AMD Radeon RX 480 Undervolting Performance

The AMD Radeon RX 480 has already proven itself to be an affordable solution for 1080P and 1440P gaming displays in our launch review and then later when we looked at 4GB versus 8GB frame buffer performance. We don’t have a proper Data Acquisition (DAQ) tools needed to measure board power, so we’ve had to sit on the sidelines for the Radeon RX 480 powergate debacle. It appears that the Radeon RX 480 is exceeding the 75W power limit over the motherboards x16 PCIe slot. AMD will be providing a software update in a driver that implements a fix and there will be an update on its progress tomorrow (July 5th, 2016).

AMD Radeon RX 480 Video Card

 

While we wait for AMD to come up with a fix for the Radeon RX 480’s power woes we ran across an article by Keith May over at WCCFTech about how he was able to undervolt the Radeon RX 480 and improve performance. With the AMD Radeon RX 480 exceeding ‘safe’ power levels in stock form at default clock speeds, proven by Ryan Shrout over at PC Perspective), we figured we’d give undervolting a shot.

AMD Radeon RX 480 Stock Voltage Settings

AMD Radeon RX 480 Stock Voltage Settings

To undervolt the AMD Radeon RX 480 you can go into AMD Radeon Settings and under the Global Wattman settings you can control the voltage levels for the GPU at seven different power states. When playing games you are usually at the highest 2-3 power states, so to keep things simple we’ll just be adjusting the two top power states. By default Power State 6 is 1100mV (1.1V) and Power State 7 is 1137mV (1.137V). We are simply going to lower those to 1050mV.

AMD Radeon RX 480 Lowered Voltage Settings

AMD Radeon RX 480 With Manually Lowered PS6 and PS7 Voltage Settings

With Voltage Control set to manual we were able to type in the voltages that we wanted and now we have PS5, PS6 and PS7 all running at 1050mV. We tried to go down to 1006mV, but the system crashed and rebooted. Just like overclocking, undervolting too much will result in system instability and could possible cause crashes.

rotr-overall

When running the AMD Radeon RX 480 8GB reference card at default settings (1136mV) we had an overall score of 67.25 FPS when we averaged our runs together with the built-in DX12 benchmark for Rise of The Tomb Raider. When we repeated the test we found that our performance and undervolted Power State 6 and Power State 7 to 1050MV we found that the performance went up! We went from 67.25 FPS to 69.59 FPS. This 3.5% performance gain isn’t massive by any means, but we were able to get a measurable difference in performance and that is exciting.

Return of the Tomb Raider Undervolting Performance

By taking a closer look at the clock speeds and voltages on the cards we can see why we are getting better performance while gaming. At stock settings the AMD Radeon RX 480 runs at it’s peak boost clock speed of 1266Mhz for about 18 seconds on this particular game title before retreating down to as low as 1211Mhz for the rest of the benchmark run. The voltage on the card was reported by GPU-Z as being 1.1187 Volts for most of the first portion of the run before dropping down and jumping around for the remainder of the benchmark. Note that one of the polling times saw the VDDC running at 1.3125V, which is a pretty high spike that is well over the 1136mV the card is set to by default. By undervolting the card we were able to keep the clock speed at the top boost clock of 1266MHz for the entire duration of the benchmark run and that meant better performance, lower GPU temperatures and less power consumption. We noticed 10-30W power reductions at the power outlet meter by undervolting the Radeon RX 480! A one voltage fits all approach obviously leaves some room for adjustment!

division-overall

In The Division we again found that manually undervolting the AMD Radeon RX 480 improved performance. On this game title we were seeing a 4% performance improvement by simply reducing the voltage in AMD’s WattMan app!

division-voltage

With the Radeon RX 480 graphics card undervolted it remained at 1266MHz for most of the benchmark run whereas the card with all default settings in WattMan only managed to hit 1260MHz at the end of the benchmark run.

bf4-overall

Not all game titles we tried out had performance differences of over 1 FPS on average and Battlefield 4 was one of those titles. We still saw performance increased thanks to undervolting, but it was a 0.8% performance increase.

bf4-voltage

This is because in Battlefield 4 the clock speeds are very close to one another. We averaged 1266MHz with the undervolted card and 1245MHz with the Radeon RX 480 running default settings, so there is only a 1.7% average clock speed difference. The biggest improvement in Battlefield 4 was at the power outlet where we saw solid 10-20 Watt power decreases thanks to the undervolt!

