AMD Radeon RX 480 8GB Review

[InvalidError]

Your mb determines how much power your pcie slot can give your device. 75w is a baseline, not a ceiling.

It is a ceiling for the PCIe slot. The card can use however much power from auxiliary power connectors as it wants but to pass PCIe compliance, it must not exceed 75W (which may actually be 60W on 12V + 15W on 3.3V according to some sources) from the PCIe slot.

I find it quite frustrating how it is so difficult to find a definitive source for such basic specifications of an "open" standard but I have not found any single source stating that a PCIe card can draw any more than 75W from the slot's five 12V and four 3.3V pins. They all say 75W max.

 

[RedJaron]

A lot of you bashing on the 480 seem to be missing something, and that is the 480's effect on the GPU market and pricing. Here's another way to think about it.

Recall when the 970 was released. It was roughly 20% faster than its predecessor, the 770. And it came out at the $330 - $350 range. Suddenly top-shelf gaming performance didn't double the cost of a budget PC. There was absolutely nothing in that price range that came close to the 970's performance.

However, until the 1070 released a few weeks ago, the 970 cost the same as it did when it was released ( barring sales and rebates of course ). That's almost two years without any price drop at all. This has been a sore spot for a lot of upper mid-range gamers. NVidia blatantly refused to offer a 960 Ti or anything to span the $130 gap between the 960 and 970. They were getting too much margin on 970 sales to risk undercutting it. And they did their research well and found the exact price people were willing to pay a lot more money than they wanted to get a card that exceeded their needs to avoid getting a card that was just under their needs. Even the lashback firestorm of the 3.5 GB VRAM didn't sink the 970 sales.

Now, look at the 480 in the same light. For $200 there isn't a card that can come close to its performance. You're getting more than playable performance for 1440p for only $200. It utterly destroys the 960 while only using slightly more power. The 380 and 380X are now completely irrelevant as the 480 is faster, cheaper, and more efficient. As of right now, the 480 is to upper mid-range cards what the 970 was to high-end cards. Have you checked prices on the mid-range GPUs right now? The 950 is down to $130. The 380 can be had for $150. As was often said of the 970, this is incredibly disruptive in the GPU pricing scheme.

Of course, like the 970's 3.5 GB problem, the 480 also has a power draw issue. Arguably, that's a worse problem because the limited VRAM didn't mean the card might not work correctly on your mboard. Hopefully this is something that can be easily fixed in firmware and/or 3rd party PCB designs.

If it can, it's possible the 480 will see similar commercial success as the 970. Except while the 970 ran mostly unopposed ( unless you wanted a very hot, very power hungry 290(X) ), NVidia can still release a GPU that upstages the 480. I hesitate to say that will be the 1060, though. Look at the past X60 variants. The 560 went up against the 6870. The 660 (Ti) variant paired with the 7870 (LE). Direct generation comparisons get muddy there after the rebranding AMD did for the 200 and 300 cards. These were not cards aimed at upper level performance. The point of the matter is the 1060 would have to be faster than the 970 in order to beat the 480. The 960 was not faster than the 770, nor the 760 faster than the 670, so this seems unlikely based on past results.

If anything, the 1060 Ti ( if we see one this time ) would be the most likely card to do it. But until that gets launched ( and if the 480's power draw issues get solved ), AMD can claim the mainsteam and upper-mid level sales right now. And unlike the 970, the 480 is much more affordable so a wallet-conscious consumer doesn't have do much hand-wringing before buying one.

 

[n0ns3ns3]

Read again: "PCIe slot specifications"

PCIe slots - the things you plug your PCIe cards into that provide PCIe x1/x4/x8/x16 data connectivity to the CPU or chipset, are only rated for 75W. While benchmarking, the reference RX480 was pulling over 100W from the motherboard's PCIe slot. Those tiny pins in the x1/x4/x8/x16 slots are only rated for about 1A each. (75W / 5 guaranteed 12V pins / 12V = 1.25A each.)

LOL ... It was bad idea to post something that late. it was not about PCIe.
The "wrong" was referring to is this statement:

Instead of pushing the very limit of the PCIe slot specification and beyond, AMD should have simply used an 8-pins PCIe connector and powered just about the whole GPU from that.

There is no need in 8pin. 6 pin is perfectly able to deliver over 200W by the spec according to molex spec

 

[n0ns3ns3]

@RedJaron
I mostly agree with you. I even hoped that RX 480 would be better performer than it is in order to see the price drop that was done to GTX 7xx, applied to 1070/1080.

