the Sapphire RX 580 BIOS removes those restrictions but how long will the VRM section survive those spikes to 250 or above watts?

It won't last long. I would advise against running any sort of power virus stress tester with the 580 BIOS as well.

Worth warning users before they try this... remember AMD's reference design 480's did not adhere to PCIe slot voltage specs! 6-pin 480 cards drew too much power from the PCIe slot, so if you flash one of those 6-pin 480's to an RX580 you risk frying the PCIe slot on your motherboard, and probably the card along with it.

I'm not flashing mine. Waiting to see what Vega brings us.

End User wrote:

I'm not flashing mine. Waiting to see what Vega brings us.

Myself, I'm waiting for the RX 560 (I put my 460 purchase on hold, because why get an older model?).

That said, when the heck is it coming out? Seems like the RX 580 was out ages ago...

You see, this one goes to 580.
All the other blokes are stuck at 480 and where are they going from there?

Couldn't you just overvolt a 480 and give it a clock bump to be a 580?

But this one goes to 580.

chuckula wrote:

You see, this one goes to 580.
All the other blokes are stuck at 480 and where are they going from there?

Couldn't you just overvolt a 480 and give it a clock bump to be a 580?

But this one goes to 580.

For an extra couple hundred buck I'll give you one that goes to 1080.

(Obligatory XKCD reference: https://xkcd.com/670/)

I picked up a Sapphire Nitro+ 580 today. Going to see how fast it'll run on 1.0V when I have some free time on Sunday.

That would be FANTASTIC Chrispy. I really wanted TR to do some testing on that for their review so we could see what the actual "process improvements" were, but it looks like I was ignored.

End User wrote:

I'm not flashing mine. Waiting to see what Vega brings us.

Flash it to Vega

Chrispy_ wrote:

I picked up a Sapphire Nitro+ 580 today. Going to see how fast it'll run on 1.0V when I have some free time on Sunday.

Bump. Chrispy, did you get time to do any testing over the weekend?

DPete27 wrote:

Pretty good freq/voltage you've got also. Thanks.

Switched my card back from the Sapphire BIOS a couple days after to the XFX OEM BIOS since I can manage a undervolt of 1.105 plus overclock the Polaris GPU to 1344 MHz. Plenty fast for me, and the games I've played. Currently playing Prey atm.

DPete27 wrote:

Chrispy_ wrote:

I picked up a Sapphire Nitro+ 580 today. Going to see how fast it'll run on 1.0V when I have some free time on Sunday.

Bump. Chrispy, did you get time to do any testing over the weekend?

Not in-depth. I lack confidence in Sapphire's BIOS, but I don't think they've done anything special at all, this is basically an RX480 with a label upgrade on the cardboard box.

I'll take a screenshot when I get home, but in words this is what I saw in global Wattman settings.

• Fixed 1140mv for the last three power states
• The highest state is set unrealistically high; It'll spend little time at this state and constantly bounce between 1416MHz and 1333MHz
• The 4th power state is a higher voltage than the last three states. It's like they didn't even bother making a proper profile, just edited the last three states (poorly) and called it a day.
• Nothing of this interim memory state that is supposedly an improvement over the 480's two states. I'm only seeing two memory states still.

So yeah, it'll briefly top 1416MHz, but for the sake of wanting to game on Sunday I just stuck it at 1300MHz 1050mv and left it at that to see if it was stable. No crashes so far, maybe I'll test tonight if I'm feeling inclined!

On the upshot, the fans are remarkably quiet and the cooler is excellent - very low temps despite the high voltage and clock boost.

Quick update:
I was wrong, it's 1150mv and it bounces between 1416 and 1300MHz for the two highest boost states :\



Quick and dirty test tonight:
1000mv - stable at 1330MHz, resets graphics driver at 1340MHz, ASIC quality is 79.5%

That stock voltage curve is odd. My RX480 ASIC quality is 80.1% (FWIW) and can do 1305MHz at 1030mV, but that's not the ragged edge. Generously say they got a ~3% process efficiency improvement with Polaris 20. Basically insignificant. Once you finish your testing, maybe post results in the GPU undervolt thread so others can reference.

The good thing about the 5xx series cards seems to be that their cooling and power delivery is beefier due to the increased design power. The fact that the 5xx series is selling for basically the same price as the 4xx, buyers might as well get the RX5xx. Undervolting is your friend. Doesn't look like AMD has improved their power delivery algorithms.

I'm not even sure I can be bothered to test much further tbh. I am seeing preliminary results exactly like the several RX480 cards I've undervolted.

These were slightly longer 5-minute burns for less superficial stability testing, excuse the shortcuts and inconsistency, I'm just not prepared to spend much time on it since Sapphire/AMD clearly haven't either

1325MHz: failed @1000mv, stable @1025mv
1350MHz: failed @1050mv, stable @1075mv
1375MHz: stable @1100mv, became impatient and clocked upwards instead of downvolting
1400MHz: failed @1100mv,
1416MHz is stock, stable @1150mv.

So, unsurprisingly it is exactly like the other RX480 cards in that it's completely fine in the 1250-1300MHz ballpark on 1V, but anything north of that requires significant voltage bumps.I'm probably just going to set this one to 1300MHz @1V and call it a day. Maybe I'll go back to the awesome 925mv reference card I was using but either way it doesn't matter which card I return to the office spares pool (that's where I've pinched this 580 from) since there really doesn't appear to be any point to the RX580 other than a bigger model number.

I was able to get my Rx480 to 1415mhz with a undervolt of 1115mV and the memory up to 2250mhz temp runs on 100% load bout 65C