Final Thoughts 

It appears that AMD was pretty easy going with regards to the higher P-state voltages as we were able to reduce ours from 1.15V to 1.05V with the new AMD WattMan software and still had full gaming stability. By lowering the power level down it increases the efficiency of the GPU and for gamers that means you get to hold the boost clock speed longer and get better performance and use less power. Having the ability to turn all the P-State knobs is pretty nice and we are already seeing the new WattMan software paying off. Being able to get a small performance boost while lowering the power consumption and heat levels was pretty nice and it’s something you might want to try out if you happen to have an AMD Radeon RX 480 graphics card. It’s easy to geek out and enjoy seeing performance go up in benchmarks while seeing double digit Wattage drops at the wall!

As usual be sure to let us know if you have any questions and Happy 4th of July to all those in the United States! If we get some free time we might try raising the GPU clock and overclocking with the lower voltage to see what happens to performance!

 

Print
  • Krisnadi Imam

    my Wattman setting for lowest stable volt is 1.1v @ 1280, but in GPUZ, the sensor log shows my may volt is at 1.08v. which one should i trust?
    Thanks

  • GraveUypo

    So…. what about the temperatures before and after? Thats what i care about most.

  • dave

    All that work, and the author didn’t think to test out if the RX480 overclocks better with 1.05v – 1.075v compared to stock voltages? *sigh*

  • mykysamurai77

    I wonder if some custom card will be released undervolted by default

  • Ace66696

    Nice to know, but i wait for Sapphire Nitro before i buy 2 of them

    • Nathan Kirsch

      Coming soon! Rumors are saying July 14-16th they should be widely available.

  • Todd Evo

    If you are lucky enough to own an RX480, undervolt it, if it’s stable,
    don’t give it a second thought. If you can’t undervolt it enough (which
    is probably rare but technically possible) that sucks, get a
    replacement like any other flawed thing you bought and had to
    (unfortunately) replace, and move on.

    If you are head deep in the
    industry and fabrication of this stuff (I design custom intel/amd motherboards), your take is “Bad AMD, bad bad
    bad, but you got lucky this time (I hope)”. Then, next week, replace
    AMD with another vendor and random product and repeat until the end of
    time, because there always is. It’s why there are errata pages. For those of us buying stuff under the pretense that the MFG did their job entirely correctly, it sucks. If you even remotely think stuff like this is specific to AMD, then you’re just not paying attention. I’ve dealt and will continue to deal with, way, way worse.

    I have this question to all of the reviewers though, or anyone with an RX480 and a supply with an 8-pin PCIe card connector on it. If the extra 2 pins on an 8-pin connector are simply a “sense” and ground pin, take said 8-pin connector, put that on your RX480, take a paperclip, short the last two pins (sense and gnd) together, and trick your PSU into the “enhanced power mode” it would’ve used if it had the 8-pins. I assume this would work, since, as I understand it, that is all a video card would’ve done, shorted the pins to unlock the higher power mode. The PSU should now be able to provide more power to the card’s connector. Those pins are typically 4-6amps, a 6-pin connector “should” be able to do 150W no problem, the PCIe guys limit it to 75W artificially though (again, for presumptive reasons… that you’d have the lowest spec possible PSU) but if you have a quality 500+ with over 12A available for your PCIe card rail, it would be the 1st thing I’d try. Unsure why no one else did, unless I’m wrong about how it might work. NOTE: I would only try this on a PSU that already has an 8-pin PCIE connector because then you can safely assume you have the power for it. Otherwise it’s at your own risk, do your PSU math, make sure you have enough current, and that your PSU used the ‘good’ pins with ‘tight’ tolerances, so that 4-6A is really 5-6A…

    My speculation: Person A @ AMD: “When we make this card, we need an 8-pin connector, the power between that and the bus would be 150W, sure, but that’s too close for comfort.” Person B @ AMD months later “We can’t require an 8-pin connector, we’re going for mainstream and some random Dell’s PSU might not handle it, we’re going for market share and can’t assume enthusiast PSUs!” so it was removed with the assumption that the 150W mark was more or less solid and it should be fine. In most cases and probably on many boards it would be. But that’s assuming your motherboard’s mfg followed their own guidelines well enough for their own power delivery. A dangerous assumption, AMD should’ve looked harder.

  • keithplayspc

    Sweet 😀 thanks for the shout-out Nathan, and good work on the results 🙂

  • Tye Roberts

    Ok – here’s something odd.

    My MSI RX480 has the following DEFAULT core voltages :
    818, 824, 875, 931, 975, 1025, 1056

    So state 7 is already 1056 and shows its frequency state of 1265.

    • Nathan Kirsch

      Interesting… What is the model number on your card?

      • Tye Roberts

        912 – V803 – 862

        Also temps are being reported much lower than reviews.
        Also just took this for you – 1320 OC – Ive been playing with voltages here they aren’t the defaults I listed – ignore that. But look at temp / fan speed under load.

        • Tye Roberts

          Here is the same test – reopened radeon settings and left it open for a bit. Not breaking 80 degrees on an overclock – and fan at reasonable auto level. This doesn’t make any sense to me based on reviews of the card.