Here how the pricing works regardless of absolute performance numbers.
multi 4K monitors setup (GTX 1080 in SLI)
single 4K monitor (GTX 1080)
2K or 2k 21:9 @144Hz (GTX 1070 - GTX 1080)
2K or 2k 21:9 @60Hz (GTX 1070)
1080p @144/200Hz or 1080p 21:9 (<GTX1060/1060ti> - GTX 1070)
1080p@60Hz - GTX 1060/1060ti

The above list is OC for the ultimate game play with high settings and decent FPS. But generally that's what this cards will be aiming at.
Now, the huge chunk of sales, is around 200$.
so:
the 1050 will have to be below 200$(150-180) with ~960 performance.
the 1060 will have to be directly competing with RX 480 -> 200$ 970+ performance.
it will be idiotic to have over 200$ gap to 1070, so 1060ti at 300$ with 980+ performance

 

[Zoolook13]

I think a lot of people fail to acknowledge that those extra 4GB of 8GHz GDDR5 doesn't come without a power cost, they are quite hungry, maybe even AMD's engineers forgot that after finding the 4GB variant come well within the 150W TDP.

 

[Onus]

Red, I think you're right about the absolute performance, and bang/buck, for the RX 480. If I needed a powerful graphics card (and didn't have a GTX970), it would be at the top of my list; except for the potentially damaging power issue. The possibility of smoking a trace in a PCIe slot is, and should be, an absolute dealbreaker. I would think it's an easy fix by any number of board partners, but until then, this card is a bad idea, at any price.

 

[quilciri]

AMD needs to start focusing on the higher end as they've been mastering the main stream/mid-tier for a while now. That may change now with the less than stellar 480 release and upcoming GTX 1060. Their Fury X was a monumental failure: it didn't even stay on the market six months after intro nearly a year ago, while the 980Ti is still selling a year and a half later after introduction.

*Start* focusing on the high end? That's what AMD has done every previous product cycle since they bought ATI, and that's what ATI did before them. It hasn't worked well for them the last few years and as you yourself pointed out, that's not where the profit or market share is. That argument contradicts itself.

The point of the steam hardware reference: It doesn't matter how far one card is out in the lead against the entire body of available cards. You say AMD needs to gain market share against Nvidia...they have been for a year now and it's been with mid range cards that they've made inroads.

 

[InvalidError]

The "wrong" was referring to is this statement:

Instead of pushing the very limit of the PCIe slot specification and beyond, AMD should have simply used an 8-pins PCIe connector and powered just about the whole GPU from that.

There is no need in 8pin. 6 pin is perfectly able to deliver over 200W by the spec according to molex spec

The connector may be able to handle it but if you pulled 5A per #18 wire and have 75cm cables, that's 125mV of voltage drop across each wire, up to 250mV total if most of that current gets returned to the ground wires.

 

[n0ns3ns3]

[
The connector may be able to handle it but if you pulled 5A per #18 wire and have 75cm cables, that's 125mV of voltage drop across each wire, up to 250mV total if most of that current gets returned to the ground wires.

My 250w GPU is powered by a single 8 wire cable that has split to 2x6/8 pins connectros at the end.
This 6/8 pins power option is purely a decision of board makers. And power delivery on each is limited by the graphics card BIOS/HW (depends on the model)

 

[IInuyasha74]

Red, I think you're right about the absolute performance, and bang/buck, for the RX 480. If I needed a powerful graphics card (and didn't have a GTX970), it would be at the top of my list; except for the potentially damaging power issue. The possibility of smoking a trace in a PCIe slot is, and should be, an absolute dealbreaker. I would think it's an easy fix by any number of board partners, but until then, this card is a bad idea, at any price.

Yup, that is what I was saying when the review went live. Lot of people complaining because they aren't even thinking about price/performance.

There is that power issue, but fortunately it won't end up being a big deal. Soon the OEMs will make better boards with extra power connections and the issue will vanish. I'm honestly surprised AMD didn't work around the problem in the first place. Really shouldn't need a board partner to fix this sort of problem.

 

[Onus]

...I'm honestly surprised AMD didn't work around the problem in the first place. Really shouldn't need a board partner to fix this sort of problem.

Indeed; this was not even yet a beta release. AMD must have figured if game developers can do it, then why not them?

 

[InvalidError]

I'm honestly surprised AMD didn't work around the problem in the first place. Really shouldn't need a board partner to fix this sort of problem.

AMD could drop the stock clock to 1GHz on reference cards and that should fix the excess PCIe power draw issue. Now that the GPUs have launched though, it becomes a matter of weighing the liabilities from possibly frying PCIe slots vs the liabilities from lowering the performance post-launch/sale.

 

[Math Geek]

looks like tom's called the pcie slot issues. too bad folks were not listening!! http://wccftech.com/amd-rx-480-killing-pcie-slots/

not good at all for amd and these cards

 

[Onus]

...and people wonder why TH uses that "fancy" equipment...

As much as I dislike money-grubbing lawyers, I could see this turning into a class action. I would not be surprised if AMD fingers a scapegoat for the premature release and throws him/her under the bus.

 

[jimmysmitty]

looks like tom's called the pcie slot issues. too bad folks were not listening!! http://wccftech.com/amd-rx-480-killing-pcie-slots/

not good at all for amd and these cards

If true it makes me wonder what was going through their minds and if they knew about this. They had to know about this.

Hopefully it is just a issue in the BIOS and they can get an update out that puts the PCIe back to 75W and then the rest of the power us drawn from the PCIe cable from the PSU.