        • Nathan Kirsch

          I pinged my MSI Taiwan contact to confirm your voltages. I won’t get an answer until tomorrow though as it is 3am there. I’ve yet to see a reference card shipping with anything lower than 1.15V on the GPU, so crazy your card shipped lowe. Care to share a screenshot of it with default settings? Is that a custom board or reference? I can’t even find that model here in the states.

        • Tye Roberts

          Reset settings and had to use the switch to expose defaults and hence the ‘apply’. If you’d like confirmation via Skype or some other means I’ll assist.

          Don’t think it matters, but I’m Canadian. Bought this card in Calgary last night.
          * It is a reference board purchased at a chain here called “Memory Express”

        • Petar Posavec

          Out of curiosity… have you tried lowering the voltages of the VRAM as well?
          Some have stated that the larger power draws/spikes occur due to VRAM usage increasing quite a bit (and GDDR5 can be power hungry), so would be interesting to see the results if you can (in addition to the core) undervolt the GPU RAM.

        • Tye Roberts

          Not yet, no. I read everywhere that RAM was alright to go straight to 2200 MHz without changes so that’s what I’ve done. Might look later when I get some feedback as to what may be safe.

          As of now, I’ve OC’d the core to 1360 with 1100 mV at that final state. My temps with the fan hovering at 3000rpm range are lower than 80 once the fan catches up. Maybe I got lucky, but I don’t think I got that lucky. I think all these cards are over volted.

        • Petar Posavec

          That’s pretty much my explanation behind this ‘power draw issue’ as well (jacked up voltages).
          Now, if the VRAM is the culprit as some claim it to be (and AMD did sort of confirm it in an official statement), it will be interesting to see what the total power draw of the GPU will be after undervolting that as well.

          If the GPU is maxing out at 135W of power draw while undervolted, (or just above that), then undervolting the RAM might slash the power by quite a bit (after all, transitioning to HBM from GDDR5 last year did bring a lot of power savings on the Fury line – though that line of GPU’s was also relatively new on a well understood process).

          Could you post power draw and performance results after undervolting the VRAM as well in addition to the core?

        • Tye Roberts

          I might when I read/see a bit more about undervolting RAM. I read somewhere that someone halved the mV setting on RAM. That seems insane.

          The big take away for me here is that I’m OC’d to 1360 core at 1110 mV. I’m still under 80 degrees once the fan (which seems a tad slow to ramp up… but does get there in time). Max fan setting is currently set to 3200 – but that’s only because it never really gets that high and I wanted to “push” the fan curve a bit higher. Reviews of this card seem to think it gets way too hot – I’m not seeing that considering I’m almost at a 100 MHZ OC now. I think the heat isn’t due to a ‘poor’ cooler – although I’m sure aftermarket coolers will be much better – I think it is due to over volting the card. I have a feeling that in a few months perhaps we will see that MOST folks are able to undervolt the card a great deal – and considering that my DEFAULTS are lower (we still don’t know why) it seems that it was the issue all along.

        • Petar Posavec

          Out of curiosity, what is your GPU power draw with the undervolts on 1360 core?
          Also, is it possible to lower the voltage further than 1.11V on 1360 or is that the lowest the gpu can handle with those clocks?

          Either way, I hope you will try undervolting the VRAM today or tomorrow.
          🙂
          I don’t know whether slashing the voltage in half on the VRAM would necessarily be doable (seems too much), but I imagine that undervolting it should be possible – just by how much I have no clue (but if you do manage to get it down quite a bit, you might start seeing lower temperatures – just make sure the undervolt on the RAM is also stable).

          And please report here.

        • Tye Roberts

          Well I don’t pretend to be an expert – I used to build machines more when I was younger but I’m just a average knowledge stay at home father now.

          I did have a crash at 1100mV when running at 1360 mhz during a ‘stress test’ – so I upped it to 1110 mV and tried again and it worked. I have no idea what the total power draw of the card is – not sure how to measure that here – but I have a lot of headroom on the power supply and I trust it – same with the motherboard – so I’m fine there.

          Haven’t undervolted RAM and dont intend to soon, however, I did try to push that overclock a little past 2200 mhz and had a crash – so I can’t imagine at 2200 mhz I’d be able to undervolt the RAM a whole lot. Maybe I’m wrong about that.

        • Tye Roberts

          Maybe the BIOS of the card was updated with different values?

      • Todd Evo

        Nathan, can you try stock voltages, and an 8-pin connector to the card, shorting the last two unused pins to force the cpu into it’s enhanced power mode? It would do the same thing a legit 8-pin connector on the card would’ve done.
        I’d like to see if it helps, and anyone with a sufficient supply may be able to resolve it that way. The 8-pins, as I understand it, are just there to easily differentiate between a “really good PSU with enough beefy power for some insane card.” and “any random power supply with a PCIe connector and who knows what supply voltage or quality.” and nothing to do (assuming the pins and connector are quality) with how much power the connector itself can handle. A 6-pin CAN do 150W, it just doesn’t because mfg’s recommended to keep 6-pin at 75W for PSU quality assumptions and nothing to do with the physics of it itself, and as a result the card won’t draw more from the connector because it doesn’t “sense” that it can. I’d do it myself but I couldn’t get a card.

        • Todd Evo

          Just an amperage or voltage swing on the 12v wires of the pcie connector both with 8 and 6 pins would tell you. I have multimeters for days but no cards 🙁

  • Sachin Lakshminarayana

    whatever its is my R9 390 is best

  • I did the same trick with R9 290 Reference. Yes, performance is improved compared to even when the card running overclocked. Simply because of the poor cooling solution. with lowered volt, the card run cooler and throttle less when hitting target temp. Power reduction is significant because the GPU is much power hungry than Polaris. I think this is the trick AMD did with R9 Nano to make the GPU consume significantly less power.

  • Kyle

    So the RX 480 is better when you undervolt it? It sounds incredibly silly, but the evidence is here. It’s highly possible that the RX 480 Legitreviews has is good. Not everyone gets lucky at the silicon lottery. Some RX 480s may not undervolt too well. There must be a reason AMD chose those stock voltages.

    • Todd Evo

      Assume for a moment your GPU cooler is inadequate. It’s RETAINING heat too long. Now, run your GPU at full speed. At this point, what AMD expected to happen and what is actually going to happen are two different things. AMD assumes you’re using a good enough cooler. Even if it is the reference one a bunch of vendors make them, ymmwv. So, your GPU is now retaining too hot for too long, when that happens it drops speed to cool off. That switching time has it’s own performance cost especially when its happening over and over again because your temperatures aren’t stable. There’s your performance hit. You’re basically losing what you expected to gain. The same will happen with any modern CPU with speed throttling. I work in with high performance computing and I have clients disabling turbo boost regularly because the performance can’t be guaranteed. These techs work great when the results can be stabilized (good cooling, good power) but cost you when they don’t. Undervolting stabilizes the clocks by making the cpu consume less power for the same effort. Less heat, less throttling.

      I really wish I had a card because I have the rest of the equipment to test and measure using an 8-pin connector and a paperclip. PSU goes into higher power mode, done. Those with insufficient power supplies are out of luck, or need to upgrade, but if 8-pin power fixes this for most it’s a free – $2 solution for a custom cable and power solution resolved. If anyone is sketchy, I’ll happily buy your card for the full purchase price + shipping and test it myself.

      If you’d rather undervolt, and it doesn’t work, try ticking down the clocks a bit. It sucks, but who knows maybe you’ll only lose a few frames and you’re back to square 1, but with less power draw.

  • John Pombrio

    In order to stay within the PCI spec, expect to under volt by 30%. PCper found that a demanding game had the board go over the spec for all the time while playing the game by 27% at stock speeds.

    • Mr.Gold

      Thats now how it works. Its not a linear curve from volt to power.
      Also PCPer didn’t show 30% over the 66w limit.
      Most game didn’t draw an average over 70w, the max peak where 80w.

      So the peak overage was about 20% (66 to 80w)

      Quick ref: For the R9 290, 50mv represents about 60w of power saving.

      It doesn;t help AMD, but its very possible 95% of the RX 480 can run at lower voltage and reduce the wattage from ~160w to ~130w.
      Making it within spec.

      Like other have said, it seem AMD, at the last minute, at to pump the voltage to increase the binning… because Global Foundry wafers came out sub standard to what AMD expected.

      Again AMD got royally screwed by GF . And AMD will pay for it, and GF will profit.

  • brucer

    meh, I’ll be returning or selling my unopened rx480.. I cant trust amd anymore.

    • GmailIsDown

      I have an unopened Rx 480 sitting on my desk and I m on the fence. I will wait until AMD’s software solution tomorrow and then the responses from review sites testing their solution before deciding what to do with this card.

      • Todd Evo

        I’ll buy them. I’m fortunate enough to work deep in the industry to understand how stuff like this happens and what the more probable risks are. I’m not saying I don’t blame you for feeling this way, I’m saying it means you’ll have to blindly trust:
        http://www.theinquirer.net/inquirer/news/1004378/why-nvidia-chips-defective
        and
        http://techreport.com/news/29585/prime95-can-cause-intel-skylake-cpus-to-freeze
        as two completely random examples. But that stuff will never happen again, right?
        Do you trust microsoft and valve? Do you have a choice?