 

[InvalidError]

RIP, AMD RX-480 reference design.

Morale on the story? Don't design boards that draw 75+W from the PCIe slot, especially when you already need at least a 6-pins PCIe AUX power connector anyway which can handle a 300W overload much more gracefully than the PCIe slot can - there is a heck of a lot more copper in #18 wires than in PCIe edge connectors.

 

[Math Geek]

hopefully this can be fixed with a bios update. if it just a really bad design, then it could get real bad for amd trying to recall these boards!!!

with the 470 and 460 still not out, those cards will suffer in sales for sure as folks will hesitate until they can be tested too.

 

[n0ns3ns3]

hopefully this can be fixed with a bios update. if it just a really bad design, then it could get real bad for amd trying to recall these boards!!!

with the 470 and 460 still not out, those cards will suffer in sales for sure as folks will hesitate until they can be tested too.

sales would not hurt that much.
Most people would never know such an issue existed.
congrats, you are a member of a thinking minority

 

[InvalidError]

hopefully this can be fixed with a bios update. if it just a really bad design, then it could get real bad for amd trying to recall these boards!!!

A BIOS update cannot change how the board's VRM phases are split between PCIe 12V and AUX 12V. The only thing AMD can do about it is lower the maximum allowed clocks and voltages to lower worst-case power draw.

I cannot imagine how AMD could have done that sort of oversight nor how none of their board partners apparently caught it either. The boards shouldn't have passed PCIe certification. Did boards sent for qualification have different power limits from retail ones?

 

[-Fran-]

hopefully this can be fixed with a bios update. if it just a really bad design, then it could get real bad for amd trying to recall these boards!!!

A BIOS update cannot change how the board's VRM phases are split between PCIe 12V and AUX 12V. The only thing AMD can do about it is lower the maximum allowed clocks and voltages to lower worst-case power draw.

I cannot imagine how AMD could have done that sort of oversight nor how none of their board partners apparently caught it either. The boards shouldn't have passed PCIe certification. Did boards sent for qualification have different power limits from retail ones?

I would imagine they certify them at "stock clocks" only. Since OC'ing has always been considered "out of scope", I'd say they didn't even bother testing.

It might have been caught after the partners started toying with OC numbers I would imagine?

Cheers!

 

[n0ns3ns3]

They didn't even had the resources to test
And yes, BIOS can change the power draw balance from different sources. Not always, but possible.

 

[jimmysmitty]

hopefully this can be fixed with a bios update. if it just a really bad design, then it could get real bad for amd trying to recall these boards!!!

with the 470 and 460 still not out, those cards will suffer in sales for sure as folks will hesitate until they can be tested too.

sales would not hurt that much.
Most people would never know such an issue existed.
congrats, you are a member of a thinking minority

Yes and no. It is still something that AMD does not need since most people who wil buy these are people who read tech news. The only ones who will buy anyways will be the more hardcore fans, after this information has come out.

It is still bad press for AMD and of course we are currently awaiting some sort of response from AMD.

I hope the Zen launch goes better than this. I don't think AMD can take having to find another CEO.

 

[RedJaron]

except for the potentially damaging power issue. The possibility of smoking a trace in a PCIe slot is, and should be, an absolute dealbreaker. I would think it's an easy fix by any number of board partners, but until then, this card is a bad idea, at any price.

Correct, which is why I said this is a worse problem than the 970's 3.5 VRAM. I share in the befuddlement of how AMD released it with this problem. I can't say how easy or hard it is to fix, I only wish that it does.

 

[jfive194]

Toms Hardware: Considering all the buzz about yours and others rx 480 PCIE power draw tests are we going to see every new GPU and even some past ones put through this testing?

What I want to see is that every review site that has tested the RX 480 power draw with those home brewed test rigs go back and test as many different GPUs as they can. Do other cards fall out of spec? Which ones? If so by how much? That should keep things scientific and get rid of the claims of anti AMD bias. Furthermore it will help inform the consumers and enthusiasts who consider your work before parting with our hard earned dollars.

Thanks and keep up the good work!

 

[10tacle]

Correct, which is why I said this is a worse problem than the 970's 3.5 VRAM. I share in the befuddlement of how AMD released it with this problem. I can't say how easy or hard it is to fix, I only wish that it does.

I never understood what the big gripe was about that 3.5GB direct access VRAM issue anyway as an SLI 970 owner. Not one of my games in the past 2-3 years (nearly 2 years of owning 970) came close to allocating* even 3GB VRAM at 1440p with ultra and 4xAA in Afterburner (Crysis 3, BF3/4, Witcher 3, FC3/4, Battlefront, Project Cars, DiRT Rally, Assetto Corsa, RaceRoom Experience). Granted I don't have GTA-V or Shadow Of Mortar which are brutal on VRAM use.

*A lot of people still confuse VRAM memory allocation that shows up in GPU programs like Afterburner and GPU-Z with what is actually being used.