        This stuff happens, it sucks, but it happens and sometimes it’s really bad but it doesn’t feel this is “really bad”. That said, if you’re both sincere, I’ll buy them if they’re the $200 version, I’ll pay shipping of course. I’ll either undervolt them or more likely replace the cooling and try an 8-pin connector. Either voltages were increased for binning, or the consequence if sticking with a 6-pin connector weren’t fully understood. Both are absolutely plausible and I’ve seen it with every manufacturer just about, and way more than once. I’ll take the chance.

        • GmailIsDown

          Mine is the 8GB version from sapphire so I guess you don’t want. I thought about switching to a 8-pin connector but I am not sure if it will just work with a cheap 6-pin to 8-pin adapter or I have to open up the card and replace it. I don’t think I will replace the stocker cooler with an aftermarket one. My case is compacted and a hooded version actually offers better cooling than an open design, that I’ve been told. I will try undervolting it as sites reported that that actually consistently improved performance, though YMMV as to how low a card could go before becoming unstable. Maybe the yet to be announced AMD solution will do the same thing.

        • Todd Evo

          As I understand it, it’s like this. The PCIe guys said: “6 pin adapters are only allowed to supply 75W”. That was a kindof arbitrary number, the general idea being that any PCIe with a 6-pin connector could supply it no problem and your card can’t draw more than that. If that’s the case, it draws more power where it can, the slot. PSUs with 8-pin connectors on their PCIe (repeat PCIE 🙂 not 12v+ or EPS 12v… etc..) on the other hand would have to be able to supply 150W through that connector.
          If your PSU has an 8-pin power connector for the PCIE Card, you’ll see that the 2-pins (which the plastic part usually breaks off the 6-pin connector on some PSUs) are connected to each other and a nearby black cable going to the 6-pins. The bottom black row of wires is ground, those 2 pins are connected to ground, but not to each other. Not until you plug it into an 8-pin video card. All that 8-pin video card does is connect those two wires together internally. So, you can do the same to those 2 pins with a paperclip or something.
          What that would do is tell your PSU: “Hey, your sense wire is going off, this is an 8-pin card, give it 150W of power if it needs it, not that 75W 6-pin stuff.” Now the card can get power from there, when it does it would relieve the slot. From there, you’d be left with the card you expected to buy. All undervolting does is get the card to consume less power overall, to compensate for that 6-pin 75W limit. Doing my suggestions breaks that 75W limit by telling your PSU it just needs a bit more power. I honestly think AMD forgot about the 75W limitation on the 6-pin connector. I honestly think they were being clever by not using 8-pin versions so more people with older supplies could use it, but didn’t consider this problem. I’m not saying I’m right, but I am saying I’m willing to pay for a pair of cards to find out myself, unmatched is fine, but I have a budget and $200 + shipping is what I was looking for.

        • GmailIsDown

          I am a bit confused. If I have a 6+2 pin connector from PSU, as the two extra two pins are dummy pins, won’t connecting the 6-pin to the card allow it to draw up to 150W from the 6-pin?

        • Todd Evo

          There are two types of 6-pin connectors, 6 and 6+2. The “6” in 6+2 are technically interchangeable with a standard 6..
          However, on a PSU that ONLY has 6-pin connectors, you might notice that there are only two power wires. But on a 6+2, it has them all populated. Other manufacturers might have gone ahead and wired all 6 circuits anyway even if they didn’t put a “+2” but that really would be pointless to do.
          So, even without the +2, the “6” in a standard 6 and a “6” in the 6+2 are not the same.
          The +2 adds an extra sense pin, when it is shorted to ground, it allows the 6+2 connector to add more available current (in part because of the additional power wire)

          When you short those additional two add-ons (pins ‘4 & 6’ on your 8-pin) it’s telling your supply that your card is an 8-pin card and needs to use the additional power it has. A “real” 6-pin connector doesn’t have it.
          https://en.wikipedia.org/wiki/PCI_Express#Power

          Caveats:
          1) I don’t know for certain I’m right, but I’d try it, nothing should break, I’ll either consume more power from the PCIe aux port (meaning: problem solved) or I won’t (meaning it didn’t make a difference.)
          2) If AMD designed their card exactly to spec, then one of the power-side of the 6-pins doesn’t go anywhere. If it connects to the other 12V, great, if not, oh well, but not a big deal, that just, means it’s only drawing power from 2 pins. That doesn’t scare me personally, the card only needs an amp or two more from that connector and the 2 pins it uses should be sufficient as I have a have a decent supply (anyone should if you have the 8-pin conn as an option in the first place. )

        • GmailIsDown

          Thank you 🙂 I understand now. I will still need to short the two +2 pins together for it to be able to draw 150W. AMD just announced its solution (undervolting and changing power distribution) and it confirmed what you said.

        • Ascaris

          The current draw is determined by the card, not the PSU. It’s the card that has to sense that it is connected to an 8-pin connector, and the PSU has no idea or concern about that. The PSU just produces as much current as the device draws from it until it reaches its overcurrent limit or it self-destructs; it has no idea of the power requirements or limits of any device attached.

          As you mentioned, the two additional pins in the 8-pin PCIE power connector are already connected together and grounded via the wires in the power lead itself, and connecting two wires that are already both grounded doesn’t change anything.

          The thing that needs to change is that the 8 pin sense pin on the card has to be grounded. That’s what lets the card know it can pull more current from the PCIE connector. Since this is a reference design that was meant to have only 75w drawn through the PCIE connector from the start, it is unlikely that simply adding a jumper wire will be possible or effective.

          If the PCB itself has the provision for an 8-pin connector to have been optionally installed at the time of manufacture (which might be used later with other reference cards from the same series, for example), you could try adding a jumper wire between the two unused pins, but it’s going to take soldering on your brand new, under-warranty card to do it properly– and even then, the odds are that the card’s firmware/driver combo won’t be paying any attention to the pin that senses an 8-pin connector, as it won’t be expecting the possibility.

          Someone could modify the card’s firmware or driver (depending on where the limit is enforced; it could be either or both) to raise the power limit over the 6-pin connector, and it might not be a bad idea to drop it a bit for the PCIE slot at the same time, given the reports that have surfaced. How easy or difficult this would be remains to be seen, but it is probably the only way to achieve what people are talking about here.

        • Todd Evo

          Thank you for clarifying! Now that AMD has shifted the load in software, I’d still use the 6-pin part ignoring the +2 and have the extra wire to share the load.

    • Todd Evo

      if you’re serious and it’s the 4GB version I’ll buy it (if you’re in the US :p)

      • brucer

        its the 8gig sapphire model

        • Todd Evo

          ahhh… sorry, I really wanted the 4GB one…

        • Todd Evo

          though i’ll throw $200 + shipping out there any way, I won’t blame you for saying no 🙂 it’s there.

        • brucer

          I’ve already spoke to the place I ordered from. They accept a full refund being the package is unopened.. so unless you can beat $250 plus shipping I’ll have to return it.. I thought about putting it on ebay but I dont care to mess with it. I’m probably just going to return it and get the amount refunded to my account and put it towards another 1070 or a 1080. I’m done with amd’s incompetence.

  • AMDExcites

    Don’t have to do this on Nvidia cards, they automatically do ti for you and have done it for YEARS.

    • Maurice Fortin

      Nvidia relies on chopping things away and game code trick to make seem faster just wait a few drivers and what performance it has will be lost magically as the new cards are coming out, ignorance is bliss.

    • Todd Evo

      Hey, as long as they properly attach their silicon to their substrate, nVidia makes great stuff.

  • Athenasius

    was reading elsewhere, people have been getting stock clocks and undervolting the ram by 0.2V and it staying stable. some people getting as low at 0.650V

    • Nathan Kirsch

      yeah, underclocking the memory could also be done and it’s the same method as this. I totally ignored memory on this article as it gets complicated to explain and show it all!

      • Athenasius

        good article, the graphs show that something is not right with the AVFS settings. Hopefully a software/firmware update should sort it, from what i was reading on their graphs, the system should be lowering to the most stable voltage for a set clock. Is obviously not working with how it appears to spike and never stays as low as your set voltage.

  • Richard Krupski

    The performance issues might be related to the power issues so AMD’s fix might improve performance… That would be a nice win-win for them.

  • Dmitri

    Lol …
    It’s almost sad to read most of the comments as they belong to fanboys.
    AMD had released a card that has a built quality of 100$ junk. That cooling system is so cheap … . They released the card that overdraws power – either idiots, or greedy morons.
    The fact that cards can operate at 10% less voltage, tells us that probably they wanted to use even shittiest chips to sell as more expensive. Nvidia usually just makes a less expensive cut down version of them.
    And they (red fanboys) are just happy when someone peeing on their face.

    • gargoyals

      Savage

    • jpteknoman .

      the problem is that AMD never had boost clock so the implementation is all over the place as seen by the graphs.

    • Maurice Fortin

      and green fans are happy that they continue buying underspec overpriced slower as it ages junk, its ok, we understand your ignorance.

    • gueststar

      if amd is going to fix the power issue whats the big deal… ofcourse the card has a cheap cooling solution its a reference design thats cost $200.. would you rather they do what Nvidia did and charge an extra 100 dollars for a better cooling solution that is still inferior to AIB designs.

      And nvidia uses shitty chips too all cpu/gpu manufacturers use shitty chips in there line up. thats why some can barely OC and others are great at it.

      • Dmitri

        I’d like to see something other than this simple aluminum heatsink under shroud. In that strive to make big news (affordable performance) they missed the point. It does not meet the build quality standards of a 250$ video card.
        And frankly, 150w is kinda too high for this card. Nvidia offering GTX 1070 with 150w. OC everyone is using shitty GPUs, it’s the way to use them is different.
        I had an AMD GPU that was working on stock clock in some games, but would crash on others … I figured out few month later, that if i under clock it a bit it will work fine. Since it was already second card (first produced artifacts) I just went and bought GTX 570 that served me for 6 years despite 1.1V and 20% overclock.
        I was never a fanboy of any company and was going back and forth between vendors depending on the offer since rage, tnt and 3dfx cards.
        It’s just Nvidia does better cards recently. AMD might offer better price performance in some cases, but i have not bought a graphics card for less than 300$ for over a decade. So I care about additional value (for me that would be Linix drivers, Shadowplay, GF experience) and I care about build quality since i like them to run for long time. If i was in to 200-250$ cards, I would be waiting to see the gtx 1060 and would definitely consider good 970 (already seen asus strix card for 240$) over reference 480.

        • gueststar

          Dmitri all reference coolers are shit. yes AMDs is even more shit then Nvidias but this card is a lower mid range card. and amd isn’t charging you a 100 dollars for it. the 1060 has an even shitier cooler then the 1070/80 and they are charging $50 dollars for it.

          if power is a major deciding factor then go with Nvidia. not everyone has expensive electricity and some people even have it included in their rent if they don’t own a place so its a non issue for some.

        • Dmitri

          Power is important since it related to heat. I live in a place with over 30c ambient for about 6-7months a year.
          So i generally prefer parts that run cooler (consume less power). As for the reference cooler, nvidia’s reference coolers are the best when you need a blower type (small form factor, sli etc.). The actually do the job.
          If i compare visually the reference build quality of 1060 and 480, somehow the 1060 seems to be a bit more invested.

  • Athenasius

    From the looks of the jumps in the voltage when set to default/auto, it appears that the new AVFS engine needs a lot more tweaking. If it was working, it should have optimised the voltage itself down to the levels shown in this review.

    • Maurice Fortin

      that is my thinking, they simply did not have time to thoroughly tune tune the AVFS which is in a league of its own for what it is, and more then likely is very new to all those involved so it will take a bit to “tune” and optimize, IMO performance and power may not look amazing right now, but give a month or so and should be quite better. I know myself am waiting on a revised blower style reference cooler as I much prefer card to vent its own heat and just via spec alone, paying ~$20 less then I did for my 7870 years ago that RX480 8gb is WAY faster and using approx 20w less in the process as well so to me, is quite amazing card, just waiting on the $ or would have one right now.

      AMD builds very good products that last and get better as they age, this has been proven many times over many decades, with many of what they do release “living” just as long unlike some green products 🙂

      • Athenasius

        Indeed, other factors in terms of the performance improving come from tuning the ‘Primitive discard accelerator’ and the command processor in newer drivers. Since the former is controlled partially by drivers while the latter is microcode update-able from what it shows in the AMD slides.

    • Petar Posavec

      May I ask… do you happen to know how much further did entire GPU power consumption go down after reducing the voltage of 480 VRAM by half?

      • Athenasius

        think some people save a good 15 – 30w when undervolting overall, but a nice amount of that is from vram alone.

        • Petar Posavec

          I was under the impression that 30W power savings came from just the core undervolt and untouched VRAM. Hence why I asked about a combination of both

  • 200380051

    Well, well. Hawaii did that too…

    That explains a lot. AMD must have set te bar pretty low with regards to binning; this must help yields enough for a viable 200$ price point. I expect the AIB versions to be much better than the reference models.

    • KVragec

      But if the chips would be bad undervolting them would result in crashes??? Don’t you think.

      • 200380051

        Some, yes, but that would be the worst case (barely passing binning). In the majority of cases though (mean binned chips), you would get undervolting headroom, and plenty of it in the best cases (Binned juuuust under the threshold of a better bin spec, probably reserved for AIBs).

        • KVragec

          Well, we will see. We still do not know what AMD made wrong. But problem for the buyers of the reference model is that AMD could simply limit the boost clock since the card is advertised up to 5,8 tflops. So the card could end up with lowered clocks. If they made a different mistake, with voltages accros the board and faulty tuning then performance could end up the same. Maybe boost clock is responsibel for erratic voltage jumps. But that are all my speculations. I’m buying custom model anyway and i don’t care. Will see what 1060 will bring to the fight and for what price also

    • Shamz

      I’d be surprised if AIB would be worse than reference. If so, that would be a first.

  • KVragec

    Only problem right now is if AMD is aware of to high voltages or they will simply throttle the card to 150 watts and all the downvolting is on the users themselfess. And also is the to high voltage deliberately like that or the mistake because they wanted to give the GPU some headroom.

    • Nathan Kirsch

      They have several options, so we will have to wait and see what they do this week after they announce more information tomorrow. If they do have to dial the reference card back to get it within specifications it will just help the custom cards sell better. The performance numbers are solid as is the $199 entry price.

      • KVragec

        True. My general opinion is that no one shoulf buy referencd model. And by that i mean AMD and Nvidija. Nvidija dissapointed mi with their pricing of the 1070 and 1080 and on top of that retailers are pushing the price even further. In Croatia the price for the Gainward GTX 1080 Phantom edition is 850 USD, and that is the lowest, average is around 950 to 1050. 1070 is 620 to 650. Strix even 700

    • The_Countess

      usually voltages are selected to hit yield targets at the target clockspeed.

      so its likely that a number of cards needs this voltage to remain stable (with a bit of headroom to account for less then ideal PSU ect).

      basically AMD picked these voltage deliberately for a reason.

      AMD could make the voltage variable to what the individual card actually need but that would likely create a PR problem when people that got a card that has a higher voltage then the ones in the reviews starts to complain.

      so it’s up to the users to under-volt their own cards and find out what each is capable of individually.

      • KVragec

        Unfortunately that is what it looks like.

  • Felipe Queirolo

    Good to know, everybody should start tinkering with this

    • Djordje

      It’s why they made the tool with all the options, is it not?

      • Felipe Queirolo

        Of course, but few reviewers have discovered this!

        • Djordje

          Enthusiasts shouldn’t rely on reviewers or at least most of them.

        • Felipe Queirolo

          This cards are for gamers, and we all know that gamers aren’t the cleverer ones, they are marketing bit..es, they do what the industry make them do

        • MetalQuintessence

          Unfo very little of the majority of PC users are actual “enthusiasts” and would go at such lengths to tinker with this on their own. Some simply don’t have the time. So they need a little bit of help. And making more pl aware of that will make the companies care more about pleasing their customers in that regard. 😉

        • SlutMagnet

          Your comment makes no sense.

  • triplezero1

    Wow! Will be doing this when I install my new Sapphire model.

    • KVragec

      I’m also waitng for the nitro model and definitely will be trying the same thing. But thing is that maybe AIB partners will have different voltage settings across the board cause the pcb is different. Sapphire said that pcb is custom, also there is 8 pin on it

      • Nathan Kirsch

        ^^ I’m hoping the AIB’s might tinker with the voltages a bit on some of the higher end cards. I know Sapphire also might have a Toxic card in the works down the road, so once powergate is figured out maybe the AIB’s can really get the voltage and clock frequencies dialed in.

        • KVragec

          Well in all honesty i think they can. I have two examples. One is my R9 270X which is running using 1,150 volts and second is friends GTX1080 FE which uses only 1,02. I know that none of those are the same manufacturing nodes, or even foundries but that makes me thinking that voltages are simply to high on RX480 reference

        • triplezero1

          I don’t really care for overclocking, so I just got a Reference model. I wanted a cheap, efficient way to run DOOM and other AAA games at 1080p Ultra 60FPS (My R9 280 performs at about Medium 60FPS). In addition, I plan to get a 25″ Ultrawide, which I don’t think my former card would’ve been able to handle.

        • KVragec

          That is a nice setup. Well, i’ll just might give it a go to see how much can it be pushed on a stock voltages with minor increase in power range

        • triplezero1

          Update: I’m running DOOM at Ultra at 140FPS!

        • Maurice Fortin

          I have experience with many cpu/gpu undervolting it is well worth doing, but comparing Radeon to Geforce in this regard, theory is same but that is it. Geforce tended to use either very high volts or very high amps to feed VRM which usually was underspec cause it was less costly, and less phases, the RX480 has a VERY robust VRM all Radeons going back to at least the 2k series have had awesome VRM and more often then not used a balanced approach of volt and amp to power things, not pushing one extreme or the other, also part of the reason Radeons tend to not drop dead nearly as often.

        • KVragec

          Ok. Get your point.

        • KVragec

          Well,partners will thorougly test the cards with various voltages to see what is optimal. I saw a stream with Sapphire Ed and he said, despite the ban until launch, that card has 225 watts at disposal but that needs nowhere near that much. We’ll see about that, also, people forget that to high temperatures have huge impact on consumption also and better cooling on vrm and GPU have a nice impact on that

    • KVragec

      But since Nitro will have 8 pin and higher tdp i’ll probably opt for overclocking on the stock voltages